Merge branch 'str.format'

CMakeLists.txt

@@ -28,7 +28,7 @@ if(ENABLE_CCACHE)
   find_program(CCACHE ccache)
   if(CCACHE)
     message(STATUS "found ccache ${CCACHE}")
-    set_property(GLOBAL PROPERTY RULE_LAUNCH_COMPILE ${CCACHE})
+    set_property(GLOBAL PROPERTY RULE_LAUNCH_COMPILE "CCACHE_CPP2=yes ${CCACHE}")
     set_property(GLOBAL PROPERTY RULE_LAUNCH_LINK ${CCACHE})
   endif()
 endif()

Makefile

@@ -249,6 +249,8 @@ ifeq ($(USE_CCACHE),1)
 	CXX := ccache $(CXX)
 	CXX_PROFILE := ccache $(CXX_PROFILE)
 	CLANG_CXX := ccache $(CLANG_CXX)
+	CXX_ENV += CCACHE_CPP2=yes
+	CC_ENV += CCACHE_CPP2=yes
 	ifeq ($(USE_DISTCC),1)
 		CXX_ENV += CCACHE_PREFIX=distcc
 	endif
@@ -266,6 +268,7 @@ ifeq ($(USE_CCACHE),1)
 endif
 CXX := $(CXX_ENV) $(CXX)
 CXX_PROFILE := $(CXX_ENV) $(CXX_PROFILE)
+CC := $(CC_ENV) $(CC)
 CLANG_CXX := $(CXX_ENV) $(CLANG_CXX)
 # Not sure if ccache_basedir actually helps at all (I think the generated files make them different?)
 LLVM_BUILD_ENV += CCACHE_DIR=$(HOME)/.ccache_llvm CCACHE_BASEDIR=$(LLVM_SRC)
@@ -278,8 +281,8 @@ STDLIB_OBJS := stdlib.bc.o stdlib.stripped.bc.o
 STDLIB_RELEASE_OBJS := stdlib.release.bc.o
 
 STDMODULE_SRCS := errnomodule.c shamodule.c sha256module.c sha512module.c _math.c mathmodule.c md5.c md5module.c _randommodule.c _sre.c operator.c binascii.c pwdmodule.c posixmodule.c $(EXTRA_STDMODULE_SRCS)
-STDOBJECT_SRCS := structseq.c capsule.c $(EXTRA_STDOBJECT_SRCS)
-STDPYTHON_SRCS := pyctype.c getargs.c $(EXTRA_STDPYTHON_SRCS)
+STDOBJECT_SRCS := structseq.c capsule.c stringobject.c $(EXTRA_STDOBJECT_SRCS)
+STDPYTHON_SRCS := pyctype.c getargs.c formatter_string.c pystrtod.c dtoa.c $(EXTRA_STDPYTHON_SRCS)
 FROM_CPYTHON_SRCS := $(addprefix lib_python/2.7_Modules/,$(STDMODULE_SRCS)) $(addprefix lib_python/2.7_Objects/,$(STDOBJECT_SRCS)) $(addprefix lib_python/2.7_Python/,$(STDPYTHON_SRCS))
 
 # The stdlib objects have slightly longer dependency chains,
@@ -916,7 +919,7 @@ watch: watch_pyston_dbg
 watch_vim:
 	$(MAKE) watch WATCH_ARGS='COLOR=0 USE_DISTCC=0 -j1 2>&1 | tee compile.log'
 wdbg_%:
-	$(MAKE) $(patsubst wdbg_%,watch_dbg_%,$@) GDB_CMDS="--ex quit"
+	$(MAKE) $(patsubst wdbg_%,watch_dbg_%,$@) GDB_POST_CMDS="--ex quit"
 
 .PHONY: test_asm test_cpp_asm
 test_asm:

include/Python.h

@@ -63,6 +63,7 @@
 #include "dictobject.h"
 #include "tupleobject.h"
 #include "methodobject.h"
+#include "classobject.h"
 #include "fileobject.h"
 #include "pycapsule.h"
 #include "sliceobject.h"
@@ -84,6 +85,8 @@
 #include "abstract.h"
 
 #include "pyctype.h"
+#include "pystrtod.h"
+#include "dtoa.h"
 
 // directly from CPython:
 /* Argument must be a char or an int in [-128, 127] or [0, 255]. */

include/classobject.h

+// This file is originally from CPython 2.7, with modifications for Pyston
+
+/* Class object interface */
+
+/* Revealing some structures (not for general use) */
+
+#ifndef Py_CLASSOBJECT_H
+#define Py_CLASSOBJECT_H
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+// Pyston change: this is not the format we're using
+#if 0
+typedef struct {
+    PyObject_HEAD
+    PyObject	*cl_bases;	/* A tuple of class objects */
+    PyObject	*cl_dict;	/* A dictionary */
+    PyObject	*cl_name;	/* A string */
+    /* The following three are functions or NULL */
+    PyObject	*cl_getattr;
+    PyObject	*cl_setattr;
+    PyObject	*cl_delattr;
+    PyObject    *cl_weakreflist; /* List of weak references */
+} PyClassObject;
+
+typedef struct {
+    PyObject_HEAD
+    PyClassObject *in_class;	/* The class object */
+    PyObject	  *in_dict;	/* A dictionary */
+    PyObject	  *in_weakreflist; /* List of weak references */
+} PyInstanceObject;
+
+typedef struct {
+    PyObject_HEAD
+    PyObject *im_func;   /* The callable object implementing the method */
+    PyObject *im_self;   /* The instance it is bound to, or NULL */
+    PyObject *im_class;  /* The class that asked for the method */
+    PyObject *im_weakreflist; /* List of weak references */
+} PyMethodObject;
+#endif
+typedef struct _PyClassObject PyClassObject;
+typedef struct _PyInstanceObject PyInstanceObject;
+typedef struct _PyMethodObject PyMethodObject;
+
+// Pyston change: these are not static objects any more
+PyAPI_DATA(PyTypeObject*) classobj_cls;
+PyAPI_DATA(PyTypeObject*) instance_cls;
+PyAPI_DATA(PyTypeObject*) instancemethod_cls;
+#define PyClass_Type (*classobj_cls)
+#define PyInstance_Type (*instance_cls)
+#define PyMethod_Type (*instancemethod_cls)
+
+// Pyston change: change these to use the Py_TYPE macro instead
+// of looking at ob_type directly
+#define PyClass_Check(op) (Py_TYPE(op) == &PyClass_Type)
+#define PyInstance_Check(op) (Py_TYPE(op) == &PyInstance_Type)
+#define PyMethod_Check(op) (Py_TYPE(op) == &PyMethod_Type)
+
+PyAPI_FUNC(PyObject *) PyClass_New(PyObject *, PyObject *, PyObject *);
+PyAPI_FUNC(PyObject *) PyInstance_New(PyObject *, PyObject *,
+                                            PyObject *);
+PyAPI_FUNC(PyObject *) PyInstance_NewRaw(PyObject *, PyObject *);
+PyAPI_FUNC(PyObject *) PyMethod_New(PyObject *, PyObject *, PyObject *);
+
+PyAPI_FUNC(PyObject *) PyMethod_Function(PyObject *);
+PyAPI_FUNC(PyObject *) PyMethod_Self(PyObject *);
+PyAPI_FUNC(PyObject *) PyMethod_Class(PyObject *);
+
+/* Look up attribute with name (a string) on instance object pinst, using
+ * only the instance and base class dicts.  If a descriptor is found in
+ * a class dict, the descriptor is returned without calling it.
+ * Returns NULL if nothing found, else a borrowed reference to the
+ * value associated with name in the dict in which name was found.
+ * The point of this routine is that it never calls arbitrary Python
+ * code, so is always "safe":  all it does is dict lookups.  The function
+ * can't fail, never sets an exception, and NULL is not an error (it just
+ * means "not found").
+ */
+PyAPI_FUNC(PyObject *) _PyInstance_Lookup(PyObject *pinst, PyObject *name);
+
+/* Macros for direct access to these values. Type checks are *not*
+   done, so use with care. */
+#define PyMethod_GET_FUNCTION(meth) \
+        (((PyMethodObject *)meth) -> im_func)
+#define PyMethod_GET_SELF(meth) \
+	(((PyMethodObject *)meth) -> im_self)
+#define PyMethod_GET_CLASS(meth) \
+	(((PyMethodObject *)meth) -> im_class)
+
+PyAPI_FUNC(int) PyClass_IsSubclass(PyObject *, PyObject *);
+
+PyAPI_FUNC(int) PyMethod_ClearFreeList(void);
+
+#ifdef __cplusplus
+}
+#endif
+#endif /* !Py_CLASSOBJECT_H */
+

include/complexobject.h

@@ -46,6 +46,7 @@ typedef struct {
     Py_complex cval;
 } PyComplexObject;     
 #endif
+typedef struct _PyComplexObject PyComplexObject;
 
 // Pyston change: this is not a static object any more
 // PyAPI_DATA(PyTypeObject) PyComplex_Type;

include/descrobject.h

@@ -96,7 +96,9 @@ PyAPI_FUNC(PyObject *) PyDictProxy_New(PyObject *);
 PyAPI_FUNC(PyObject *) PyWrapper_New(PyObject *, PyObject *);
 
 
-PyAPI_DATA(PyTypeObject) PyProperty_Type;
+// Pyston change: this is no longer a static object
+//PyAPI_DATA(PyTypeObject) PyProperty_Type;
+
 #ifdef __cplusplus
 }
 #endif

include/dtoa.h

+// This file is originally from CPython 2.7, with modifications for Pyston
+
+#ifndef PY_NO_SHORT_FLOAT_REPR
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+PyAPI_FUNC(double) _Py_dg_strtod(const char *str, char **ptr);
+PyAPI_FUNC(char *) _Py_dg_dtoa(double d, int mode, int ndigits,
+                        int *decpt, int *sign, char **rve);
+PyAPI_FUNC(void) _Py_dg_freedtoa(char *s);
+
+
+#ifdef __cplusplus
+}
+#endif
+#endif

include/import.h

@@ -49,7 +49,8 @@ struct _inittab {
     void (*initfunc)(void);
 };
 
-PyAPI_DATA(PyTypeObject) PyNullImporter_Type;
+// Pyston change: this is no longer a static object
+//PyAPI_DATA(PyTypeObject) PyNullImporter_Type;
 PyAPI_DATA(struct _inittab *) PyImport_Inittab;
 
 PyAPI_FUNC(int) PyImport_AppendInittab(const char *name, void (*initfunc)(void));

include/iterobject.h

@@ -7,13 +7,15 @@
 extern "C" {
 #endif
 
-PyAPI_DATA(PyTypeObject) PySeqIter_Type;
+// Pyston change: this is no longer a static object
+//PyAPI_DATA(PyTypeObject) PySeqIter_Type;
 
 #define PySeqIter_Check(op) (Py_TYPE(op) == &PySeqIter_Type)
 
 PyAPI_FUNC(PyObject *) PySeqIter_New(PyObject *);
 
-PyAPI_DATA(PyTypeObject) PyCallIter_Type;
+// Pyston change: this is no longer a static object
+//PyAPI_DATA(PyTypeObject) PyCallIter_Type;
 
 #define PyCallIter_Check(op) (Py_TYPE(op) == &PyCallIter_Type)
 

include/methodobject.h

@@ -12,7 +12,8 @@ extern "C" {
    not Python methods in user-defined classes.  See classobject.h
    for the latter. */
 
-PyAPI_DATA(PyTypeObject) PyCFunction_Type;
+// Pyston change: this is no longer a static object
+//PyAPI_DATA(PyTypeObject) PyCFunction_Type;
 
 #define PyCFunction_Check(op) (Py_TYPE(op) == &PyCFunction_Type)
 

include/object.h

@@ -495,8 +495,9 @@ PyAPI_FUNC(int) PyType_IsSubtype(PyTypeObject *, PyTypeObject *);
 PyAPI_DATA(PyTypeObject*) type_cls;
 #define PyType_Type (*type_cls)
 
-PyAPI_DATA(PyTypeObject) PyBaseObject_Type; /* built-in 'object' */
-PyAPI_DATA(PyTypeObject) PySuper_Type; /* built-in 'super' */
+// Pyston change: this is no longer a static object
+//PyAPI_DATA(PyTypeObject) PyBaseObject_Type; /* built-in 'object' */
+//PyAPI_DATA(PyTypeObject) PySuper_Type; /* built-in 'super' */
 
 // Pyston changes: these aren't direct macros any more [they potentially could be though]
 PyAPI_FUNC(bool) PyType_Check(PyObject*);
@@ -511,7 +512,8 @@ PyAPI_FUNC(PyObject *) PyType_GenericAlloc(PyTypeObject *, Py_ssize_t);
 PyAPI_FUNC(PyObject *) PyType_GenericNew(PyTypeObject *,
                                                PyObject *, PyObject *);
 PyAPI_FUNC(PyObject *) _PyType_Lookup(PyTypeObject *, PyObject *);
-PyAPI_FUNC(PyObject *) _PyObject_LookupSpecial(PyObject *, char *, PyObject **);
+// Pyston change: modified this to take a const char*
+PyAPI_FUNC(PyObject *) _PyObject_LookupSpecial(PyObject *, const char *, PyObject **);
 PyAPI_FUNC(unsigned int) PyType_ClearCache(void);
 PyAPI_FUNC(void) PyType_Modified(PyTypeObject *);
 

include/pyconfig.h

@@ -36,6 +36,13 @@
 #define HAVE_EXPM1 1
 #define Py_USING_UNICODE 1
 #define Py_UNICODE_SIZE 4
+#define HAVE_UINT32_T 1
+#define HAVE_UINT64_T 1
+#define HAVE_UINTPTR_T 1
+#define HAVE_INT32_T 1
+#define HAVE_INT64_T 1
+
+#define DOUBLE_IS_LITTLE_ENDIAN_IEEE754 1
 
 #define WITH_THREAD
 

include/pyport.h

@@ -15,6 +15,96 @@ typedef ssize_t         Py_ssize_t;
 
 // Pyston change: the rest of these have just been copied from CPython's pyport.h, in an arbitrary order:
 
+#if defined(_MSC_VER)
+#if defined(PY_LOCAL_AGGRESSIVE)
+/* enable more aggressive optimization for visual studio */
+#pragma optimize("agtw", on)
+#endif
+/* ignore warnings if the compiler decides not to inline a function */
+#pragma warning(disable: 4710)
+/* fastest possible local call under MSVC */
+#define Py_LOCAL(type) static type __fastcall
+#define Py_LOCAL_INLINE(type) static __inline type __fastcall
+#elif defined(USE_INLINE)
+#define Py_LOCAL(type) static type
+#define Py_LOCAL_INLINE(type) static inline type
+#else
+#define Py_LOCAL(type) static type
+#define Py_LOCAL_INLINE(type) static type
+#endif
+
+/*  The functions _Py_dg_strtod and _Py_dg_dtoa in Python/dtoa.c (which are
+ *  required to support the short float repr introduced in Python 3.1) require
+ *  that the floating-point unit that's being used for arithmetic operations
+ *  on C doubles is set to use 53-bit precision.  It also requires that the
+ *  FPU rounding mode is round-half-to-even, but that's less often an issue.
+ *
+ *  If your FPU isn't already set to 53-bit precision/round-half-to-even, and
+ *  you want to make use of _Py_dg_strtod and _Py_dg_dtoa, then you should
+ *
+ *     #define HAVE_PY_SET_53BIT_PRECISION 1
+ *
+ *  and also give appropriate definitions for the following three macros:
+ *
+ *    _PY_SET_53BIT_PRECISION_START : store original FPU settings, and
+ *        set FPU to 53-bit precision/round-half-to-even
+ *    _PY_SET_53BIT_PRECISION_END : restore original FPU settings
+ *    _PY_SET_53BIT_PRECISION_HEADER : any variable declarations needed to
+ *        use the two macros above.
+ *
+ * The macros are designed to be used within a single C function: see
+ * Python/pystrtod.c for an example of their use.
+ */
+
+/* get and set x87 control word for gcc/x86 */
+#ifdef HAVE_GCC_ASM_FOR_X87
+#define HAVE_PY_SET_53BIT_PRECISION 1
+/* _Py_get/set_387controlword functions are defined in Python/pymath.c */
+#define _Py_SET_53BIT_PRECISION_HEADER                          \
+    unsigned short old_387controlword, new_387controlword
+#define _Py_SET_53BIT_PRECISION_START                                   \
+    do {                                                                \
+        old_387controlword = _Py_get_387controlword();                  \
+        new_387controlword = (old_387controlword & ~0x0f00) | 0x0200; \
+        if (new_387controlword != old_387controlword)                   \
+            _Py_set_387controlword(new_387controlword);                 \
+    } while (0)
+#define _Py_SET_53BIT_PRECISION_END                             \
+    if (new_387controlword != old_387controlword)               \
+        _Py_set_387controlword(old_387controlword)
+#endif
+
+/* get and set x87 control word for VisualStudio/x86 */
+#if defined(_MSC_VER) && !defined(_WIN64) /* x87 not supported in 64-bit */
+#define HAVE_PY_SET_53BIT_PRECISION 1
+#define _Py_SET_53BIT_PRECISION_HEADER \
+    unsigned int old_387controlword, new_387controlword, out_387controlword
+/* We use the __control87_2 function to set only the x87 control word.
+   The SSE control word is unaffected. */
+#define _Py_SET_53BIT_PRECISION_START                                   \
+    do {                                                                \
+        __control87_2(0, 0, &old_387controlword, NULL);                 \
+        new_387controlword =                                            \
+          (old_387controlword & ~(_MCW_PC | _MCW_RC)) | (_PC_53 | _RC_NEAR); \
+        if (new_387controlword != old_387controlword)                   \
+            __control87_2(new_387controlword, _MCW_PC | _MCW_RC,        \
+                          &out_387controlword, NULL);                   \
+    } while (0)
+#define _Py_SET_53BIT_PRECISION_END                                     \
+    do {                                                                \
+        if (new_387controlword != old_387controlword)                   \
+            __control87_2(old_387controlword, _MCW_PC | _MCW_RC,        \
+                          &out_387controlword, NULL);                   \
+    } while (0)
+#endif
+
+/* default definitions are empty */
+#ifndef HAVE_PY_SET_53BIT_PRECISION
+#define _Py_SET_53BIT_PRECISION_HEADER
+#define _Py_SET_53BIT_PRECISION_START
+#define _Py_SET_53BIT_PRECISION_END
+#endif
+
 /* Py_DEPRECATED(version)
  * Declare a variable, type, or function deprecated.
  * Usage:
@@ -137,5 +227,50 @@ typedef ssize_t         Py_ssize_t;
 /* Smallest negative value of type Py_ssize_t. */
 #define PY_SSIZE_T_MIN (-PY_SSIZE_T_MAX-1)
 
+#ifdef uint32_t
+#define HAVE_UINT32_T 1
+#endif
+
+#ifdef HAVE_UINT32_T
+#ifndef PY_UINT32_T
+#define PY_UINT32_T uint32_t
+#endif
+#endif
+
+/* Macros for a 64-bit unsigned integer type; used for type 'twodigits' in the
+ * long integer implementation, when 30-bit digits are enabled.
+ */
+#ifdef uint64_t
+#define HAVE_UINT64_T 1
+#endif
+
+#ifdef HAVE_UINT64_T
+#ifndef PY_UINT64_T
+#define PY_UINT64_T uint64_t
+#endif
+#endif
+
+/* Signed variants of the above */
+#ifdef int32_t
+#define HAVE_INT32_T 1
+#endif
+
+#ifdef HAVE_INT32_T
+#ifndef PY_INT32_T
+#define PY_INT32_T int32_t
+#endif
+#endif
+
+#ifdef int64_t
+#define HAVE_INT64_T 1
+#endif
+
+#ifdef HAVE_INT64_T
+#ifndef PY_INT64_T
+#define PY_INT64_T int64_t
+#endif
+#endif
+
+
 #endif /* Py_PYPORT_H */
 

include/pystrtod.h

+// This file is originally from CPython 2.7, with modifications for Pyston
+//
+#ifndef Py_STRTOD_H
+#define Py_STRTOD_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+
+PyAPI_FUNC(double) PyOS_ascii_strtod(const char *str, char **ptr);
+PyAPI_FUNC(double) PyOS_ascii_atof(const char *str);
+
+/* Deprecated in 2.7 and 3.1. Will disappear in 2.8 (if it exists) and 3.2 */
+PyAPI_FUNC(char *) PyOS_ascii_formatd(char *buffer, size_t buf_len,
+                                      const char *format, double d);
+PyAPI_FUNC(double) PyOS_string_to_double(const char *str,
+                                         char **endptr,
+                                         PyObject *overflow_exception);
+
+/* The caller is responsible for calling PyMem_Free to free the buffer
+   that's is returned. */
+PyAPI_FUNC(char *) PyOS_double_to_string(double val,
+                                         char format_code,
+                                         int precision,
+                                         int flags,
+                                         int *type);
+
+PyAPI_FUNC(double) _Py_parse_inf_or_nan(const char *p, char **endptr);
+
+
+/* PyOS_double_to_string's "flags" parameter can be set to 0 or more of: */
+#define Py_DTSF_SIGN      0x01 /* always add the sign */
+#define Py_DTSF_ADD_DOT_0 0x02 /* if the result is an integer add ".0" */
+#define Py_DTSF_ALT       0x04 /* "alternate" formatting. it's format_code
+                                  specific */
+
+/* PyOS_double_to_string's "type", if non-NULL, will be set to one of: */
+#define Py_DTST_FINITE 0
+#define Py_DTST_INFINITE 1
+#define Py_DTST_NAN 2
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* !Py_STRTOD_H */
+

include/stringobject.h

@@ -58,8 +58,9 @@ typedef struct _PyStringObject PyStringObject;
 #define SSTATE_INTERNED_MORTAL 1
 #define SSTATE_INTERNED_IMMORTAL 2
 
-PyAPI_DATA(PyTypeObject) PyBaseString_Type;
-// Pyston change: this is no longer a static object
+// Pyston change: these are no longer a static object
+PyAPI_DATA(PyTypeObject*) basestring_cls;
+#define PyBaseString_Type (*basestring_cls)
 PyAPI_DATA(PyTypeObject*) str_cls;
 #define PyString_Type (*str_cls)
 
@@ -101,8 +102,8 @@ PyAPI_FUNC(void) _Py_ReleaseInternedStrings(void);
 
 /* Macro, trading safety for speed */
 // Pyston changes: these aren't direct macros any more [they potentially could be though]
-#define PyString_AS_STRING(op) PyString_AsString(op)
-#define PyString_GET_SIZE(op)  PyString_Size(op)
+#define PyString_AS_STRING(op) PyString_AsString((PyObject*)op)
+#define PyString_GET_SIZE(op)  PyString_Size((PyObject*)op)
 //#define PyString_AS_STRING(op) (((PyStringObject *)(op))->ob_sval)
 //#define PyString_GET_SIZE(op)  Py_SIZE(op)
 

lib_python/2.7_Objects/stringlib/README.txt

+bits shared by the stringobject and unicodeobject implementations (and
+possibly other modules, in a not too distant future).
+
+the stuff in here is included into relevant places; see the individual
+source files for details.
+
+--------------------------------------------------------------------
+the following defines used by the different modules:
+
+STRINGLIB_CHAR
+
+    the type used to hold a character (char or Py_UNICODE)
+
+STRINGLIB_EMPTY
+
+    a PyObject representing the empty string, only to be used if
+    STRINGLIB_MUTABLE is 0
+
+Py_ssize_t STRINGLIB_LEN(PyObject*)
+
+    returns the length of the given string object (which must be of the
+    right type)
+
+PyObject* STRINGLIB_NEW(STRINGLIB_CHAR*, Py_ssize_t)
+
+    creates a new string object
+
+STRINGLIB_CHAR* STRINGLIB_STR(PyObject*)
+
+    returns the pointer to the character data for the given string
+    object (which must be of the right type)
+
+int STRINGLIB_CHECK_EXACT(PyObject *)
+
+    returns true if the object is an instance of our type, not a subclass
+
+STRINGLIB_MUTABLE
+
+    must be 0 or 1 to tell the cpp macros in stringlib code if the object
+    being operated on is mutable or not

lib_python/2.7_Objects/stringlib/count.h

+/* stringlib: count implementation */
+
+#ifndef STRINGLIB_COUNT_H
+#define STRINGLIB_COUNT_H
+
+#ifndef STRINGLIB_FASTSEARCH_H
+#error must include "stringlib/fastsearch.h" before including this module
+#endif
+
+Py_LOCAL_INLINE(Py_ssize_t)
+stringlib_count(const STRINGLIB_CHAR* str, Py_ssize_t str_len,
+                const STRINGLIB_CHAR* sub, Py_ssize_t sub_len,
+                Py_ssize_t maxcount)
+{
+    Py_ssize_t count;
+
+    if (str_len < 0)
+        return 0; /* start > len(str) */
+    if (sub_len == 0)
+        return (str_len < maxcount) ? str_len + 1 : maxcount;
+
+    count = fastsearch(str, str_len, sub, sub_len, maxcount, FAST_COUNT);
+
+    if (count < 0)
+        return 0; /* no match */
+
+    return count;
+}
+
+#endif

lib_python/2.7_Objects/stringlib/ctype.h

+/* NOTE: this API is -ONLY- for use with single byte character strings. */
+/* Do not use it with Unicode. */
+
+#include "bytes_methods.h"
+
+static PyObject*
+stringlib_isspace(PyObject *self)
+{
+    return _Py_bytes_isspace(STRINGLIB_STR(self), STRINGLIB_LEN(self));
+}
+
+static PyObject*
+stringlib_isalpha(PyObject *self)
+{
+    return _Py_bytes_isalpha(STRINGLIB_STR(self), STRINGLIB_LEN(self));
+}
+
+static PyObject*
+stringlib_isalnum(PyObject *self)
+{
+    return _Py_bytes_isalnum(STRINGLIB_STR(self), STRINGLIB_LEN(self));
+}
+
+static PyObject*
+stringlib_isdigit(PyObject *self)
+{
+    return _Py_bytes_isdigit(STRINGLIB_STR(self), STRINGLIB_LEN(self));
+}
+
+static PyObject*
+stringlib_islower(PyObject *self)
+{
+    return _Py_bytes_islower(STRINGLIB_STR(self), STRINGLIB_LEN(self));
+}
+
+static PyObject*
+stringlib_isupper(PyObject *self)
+{
+    return _Py_bytes_isupper(STRINGLIB_STR(self), STRINGLIB_LEN(self));
+}
+
+static PyObject*
+stringlib_istitle(PyObject *self)
+{
+    return _Py_bytes_istitle(STRINGLIB_STR(self), STRINGLIB_LEN(self));
+}
+
+
+/* functions that return a new object partially translated by ctype funcs: */
+
+static PyObject*
+stringlib_lower(PyObject *self)
+{
+    PyObject* newobj;
+    newobj = STRINGLIB_NEW(NULL, STRINGLIB_LEN(self));
+    if (!newobj)
+            return NULL;
+    _Py_bytes_lower(STRINGLIB_STR(newobj), STRINGLIB_STR(self),
+                 STRINGLIB_LEN(self));
+    return newobj;
+}
+
+static PyObject*
+stringlib_upper(PyObject *self)
+{
+    PyObject* newobj;
+    newobj = STRINGLIB_NEW(NULL, STRINGLIB_LEN(self));
+    if (!newobj)
+            return NULL;
+    _Py_bytes_upper(STRINGLIB_STR(newobj), STRINGLIB_STR(self),
+                 STRINGLIB_LEN(self));
+    return newobj;
+}
+
+static PyObject*
+stringlib_title(PyObject *self)
+{
+    PyObject* newobj;
+    newobj = STRINGLIB_NEW(NULL, STRINGLIB_LEN(self));
+    if (!newobj)
+            return NULL;
+    _Py_bytes_title(STRINGLIB_STR(newobj), STRINGLIB_STR(self),
+                 STRINGLIB_LEN(self));
+    return newobj;
+}
+
+static PyObject*
+stringlib_capitalize(PyObject *self)
+{
+    PyObject* newobj;
+    newobj = STRINGLIB_NEW(NULL, STRINGLIB_LEN(self));
+    if (!newobj)
+            return NULL;
+    _Py_bytes_capitalize(STRINGLIB_STR(newobj), STRINGLIB_STR(self),
+                      STRINGLIB_LEN(self));
+    return newobj;
+}
+
+static PyObject*
+stringlib_swapcase(PyObject *self)
+{
+    PyObject* newobj;
+    newobj = STRINGLIB_NEW(NULL, STRINGLIB_LEN(self));
+    if (!newobj)
+            return NULL;
+    _Py_bytes_swapcase(STRINGLIB_STR(newobj), STRINGLIB_STR(self),
+                    STRINGLIB_LEN(self));
+    return newobj;
+}

lib_python/2.7_Objects/stringlib/fastsearch.h

+/* stringlib: fastsearch implementation */
+
+#ifndef STRINGLIB_FASTSEARCH_H
+#define STRINGLIB_FASTSEARCH_H
+
+/* fast search/count implementation, based on a mix between boyer-
+   moore and horspool, with a few more bells and whistles on the top.
+   for some more background, see: http://effbot.org/zone/stringlib.htm */
+
+/* note: fastsearch may access s[n], which isn't a problem when using
+   Python's ordinary string types, but may cause problems if you're
+   using this code in other contexts.  also, the count mode returns -1
+   if there cannot possible be a match in the target string, and 0 if
+   it has actually checked for matches, but didn't find any.  callers
+   beware! */
+
+#define FAST_COUNT 0
+#define FAST_SEARCH 1
+#define FAST_RSEARCH 2
+
+#if LONG_BIT >= 128
+#define STRINGLIB_BLOOM_WIDTH 128
+#elif LONG_BIT >= 64
+#define STRINGLIB_BLOOM_WIDTH 64
+#elif LONG_BIT >= 32
+#define STRINGLIB_BLOOM_WIDTH 32
+#else
+#error "LONG_BIT is smaller than 32"
+#endif
+
+#define STRINGLIB_BLOOM_ADD(mask, ch) \
+    ((mask |= (1UL << ((ch) & (STRINGLIB_BLOOM_WIDTH -1)))))
+#define STRINGLIB_BLOOM(mask, ch)     \
+    ((mask &  (1UL << ((ch) & (STRINGLIB_BLOOM_WIDTH -1)))))
+
+Py_LOCAL_INLINE(Py_ssize_t)
+fastsearch(const STRINGLIB_CHAR* s, Py_ssize_t n,
+           const STRINGLIB_CHAR* p, Py_ssize_t m,
+           Py_ssize_t maxcount, int mode)
+{
+    unsigned long mask;
+    Py_ssize_t skip, count = 0;
+    Py_ssize_t i, j, mlast, w;
+
+    w = n - m;
+
+    if (w < 0 || (mode == FAST_COUNT && maxcount == 0))
+        return -1;
+
+    /* look for special cases */
+    if (m <= 1) {
+        if (m <= 0)
+            return -1;
+        /* use special case for 1-character strings */
+        if (mode == FAST_COUNT) {
+            for (i = 0; i < n; i++)
+                if (s[i] == p[0]) {
+                    count++;
+                    if (count == maxcount)
+                        return maxcount;
+                }
+            return count;
+        } else if (mode == FAST_SEARCH) {
+            for (i = 0; i < n; i++)
+                if (s[i] == p[0])
+                    return i;
+        } else {    /* FAST_RSEARCH */
+            for (i = n - 1; i > -1; i--)
+                if (s[i] == p[0])
+                    return i;
+        }
+        return -1;
+    }
+
+    mlast = m - 1;
+    skip = mlast - 1;
+    mask = 0;
+
+    if (mode != FAST_RSEARCH) {
+
+        /* create compressed boyer-moore delta 1 table */
+
+        /* process pattern[:-1] */
+        for (i = 0; i < mlast; i++) {
+            STRINGLIB_BLOOM_ADD(mask, p[i]);
+            if (p[i] == p[mlast])
+                skip = mlast - i - 1;
+        }
+        /* process pattern[-1] outside the loop */
+        STRINGLIB_BLOOM_ADD(mask, p[mlast]);
+
+        for (i = 0; i <= w; i++) {
+            /* note: using mlast in the skip path slows things down on x86 */
+            if (s[i+m-1] == p[m-1]) {
+                /* candidate match */
+                for (j = 0; j < mlast; j++)
+                    if (s[i+j] != p[j])
+                        break;
+                if (j == mlast) {
+                    /* got a match! */
+                    if (mode != FAST_COUNT)
+                        return i;
+                    count++;
+                    if (count == maxcount)
+                        return maxcount;
+                    i = i + mlast;
+                    continue;
+                }
+                /* miss: check if next character is part of pattern */
+                if (!STRINGLIB_BLOOM(mask, s[i+m]))
+                    i = i + m;
+                else
+                    i = i + skip;
+            } else {
+                /* skip: check if next character is part of pattern */
+                if (!STRINGLIB_BLOOM(mask, s[i+m]))
+                    i = i + m;
+            }
+        }
+    } else {    /* FAST_RSEARCH */
+
+        /* create compressed boyer-moore delta 1 table */
+
+        /* process pattern[0] outside the loop */
+        STRINGLIB_BLOOM_ADD(mask, p[0]);
+        /* process pattern[:0:-1] */
+        for (i = mlast; i > 0; i--) {
+            STRINGLIB_BLOOM_ADD(mask, p[i]);
+            if (p[i] == p[0])
+                skip = i - 1;
+        }
+
+        for (i = w; i >= 0; i--) {
+            if (s[i] == p[0]) {
+                /* candidate match */
+                for (j = mlast; j > 0; j--)
+                    if (s[i+j] != p[j])
+                        break;
+                if (j == 0)
+                    /* got a match! */
+                    return i;
+                /* miss: check if previous character is part of pattern */
+                if (i > 0 && !STRINGLIB_BLOOM(mask, s[i-1]))
+                    i = i - m;
+                else
+                    i = i - skip;
+            } else {
+                /* skip: check if previous character is part of pattern */
+                if (i > 0 && !STRINGLIB_BLOOM(mask, s[i-1]))
+                    i = i - m;
+            }
+        }
+    }
+
+    if (mode != FAST_COUNT)
+        return -1;
+    return count;
+}
+
+#endif

lib_python/2.7_Objects/stringlib/find.h

+/* stringlib: find/index implementation */
+
+#ifndef STRINGLIB_FIND_H
+#define STRINGLIB_FIND_H
+
+#ifndef STRINGLIB_FASTSEARCH_H
+#error must include "stringlib/fastsearch.h" before including this module
+#endif
+
+Py_LOCAL_INLINE(Py_ssize_t)
+stringlib_find(const STRINGLIB_CHAR* str, Py_ssize_t str_len,
+               const STRINGLIB_CHAR* sub, Py_ssize_t sub_len,
+               Py_ssize_t offset)
+{
+    Py_ssize_t pos;
+
+    if (str_len < 0)
+        return -1;
+    if (sub_len == 0)
+        return offset;
+
+    pos = fastsearch(str, str_len, sub, sub_len, -1, FAST_SEARCH);
+
+    if (pos >= 0)
+        pos += offset;
+
+    return pos;
+}
+
+Py_LOCAL_INLINE(Py_ssize_t)
+stringlib_rfind(const STRINGLIB_CHAR* str, Py_ssize_t str_len,
+                const STRINGLIB_CHAR* sub, Py_ssize_t sub_len,
+                Py_ssize_t offset)
+{
+    Py_ssize_t pos;
+
+    if (str_len < 0)
+        return -1;
+    if (sub_len == 0)
+        return str_len + offset;
+
+    pos = fastsearch(str, str_len, sub, sub_len, -1, FAST_RSEARCH);
+
+    if (pos >= 0)
+        pos += offset;
+
+    return pos;
+}
+
+/* helper macro to fixup start/end slice values */
+#define ADJUST_INDICES(start, end, len)         \
+    if (end > len)                              \
+        end = len;                              \
+    else if (end < 0) {                         \
+        end += len;                             \
+        if (end < 0)                            \
+            end = 0;                            \
+    }                                           \
+    if (start < 0) {                            \
+        start += len;                           \
+        if (start < 0)                          \
+            start = 0;                          \
+    }
+
+Py_LOCAL_INLINE(Py_ssize_t)
+stringlib_find_slice(const STRINGLIB_CHAR* str, Py_ssize_t str_len,
+                     const STRINGLIB_CHAR* sub, Py_ssize_t sub_len,
+                     Py_ssize_t start, Py_ssize_t end)
+{
+    ADJUST_INDICES(start, end, str_len);
+    return stringlib_find(str + start, end - start, sub, sub_len, start);
+}
+
+Py_LOCAL_INLINE(Py_ssize_t)
+stringlib_rfind_slice(const STRINGLIB_CHAR* str, Py_ssize_t str_len,
+                      const STRINGLIB_CHAR* sub, Py_ssize_t sub_len,
+                      Py_ssize_t start, Py_ssize_t end)
+{
+    ADJUST_INDICES(start, end, str_len);
+    return stringlib_rfind(str + start, end - start, sub, sub_len, start);
+}
+
+#ifdef STRINGLIB_WANT_CONTAINS_OBJ
+
+Py_LOCAL_INLINE(int)
+stringlib_contains_obj(PyObject* str, PyObject* sub)
+{
+    return stringlib_find(
+        STRINGLIB_STR(str), STRINGLIB_LEN(str),
+        STRINGLIB_STR(sub), STRINGLIB_LEN(sub), 0
+        ) != -1;
+}
+
+#endif /* STRINGLIB_WANT_CONTAINS_OBJ */
+
+/*
+This function is a helper for the "find" family (find, rfind, index,
+rindex) and for count, startswith and endswith, because they all have
+the same behaviour for the arguments.
+
+It does not touch the variables received until it knows everything 
+is ok.
+*/
+
+#define FORMAT_BUFFER_SIZE 50
+
+Py_LOCAL_INLINE(int)
+stringlib_parse_args_finds(const char * function_name, PyObject *args,
+                           PyObject **subobj,
+                           Py_ssize_t *start, Py_ssize_t *end)
+{
+    PyObject *tmp_subobj;
+    Py_ssize_t tmp_start = 0;
+    Py_ssize_t tmp_end = PY_SSIZE_T_MAX;
+    PyObject *obj_start=Py_None, *obj_end=Py_None;
+    char format[FORMAT_BUFFER_SIZE] = "O|OO:";
+    size_t len = strlen(format);
+
+    strncpy(format + len, function_name, FORMAT_BUFFER_SIZE - len - 1);
+    format[FORMAT_BUFFER_SIZE - 1] = '\0';
+
+    if (!PyArg_ParseTuple(args, format, &tmp_subobj, &obj_start, &obj_end))
+        return 0;
+
+    /* To support None in "start" and "end" arguments, meaning
+       the same as if they were not passed.
+    */
+    if (obj_start != Py_None)
+        if (!_PyEval_SliceIndex(obj_start, &tmp_start))
+            return 0;
+    if (obj_end != Py_None)
+        if (!_PyEval_SliceIndex(obj_end, &tmp_end))
+            return 0;
+
+    *start = tmp_start;
+    *end = tmp_end;
+    *subobj = tmp_subobj;
+    return 1;
+}
+
+#undef FORMAT_BUFFER_SIZE
+
+#if STRINGLIB_IS_UNICODE
+
+/*
+Wraps stringlib_parse_args_finds() and additionally ensures that the
+first argument is a unicode object.
+
+Note that we receive a pointer to the pointer of the substring object,
+so when we create that object in this function we don't DECREF it,
+because it continues living in the caller functions (those functions, 
+after finishing using the substring, must DECREF it).
+*/
+
+Py_LOCAL_INLINE(int)
+stringlib_parse_args_finds_unicode(const char * function_name, PyObject *args,
+                                   PyUnicodeObject **substring,
+                                   Py_ssize_t *start, Py_ssize_t *end)
+{
+    PyObject *tmp_substring;
+
+    if(stringlib_parse_args_finds(function_name, args, &tmp_substring,
+                                  start, end)) {
+        tmp_substring = PyUnicode_FromObject(tmp_substring);
+        if (!tmp_substring)
+            return 0;
+        *substring = (PyUnicodeObject *)tmp_substring;
+        return 1;
+    }
+    return 0;
+}
+
+#endif /* STRINGLIB_IS_UNICODE */
+
+#endif /* STRINGLIB_FIND_H */

lib_python/2.7_Objects/stringlib/formatter.h

+// This file is originally from CPython 2.7, with modifications for Pyston
+
+/* implements the string, long, and float formatters.  that is,
+   string.__format__, etc. */
+
+#include <locale.h>
+
+/* Before including this, you must include either:
+   stringlib/unicodedefs.h
+   stringlib/stringdefs.h
+
+   Also, you should define the names:
+   FORMAT_STRING
+   FORMAT_LONG
+   FORMAT_FLOAT
+   FORMAT_COMPLEX
+   to be whatever you want the public names of these functions to
+   be.  These are the only non-static functions defined here.
+*/
+
+/* Raises an exception about an unknown presentation type for this
+ * type. */
+
+static void
+unknown_presentation_type(STRINGLIB_CHAR presentation_type,
+                          const char* type_name)
+{
+#if STRINGLIB_IS_UNICODE
+    /* If STRINGLIB_CHAR is Py_UNICODE, %c might be out-of-range,
+       hence the two cases. If it is char, gcc complains that the
+       condition below is always true, hence the ifdef. */
+    if (presentation_type > 32 && presentation_type < 128)
+#endif
+        PyErr_Format(PyExc_ValueError,
+                     "Unknown format code '%c' "
+                     "for object of type '%.200s'",
+                     (char)presentation_type,
+                     type_name);
+#if STRINGLIB_IS_UNICODE
+    else
+        PyErr_Format(PyExc_ValueError,
+                     "Unknown format code '\\x%x' "
+                     "for object of type '%.200s'",
+                     (unsigned int)presentation_type,
+                     type_name);
+#endif
+}
+
+static void
+invalid_comma_type(STRINGLIB_CHAR presentation_type)
+{
+#if STRINGLIB_IS_UNICODE
+    /* See comment in unknown_presentation_type */
+    if (presentation_type > 32 && presentation_type < 128)
+#endif
+        PyErr_Format(PyExc_ValueError,
+                     "Cannot specify ',' with '%c'.",
+                     (char)presentation_type);
+#if STRINGLIB_IS_UNICODE
+    else
+        PyErr_Format(PyExc_ValueError,
+                     "Cannot specify ',' with '\\x%x'.",
+                     (unsigned int)presentation_type);
+#endif
+}
+
+/*
+    get_integer consumes 0 or more decimal digit characters from an
+    input string, updates *result with the corresponding positive
+    integer, and returns the number of digits consumed.
+
+    returns -1 on error.
+*/
+static int
+get_integer(STRINGLIB_CHAR **ptr, STRINGLIB_CHAR *end,
+                  Py_ssize_t *result)
+{
+    Py_ssize_t accumulator, digitval;
+    int numdigits;
+    accumulator = numdigits = 0;
+    for (;;(*ptr)++, numdigits++) {
+        if (*ptr >= end)
+            break;
+        digitval = STRINGLIB_TODECIMAL(**ptr);
+        if (digitval < 0)
+            break;
+        /*
+           Detect possible overflow before it happens:
+
+              accumulator * 10 + digitval > PY_SSIZE_T_MAX if and only if
+              accumulator > (PY_SSIZE_T_MAX - digitval) / 10.
+        */
+        if (accumulator > (PY_SSIZE_T_MAX - digitval) / 10) {
+            PyErr_Format(PyExc_ValueError,
+                         "Too many decimal digits in format string");
+            return -1;
+        }
+        accumulator = accumulator * 10 + digitval;
+    }
+    *result = accumulator;
+    return numdigits;
+}
+
+/************************************************************************/
+/*********** standard format specifier parsing **************************/
+/************************************************************************/
+
+/* returns true if this character is a specifier alignment token */
+Py_LOCAL_INLINE(int)
+is_alignment_token(STRINGLIB_CHAR c)
+{
+    switch (c) {
+    case '<': case '>': case '=': case '^':
+        return 1;
+    default:
+        return 0;
+    }
+}
+
+/* returns true if this character is a sign element */
+Py_LOCAL_INLINE(int)
+is_sign_element(STRINGLIB_CHAR c)
+{
+    switch (c) {
+    case ' ': case '+': case '-':
+        return 1;
+    default:
+        return 0;
+    }
+}
+
+
+typedef struct {
+    STRINGLIB_CHAR fill_char;
+    STRINGLIB_CHAR align;
+    int alternate;
+    STRINGLIB_CHAR sign;
+    Py_ssize_t width;
+    int thousands_separators;
+    Py_ssize_t precision;
+    STRINGLIB_CHAR type;
+} InternalFormatSpec;
+
+
+#if 0
+/* Occassionally useful for debugging. Should normally be commented out. */
+static void
+DEBUG_PRINT_FORMAT_SPEC(InternalFormatSpec *format)
+{
+    printf("internal format spec: fill_char %d\n", format->fill_char);
+    printf("internal format spec: align %d\n", format->align);
+    printf("internal format spec: alternate %d\n", format->alternate);
+    printf("internal format spec: sign %d\n", format->sign);
+    printf("internal format spec: width %zd\n", format->width);
+    printf("internal format spec: thousands_separators %d\n",
+           format->thousands_separators);
+    printf("internal format spec: precision %zd\n", format->precision);
+    printf("internal format spec: type %c\n", format->type);
+    printf("\n");
+}
+#endif
+
+
+/*
+  ptr points to the start of the format_spec, end points just past its end.
+  fills in format with the parsed information.
+  returns 1 on success, 0 on failure.
+  if failure, sets the exception
+*/
+static int
+parse_internal_render_format_spec(STRINGLIB_CHAR *format_spec,
+                                  Py_ssize_t format_spec_len,
+                                  InternalFormatSpec *format,
+                                  char default_type,
+                                  char default_align)
+{
+    STRINGLIB_CHAR *ptr = format_spec;
+    STRINGLIB_CHAR *end = format_spec + format_spec_len;
+
+    /* end-ptr is used throughout this code to specify the length of
+       the input string */
+
+    Py_ssize_t consumed;
+    int align_specified = 0;
+    int fill_char_specified = 0;
+
+    format->fill_char = ' ';
+    format->align = default_align;
+    format->alternate = 0;
+    format->sign = '\0';
+    format->width = -1;
+    format->thousands_separators = 0;
+    format->precision = -1;
+    format->type = default_type;
+
+    /* If the second char is an alignment token,
+       then parse the fill char */
+    if (end-ptr >= 2 && is_alignment_token(ptr[1])) {
+        format->align = ptr[1];
+        format->fill_char = ptr[0];
+        fill_char_specified = 1;
+        align_specified = 1;
+        ptr += 2;
+    }
+    else if (end-ptr >= 1 && is_alignment_token(ptr[0])) {
+        format->align = ptr[0];
+        align_specified = 1;
+        ++ptr;
+    }
+
+    /* Parse the various sign options */
+    if (end-ptr >= 1 && is_sign_element(ptr[0])) {
+        format->sign = ptr[0];
+        ++ptr;
+    }
+
+    /* If the next character is #, we're in alternate mode.  This only
+       applies to integers. */
+    if (end-ptr >= 1 && ptr[0] == '#') {
+        format->alternate = 1;
+        ++ptr;
+    }
+
+    /* The special case for 0-padding (backwards compat) */
+    if (!fill_char_specified && end-ptr >= 1 && ptr[0] == '0') {
+        format->fill_char = '0';
+        if (!align_specified) {
+            format->align = '=';
+        }
+        ++ptr;
+    }
+
+    consumed = get_integer(&ptr, end, &format->width);
+    if (consumed == -1)
+        /* Overflow error. Exception already set. */
+        return 0;
+
+    /* If consumed is 0, we didn't consume any characters for the
+       width. In that case, reset the width to -1, because
+       get_integer() will have set it to zero. -1 is how we record
+       that the width wasn't specified. */
+    if (consumed == 0)
+        format->width = -1;
+
+    /* Comma signifies add thousands separators */
+    if (end-ptr && ptr[0] == ',') {
+        format->thousands_separators = 1;
+        ++ptr;
+    }
+
+    /* Parse field precision */
+    if (end-ptr && ptr[0] == '.') {
+        ++ptr;
+
+        consumed = get_integer(&ptr, end, &format->precision);
+        if (consumed == -1)
+            /* Overflow error. Exception already set. */
+            return 0;
+
+        /* Not having a precision after a dot is an error. */
+        if (consumed == 0) {
+            PyErr_Format(PyExc_ValueError,
+                         "Format specifier missing precision");
+            return 0;
+        }
+
+    }
+
+    /* Finally, parse the type field. */
+
+    if (end-ptr > 1) {
+        /* More than one char remain, invalid conversion spec. */
+        PyErr_Format(PyExc_ValueError, "Invalid conversion specification");
+        return 0;
+    }
+
+    if (end-ptr == 1) {
+        format->type = ptr[0];
+        ++ptr;
+    }
+
+    /* Do as much validating as we can, just by looking at the format
+       specifier.  Do not take into account what type of formatting
+       we're doing (int, float, string). */
+
+    if (format->thousands_separators) {
+        switch (format->type) {
+        case 'd':
+        case 'e':
+        case 'f':
+        case 'g':
+        case 'E':
+        case 'G':
+        case '%':
+        case 'F':
+        case '\0':
+            /* These are allowed. See PEP 378.*/
+            break;
+        default:
+            invalid_comma_type(format->type);
+            return 0;
+        }
+    }
+
+    return 1;
+}
+
+/* Calculate the padding needed. */
+static void
+calc_padding(Py_ssize_t nchars, Py_ssize_t width, STRINGLIB_CHAR align,
+             Py_ssize_t *n_lpadding, Py_ssize_t *n_rpadding,
+             Py_ssize_t *n_total)
+{
+    if (width >= 0) {
+        if (nchars > width)
+            *n_total = nchars;
+        else
+            *n_total = width;
+    }
+    else {
+        /* not specified, use all of the chars and no more */
+        *n_total = nchars;
+    }
+
+    /* Figure out how much leading space we need, based on the
+       aligning */
+    if (align == '>')
+        *n_lpadding = *n_total - nchars;
+    else if (align == '^')
+        *n_lpadding = (*n_total - nchars) / 2;
+    else if (align == '<' || align == '=')
+        *n_lpadding = 0;
+    else {
+        /* We should never have an unspecified alignment. */
+        *n_lpadding = 0;
+        assert(0);
+    }
+
+    *n_rpadding = *n_total - nchars - *n_lpadding;
+}
+
+/* Do the padding, and return a pointer to where the caller-supplied
+   content goes. */
+static STRINGLIB_CHAR *
+fill_padding(STRINGLIB_CHAR *p, Py_ssize_t nchars, STRINGLIB_CHAR fill_char,
+             Py_ssize_t n_lpadding, Py_ssize_t n_rpadding)
+{
+    /* Pad on left. */
+    if (n_lpadding)
+        STRINGLIB_FILL(p, fill_char, n_lpadding);
+
+    /* Pad on right. */
+    if (n_rpadding)
+        STRINGLIB_FILL(p + nchars + n_lpadding, fill_char, n_rpadding);
+
+    /* Pointer to the user content. */
+    return p + n_lpadding;
+}
+
+#if defined FORMAT_FLOAT || defined FORMAT_LONG || defined FORMAT_COMPLEX
+/************************************************************************/
+/*********** common routines for numeric formatting *********************/
+/************************************************************************/
+
+/* Locale type codes. */
+#define LT_CURRENT_LOCALE 0
+#define LT_DEFAULT_LOCALE 1
+#define LT_NO_LOCALE 2
+
+/* Locale info needed for formatting integers and the part of floats
+   before and including the decimal. Note that locales only support
+   8-bit chars, not unicode. */
+typedef struct {
+    char *decimal_point;
+    char *thousands_sep;
+    char *grouping;
+} LocaleInfo;
+
+/* describes the layout for an integer, see the comment in
+   calc_number_widths() for details */
+typedef struct {
+    Py_ssize_t n_lpadding;
+    Py_ssize_t n_prefix;
+    Py_ssize_t n_spadding;
+    Py_ssize_t n_rpadding;
+    char sign;
+    Py_ssize_t n_sign;      /* number of digits needed for sign (0/1) */
+    Py_ssize_t n_grouped_digits; /* Space taken up by the digits, including
+                                    any grouping chars. */
+    Py_ssize_t n_decimal;   /* 0 if only an integer */
+    Py_ssize_t n_remainder; /* Digits in decimal and/or exponent part,
+                               excluding the decimal itself, if
+                               present. */
+
+    /* These 2 are not the widths of fields, but are needed by
+       STRINGLIB_GROUPING. */
+    Py_ssize_t n_digits;    /* The number of digits before a decimal
+                               or exponent. */
+    Py_ssize_t n_min_width; /* The min_width we used when we computed
+                               the n_grouped_digits width. */
+} NumberFieldWidths;
+
+
+/* Given a number of the form:
+   digits[remainder]
+   where ptr points to the start and end points to the end, find where
+    the integer part ends. This could be a decimal, an exponent, both,
+    or neither.
+   If a decimal point is present, set *has_decimal and increment
+    remainder beyond it.
+   Results are undefined (but shouldn't crash) for improperly
+    formatted strings.
+*/
+static void
+parse_number(STRINGLIB_CHAR *ptr, Py_ssize_t len,
+             Py_ssize_t *n_remainder, int *has_decimal)
+{
+    STRINGLIB_CHAR *end = ptr + len;
+    STRINGLIB_CHAR *remainder;
+
+    while (ptr<end && isdigit(*ptr))
+        ++ptr;
+    remainder = ptr;
+
+    /* Does remainder start with a decimal point? */
+    *has_decimal = ptr<end && *remainder == '.';
+
+    /* Skip the decimal point. */
+    if (*has_decimal)
+        remainder++;
+
+    *n_remainder = end - remainder;
+}
+
+/* not all fields of format are used.  for example, precision is
+   unused.  should this take discrete params in order to be more clear
+   about what it does?  or is passing a single format parameter easier
+   and more efficient enough to justify a little obfuscation? */
+static Py_ssize_t
+calc_number_widths(NumberFieldWidths *spec, Py_ssize_t n_prefix,
+                   STRINGLIB_CHAR sign_char, STRINGLIB_CHAR *number,
+                   Py_ssize_t n_number, Py_ssize_t n_remainder,
+                   int has_decimal, const LocaleInfo *locale,
+                   const InternalFormatSpec *format)
+{
+    Py_ssize_t n_non_digit_non_padding;
+    Py_ssize_t n_padding;
+
+    spec->n_digits = n_number - n_remainder - (has_decimal?1:0);
+    spec->n_lpadding = 0;
+    spec->n_prefix = n_prefix;
+    spec->n_decimal = has_decimal ? strlen(locale->decimal_point) : 0;
+    spec->n_remainder = n_remainder;
+    spec->n_spadding = 0;
+    spec->n_rpadding = 0;
+    spec->sign = '\0';
+    spec->n_sign = 0;
+
+    /* the output will look like:
+       |                                                                                         |
+       | <lpadding> <sign> <prefix> <spadding> <grouped_digits> <decimal> <remainder> <rpadding> |
+       |                                                                                         |
+
+       sign is computed from format->sign and the actual
+       sign of the number
+
+       prefix is given (it's for the '0x' prefix)
+
+       digits is already known
+
+       the total width is either given, or computed from the
+       actual digits
+
+       only one of lpadding, spadding, and rpadding can be non-zero,
+       and it's calculated from the width and other fields
+    */
+
+    /* compute the various parts we're going to write */
+    switch (format->sign) {
+    case '+':
+        /* always put a + or - */
+        spec->n_sign = 1;
+        spec->sign = (sign_char == '-' ? '-' : '+');
+        break;
+    case ' ':
+        spec->n_sign = 1;
+        spec->sign = (sign_char == '-' ? '-' : ' ');
+        break;
+    default:
+        /* Not specified, or the default (-) */
+        if (sign_char == '-') {
+            spec->n_sign = 1;
+            spec->sign = '-';
+        }
+    }
+
+    /* The number of chars used for non-digits and non-padding. */
+    n_non_digit_non_padding = spec->n_sign + spec->n_prefix + spec->n_decimal +
+        spec->n_remainder;
+
+    /* min_width can go negative, that's okay. format->width == -1 means
+       we don't care. */
+    if (format->fill_char == '0' && format->align == '=')
+        spec->n_min_width = format->width - n_non_digit_non_padding;
+    else
+        spec->n_min_width = 0;
+
+    if (spec->n_digits == 0)
+        /* This case only occurs when using 'c' formatting, we need
+           to special case it because the grouping code always wants
+           to have at least one character. */
+        spec->n_grouped_digits = 0;
+    else
+        spec->n_grouped_digits = STRINGLIB_GROUPING(NULL, 0, NULL,
+                                                    spec->n_digits,
+                                                    spec->n_min_width,
+                                                    locale->grouping,
+                                                    locale->thousands_sep);
+
+    /* Given the desired width and the total of digit and non-digit
+       space we consume, see if we need any padding. format->width can
+       be negative (meaning no padding), but this code still works in
+       that case. */
+    n_padding = format->width -
+                        (n_non_digit_non_padding + spec->n_grouped_digits);
+    if (n_padding > 0) {
+        /* Some padding is needed. Determine if it's left, space, or right. */
+        switch (format->align) {
+        case '<':
+            spec->n_rpadding = n_padding;
+            break;
+        case '^':
+            spec->n_lpadding = n_padding / 2;
+            spec->n_rpadding = n_padding - spec->n_lpadding;
+            break;
+        case '=':
+            spec->n_spadding = n_padding;
+            break;
+        case '>':
+            spec->n_lpadding = n_padding;
+            break;
+        default:
+            /* Shouldn't get here, but treat it as '>' */
+            spec->n_lpadding = n_padding;
+            assert(0);
+            break;
+        }
+    }
+    return spec->n_lpadding + spec->n_sign + spec->n_prefix +
+        spec->n_spadding + spec->n_grouped_digits + spec->n_decimal +
+        spec->n_remainder + spec->n_rpadding;
+}
+
+/* Fill in the digit parts of a numbers's string representation,
+   as determined in calc_number_widths().
+   No error checking, since we know the buffer is the correct size. */
+static void
+fill_number(STRINGLIB_CHAR *buf, const NumberFieldWidths *spec,
+            STRINGLIB_CHAR *digits, Py_ssize_t n_digits,
+            STRINGLIB_CHAR *prefix, STRINGLIB_CHAR fill_char,
+            LocaleInfo *locale, int toupper)
+{
+    /* Used to keep track of digits, decimal, and remainder. */
+    STRINGLIB_CHAR *p = digits;
+
+#ifndef NDEBUG
+    Py_ssize_t r;
+#endif
+
+    if (spec->n_lpadding) {
+        STRINGLIB_FILL(buf, fill_char, spec->n_lpadding);
+        buf += spec->n_lpadding;
+    }
+    if (spec->n_sign == 1) {
+        *buf++ = spec->sign;
+    }
+    if (spec->n_prefix) {
+        memmove(buf,
+                prefix,
+                spec->n_prefix * sizeof(STRINGLIB_CHAR));
+        if (toupper) {
+            Py_ssize_t t;
+            for (t = 0; t < spec->n_prefix; ++t)
+                buf[t] = STRINGLIB_TOUPPER(buf[t]);
+        }
+        buf += spec->n_prefix;
+    }
+    if (spec->n_spadding) {
+        STRINGLIB_FILL(buf, fill_char, spec->n_spadding);
+        buf += spec->n_spadding;
+    }
+
+    /* Only for type 'c' special case, it has no digits. */
+    if (spec->n_digits != 0) {
+        /* Fill the digits with InsertThousandsGrouping. */
+#ifndef NDEBUG
+        r =
+#endif
+            STRINGLIB_GROUPING(buf, spec->n_grouped_digits, digits,
+                               spec->n_digits, spec->n_min_width,
+                               locale->grouping, locale->thousands_sep);
+#ifndef NDEBUG
+        assert(r == spec->n_grouped_digits);
+#endif
+        p += spec->n_digits;
+    }
+    if (toupper) {
+        Py_ssize_t t;
+        for (t = 0; t < spec->n_grouped_digits; ++t)
+            buf[t] = STRINGLIB_TOUPPER(buf[t]);
+    }
+    buf += spec->n_grouped_digits;
+
+    if (spec->n_decimal) {
+        Py_ssize_t t;
+        for (t = 0; t < spec->n_decimal; ++t)
+            buf[t] = locale->decimal_point[t];
+        buf += spec->n_decimal;
+        p += 1;
+    }
+
+    if (spec->n_remainder) {
+        memcpy(buf, p, spec->n_remainder * sizeof(STRINGLIB_CHAR));
+        buf += spec->n_remainder;
+        p += spec->n_remainder;
+    }
+
+    if (spec->n_rpadding) {
+        STRINGLIB_FILL(buf, fill_char, spec->n_rpadding);
+        buf += spec->n_rpadding;
+    }
+}
+
+static char no_grouping[1] = {CHAR_MAX};
+
+/* Find the decimal point character(s?), thousands_separator(s?), and
+   grouping description, either for the current locale if type is
+   LT_CURRENT_LOCALE, a hard-coded locale if LT_DEFAULT_LOCALE, or
+   none if LT_NO_LOCALE. */
+static void
+get_locale_info(int type, LocaleInfo *locale_info)
+{
+    switch (type) {
+    case LT_CURRENT_LOCALE: {
+        struct lconv *locale_data = localeconv();
+        locale_info->decimal_point = locale_data->decimal_point;
+        locale_info->thousands_sep = locale_data->thousands_sep;
+        locale_info->grouping = locale_data->grouping;
+        break;
+    }
+    case LT_DEFAULT_LOCALE:
+        locale_info->decimal_point = ".";
+        locale_info->thousands_sep = ",";
+        locale_info->grouping = "\3"; /* Group every 3 characters.  The
+                                         (implicit) trailing 0 means repeat
+                                         infinitely. */
+        break;
+    case LT_NO_LOCALE:
+        locale_info->decimal_point = ".";
+        locale_info->thousands_sep = "";
+        locale_info->grouping = no_grouping;
+        break;
+    default:
+        assert(0);
+    }
+}
+
+#endif /* FORMAT_FLOAT || FORMAT_LONG || FORMAT_COMPLEX */
+
+/************************************************************************/
+/*********** string formatting ******************************************/
+/************************************************************************/
+
+static PyObject *
+format_string_internal(PyObject *value, const InternalFormatSpec *format)
+{
+    Py_ssize_t lpad;
+    Py_ssize_t rpad;
+    Py_ssize_t total;
+    STRINGLIB_CHAR *p;
+    Py_ssize_t len = STRINGLIB_LEN(value);
+    PyObject *result = NULL;
+
+    /* sign is not allowed on strings */
+    if (format->sign != '\0') {
+        PyErr_SetString(PyExc_ValueError,
+                        "Sign not allowed in string format specifier");
+        goto done;
+    }
+
+    /* alternate is not allowed on strings */
+    if (format->alternate) {
+        PyErr_SetString(PyExc_ValueError,
+                        "Alternate form (#) not allowed in string format "
+                        "specifier");
+        goto done;
+    }
+
+    /* '=' alignment not allowed on strings */
+    if (format->align == '=') {
+        PyErr_SetString(PyExc_ValueError,
+                        "'=' alignment not allowed "
+                        "in string format specifier");
+        goto done;
+    }
+
+    /* if precision is specified, output no more that format.precision
+       characters */
+    if (format->precision >= 0 && len >= format->precision) {
+        len = format->precision;
+    }
+
+    calc_padding(len, format->width, format->align, &lpad, &rpad, &total);
+
+    /* allocate the resulting string */
+    result = STRINGLIB_NEW(NULL, total);
+    if (result == NULL)
+        goto done;
+
+    /* Write into that space. First the padding. */
+    p = fill_padding(STRINGLIB_STR(result), len,
+                     format->fill_char, lpad, rpad);
+
+    /* Then the source string. */
+    memcpy(p, STRINGLIB_STR(value), len * sizeof(STRINGLIB_CHAR));
+
+done:
+    return result;
+}
+
+
+/************************************************************************/
+/*********** long formatting ********************************************/
+/************************************************************************/
+
+#if defined FORMAT_LONG || defined FORMAT_INT
+typedef PyObject*
+(*IntOrLongToString)(PyObject *value, int base);
+
+static PyObject *
+format_int_or_long_internal(PyObject *value, const InternalFormatSpec *format,
+                            IntOrLongToString tostring)
+{
+    PyObject *result = NULL;
+    PyObject *tmp = NULL;
+    STRINGLIB_CHAR *pnumeric_chars;
+    STRINGLIB_CHAR numeric_char;
+    STRINGLIB_CHAR sign_char = '\0';
+    Py_ssize_t n_digits;       /* count of digits need from the computed
+                                  string */
+    Py_ssize_t n_remainder = 0; /* Used only for 'c' formatting, which
+                                   produces non-digits */
+    Py_ssize_t n_prefix = 0;   /* Count of prefix chars, (e.g., '0x') */
+    Py_ssize_t n_total;
+    STRINGLIB_CHAR *prefix = NULL;
+    NumberFieldWidths spec;
+    long x;
+
+    /* Locale settings, either from the actual locale or
+       from a hard-code pseudo-locale */
+    LocaleInfo locale;
+
+    /* no precision allowed on integers */
+    if (format->precision != -1) {
+        PyErr_SetString(PyExc_ValueError,
+                        "Precision not allowed in integer format specifier");
+        goto done;
+    }
+
+    /* special case for character formatting */
+    if (format->type == 'c') {
+        /* error to specify a sign */
+        if (format->sign != '\0') {
+            PyErr_SetString(PyExc_ValueError,
+                            "Sign not allowed with integer"
+                            " format specifier 'c'");
+            goto done;
+        }
+
+        /* Error to specify a comma. */
+        if (format->thousands_separators) {
+            PyErr_SetString(PyExc_ValueError,
+                            "Thousands separators not allowed with integer"
+                            " format specifier 'c'");
+            goto done;
+        }
+
+        /* taken from unicodeobject.c formatchar() */
+        /* Integer input truncated to a character */
+/* XXX: won't work for int */
+        x = PyLong_AsLong(value);
+        if (x == -1 && PyErr_Occurred())
+            goto done;
+#ifdef Py_UNICODE_WIDE
+        if (x < 0 || x > 0x10ffff) {
+            PyErr_SetString(PyExc_OverflowError,
+                            "%c arg not in range(0x110000) "
+                            "(wide Python build)");
+            goto done;
+        }
+#else
+        if (x < 0 || x > 0xffff) {
+            PyErr_SetString(PyExc_OverflowError,
+                            "%c arg not in range(0x10000) "
+                            "(narrow Python build)");
+            goto done;
+        }
+#endif
+        numeric_char = (STRINGLIB_CHAR)x;
+        pnumeric_chars = &numeric_char;
+        n_digits = 1;
+
+        /* As a sort-of hack, we tell calc_number_widths that we only
+           have "remainder" characters. calc_number_widths thinks
+           these are characters that don't get formatted, only copied
+           into the output string. We do this for 'c' formatting,
+           because the characters are likely to be non-digits. */
+        n_remainder = 1;
+    }
+    else {
+        int base;
+        int leading_chars_to_skip = 0;  /* Number of characters added by
+                                           PyNumber_ToBase that we want to
+                                           skip over. */
+
+        /* Compute the base and how many characters will be added by
+           PyNumber_ToBase */
+        switch (format->type) {
+        case 'b':
+            base = 2;
+            leading_chars_to_skip = 2; /* 0b */
+            break;
+        case 'o':
+            base = 8;
+            leading_chars_to_skip = 2; /* 0o */
+            break;
+        case 'x':
+        case 'X':
+            base = 16;
+            leading_chars_to_skip = 2; /* 0x */
+            break;
+        default:  /* shouldn't be needed, but stops a compiler warning */
+        case 'd':
+        case 'n':
+            base = 10;
+            break;
+        }
+
+        /* The number of prefix chars is the same as the leading
+           chars to skip */
+        if (format->alternate)
+            n_prefix = leading_chars_to_skip;
+
+        /* Do the hard part, converting to a string in a given base */
+        tmp = tostring(value, base);
+        if (tmp == NULL)
+            goto done;
+
+        pnumeric_chars = STRINGLIB_STR(tmp);
+        n_digits = STRINGLIB_LEN(tmp);
+
+        prefix = pnumeric_chars;
+
+        /* Remember not to modify what pnumeric_chars points to.  it
+           might be interned.  Only modify it after we copy it into a
+           newly allocated output buffer. */
+
+        /* Is a sign character present in the output?  If so, remember it
+           and skip it */
+        if (pnumeric_chars[0] == '-') {
+            sign_char = pnumeric_chars[0];
+            ++prefix;
+            ++leading_chars_to_skip;
+        }
+
+        /* Skip over the leading chars (0x, 0b, etc.) */
+        n_digits -= leading_chars_to_skip;
+        pnumeric_chars += leading_chars_to_skip;
+    }
+
+    /* Determine the grouping, separator, and decimal point, if any. */
+    get_locale_info(format->type == 'n' ? LT_CURRENT_LOCALE :
+                    (format->thousands_separators ?
+                     LT_DEFAULT_LOCALE :
+                     LT_NO_LOCALE),
+                    &locale);
+
+    /* Calculate how much memory we'll need. */
+    n_total = calc_number_widths(&spec, n_prefix, sign_char, pnumeric_chars,
+                       n_digits, n_remainder, 0, &locale, format);
+
+    /* Allocate the memory. */
+    result = STRINGLIB_NEW(NULL, n_total);
+    if (!result)
+        goto done;
+
+    /* Populate the memory. */
+    fill_number(STRINGLIB_STR(result), &spec, pnumeric_chars, n_digits,
+                prefix, format->fill_char, &locale, format->type == 'X');
+
+done:
+    Py_XDECREF(tmp);
+    return result;
+}
+#endif /* defined FORMAT_LONG || defined FORMAT_INT */
+
+/************************************************************************/
+/*********** float formatting *******************************************/
+/************************************************************************/
+
+#ifdef FORMAT_FLOAT
+#if STRINGLIB_IS_UNICODE
+static void
+strtounicode(Py_UNICODE *buffer, const char *charbuffer, Py_ssize_t len)
+{
+    Py_ssize_t i;
+    for (i = 0; i < len; ++i)
+        buffer[i] = (Py_UNICODE)charbuffer[i];
+}
+#endif
+
+/* much of this is taken from unicodeobject.c */
+static PyObject *
+format_float_internal(PyObject *value,
+                      const InternalFormatSpec *format)
+{
+    char *buf = NULL;       /* buffer returned from PyOS_double_to_string */
+    Py_ssize_t n_digits;
+    Py_ssize_t n_remainder;
+    Py_ssize_t n_total;
+    int has_decimal;
+    double val;
+    Py_ssize_t precision;
+    Py_ssize_t default_precision = 6;
+    STRINGLIB_CHAR type = format->type;
+    int add_pct = 0;
+    STRINGLIB_CHAR *p;
+    NumberFieldWidths spec;
+    int flags = 0;
+    PyObject *result = NULL;
+    STRINGLIB_CHAR sign_char = '\0';
+    int float_type; /* Used to see if we have a nan, inf, or regular float. */
+
+#if STRINGLIB_IS_UNICODE
+    Py_UNICODE *unicode_tmp = NULL;
+#endif
+
+    /* Locale settings, either from the actual locale or
+       from a hard-code pseudo-locale */
+    LocaleInfo locale;
+
+    if (format->precision > INT_MAX) {
+        PyErr_SetString(PyExc_ValueError, "precision too big");
+        goto done;
+    }
+    precision = (int)format->precision;
+
+    /* Alternate is not allowed on floats. */
+    if (format->alternate) {
+        PyErr_SetString(PyExc_ValueError,
+                        "Alternate form (#) not allowed in float format "
+                        "specifier");
+        goto done;
+    }
+
+    if (type == '\0') {
+        /* Omitted type specifier. This is like 'g' but with at least one
+           digit after the decimal point, and different default precision.*/
+        type = 'g';
+        default_precision = PyFloat_STR_PRECISION;
+        flags |= Py_DTSF_ADD_DOT_0;
+    }
+
+    if (type == 'n')
+        /* 'n' is the same as 'g', except for the locale used to
+           format the result. We take care of that later. */
+        type = 'g';
+
+    val = PyFloat_AsDouble(value);
+    if (val == -1.0 && PyErr_Occurred())
+        goto done;
+
+    if (type == '%') {
+        type = 'f';
+        val *= 100;
+        add_pct = 1;
+    }
+
+    if (precision < 0)
+        precision = default_precision;
+
+    /* Cast "type", because if we're in unicode we need to pass a
+       8-bit char. This is safe, because we've restricted what "type"
+       can be. */
+    buf = PyOS_double_to_string(val, (char)type, precision, flags,
+                                &float_type);
+    if (buf == NULL)
+        goto done;
+    n_digits = strlen(buf);
+
+    if (add_pct) {
+        /* We know that buf has a trailing zero (since we just called
+           strlen() on it), and we don't use that fact any more. So we
+           can just write over the trailing zero. */
+        buf[n_digits] = '%';
+        n_digits += 1;
+    }
+
+    /* Since there is no unicode version of PyOS_double_to_string,
+       just use the 8 bit version and then convert to unicode. */
+#if STRINGLIB_IS_UNICODE
+    unicode_tmp = (Py_UNICODE*)PyMem_Malloc((n_digits)*sizeof(Py_UNICODE));
+    if (unicode_tmp == NULL) {
+        PyErr_NoMemory();
+        goto done;
+    }
+    strtounicode(unicode_tmp, buf, n_digits);
+    p = unicode_tmp;
+#else
+    p = buf;
+#endif
+
+    /* Is a sign character present in the output?  If so, remember it
+       and skip it */
+    if (*p == '-') {
+        sign_char = *p;
+        ++p;
+        --n_digits;
+    }
+
+    /* Determine if we have any "remainder" (after the digits, might include
+       decimal or exponent or both (or neither)) */
+    parse_number(p, n_digits, &n_remainder, &has_decimal);
+
+    /* Determine the grouping, separator, and decimal point, if any. */
+    get_locale_info(format->type == 'n' ? LT_CURRENT_LOCALE :
+                    (format->thousands_separators ?
+                     LT_DEFAULT_LOCALE :
+                     LT_NO_LOCALE),
+                    &locale);
+
+    /* Calculate how much memory we'll need. */
+    n_total = calc_number_widths(&spec, 0, sign_char, p, n_digits,
+                                 n_remainder, has_decimal, &locale, format);
+
+    /* Allocate the memory. */
+    result = STRINGLIB_NEW(NULL, n_total);
+    if (result == NULL)
+        goto done;
+
+    /* Populate the memory. */
+    fill_number(STRINGLIB_STR(result), &spec, p, n_digits, NULL,
+                format->fill_char, &locale, 0);
+
+done:
+    PyMem_Free(buf);
+#if STRINGLIB_IS_UNICODE
+    PyMem_Free(unicode_tmp);
+#endif
+    return result;
+}
+#endif /* FORMAT_FLOAT */
+
+/************************************************************************/
+/*********** complex formatting *****************************************/
+/************************************************************************/
+
+#ifdef FORMAT_COMPLEX
+
+static PyObject *
+format_complex_internal(PyObject *value,
+                        const InternalFormatSpec *format)
+{
+    double re;
+    double im;
+    char *re_buf = NULL;       /* buffer returned from PyOS_double_to_string */
+    char *im_buf = NULL;       /* buffer returned from PyOS_double_to_string */
+
+    InternalFormatSpec tmp_format = *format;
+    Py_ssize_t n_re_digits;
+    Py_ssize_t n_im_digits;
+    Py_ssize_t n_re_remainder;
+    Py_ssize_t n_im_remainder;
+    Py_ssize_t n_re_total;
+    Py_ssize_t n_im_total;
+    int re_has_decimal;
+    int im_has_decimal;
+    Py_ssize_t precision;
+    Py_ssize_t default_precision = 6;
+    STRINGLIB_CHAR type = format->type;
+    STRINGLIB_CHAR *p_re;
+    STRINGLIB_CHAR *p_im;
+    NumberFieldWidths re_spec;
+    NumberFieldWidths im_spec;
+    int flags = 0;
+    PyObject *result = NULL;
+    STRINGLIB_CHAR *p;
+    STRINGLIB_CHAR re_sign_char = '\0';
+    STRINGLIB_CHAR im_sign_char = '\0';
+    int re_float_type; /* Used to see if we have a nan, inf, or regular float. */
+    int im_float_type;
+    int add_parens = 0;
+    int skip_re = 0;
+    Py_ssize_t lpad;
+    Py_ssize_t rpad;
+    Py_ssize_t total;
+
+#if STRINGLIB_IS_UNICODE
+    Py_UNICODE *re_unicode_tmp = NULL;
+    Py_UNICODE *im_unicode_tmp = NULL;
+#endif
+
+    /* Locale settings, either from the actual locale or
+       from a hard-code pseudo-locale */
+    LocaleInfo locale;
+
+    if (format->precision > INT_MAX) {
+        PyErr_SetString(PyExc_ValueError, "precision too big");
+        goto done;
+    }
+    precision = (int)format->precision;
+
+    /* Alternate is not allowed on complex. */
+    if (format->alternate) {
+        PyErr_SetString(PyExc_ValueError,
+                        "Alternate form (#) not allowed in complex format "
+                        "specifier");
+        goto done;
+    }
+
+    /* Neither is zero pading. */
+    if (format->fill_char == '0') {
+        PyErr_SetString(PyExc_ValueError,
+                        "Zero padding is not allowed in complex format "
+                        "specifier");
+        goto done;
+    }
+
+    /* Neither is '=' alignment . */
+    if (format->align == '=') {
+        PyErr_SetString(PyExc_ValueError,
+                        "'=' alignment flag is not allowed in complex format "
+                        "specifier");
+        goto done;
+    }
+
+    re = PyComplex_RealAsDouble(value);
+    if (re == -1.0 && PyErr_Occurred())
+        goto done;
+    im = PyComplex_ImagAsDouble(value);
+    if (im == -1.0 && PyErr_Occurred())
+        goto done;
+
+    if (type == '\0') {
+        /* Omitted type specifier. Should be like str(self). */
+        type = 'g';
+        default_precision = PyFloat_STR_PRECISION;
+        if (re == 0.0 && copysign(1.0, re) == 1.0)
+            skip_re = 1;
+        else
+            add_parens = 1;
+    }
+
+    if (type == 'n')
+        /* 'n' is the same as 'g', except for the locale used to
+           format the result. We take care of that later. */
+        type = 'g';
+
+    if (precision < 0)
+        precision = default_precision;
+
+    /* Cast "type", because if we're in unicode we need to pass a
+       8-bit char. This is safe, because we've restricted what "type"
+       can be. */
+    re_buf = PyOS_double_to_string(re, (char)type, precision, flags,
+                                   &re_float_type);
+    if (re_buf == NULL)
+        goto done;
+    im_buf = PyOS_double_to_string(im, (char)type, precision, flags,
+                                   &im_float_type);
+    if (im_buf == NULL)
+        goto done;
+
+    n_re_digits = strlen(re_buf);
+    n_im_digits = strlen(im_buf);
+
+    /* Since there is no unicode version of PyOS_double_to_string,
+       just use the 8 bit version and then convert to unicode. */
+#if STRINGLIB_IS_UNICODE
+    re_unicode_tmp = (Py_UNICODE*)PyMem_Malloc((n_re_digits)*sizeof(Py_UNICODE));
+    if (re_unicode_tmp == NULL) {
+        PyErr_NoMemory();
+        goto done;
+    }
+    strtounicode(re_unicode_tmp, re_buf, n_re_digits);
+    p_re = re_unicode_tmp;
+
+    im_unicode_tmp = (Py_UNICODE*)PyMem_Malloc((n_im_digits)*sizeof(Py_UNICODE));
+    if (im_unicode_tmp == NULL) {
+        PyErr_NoMemory();
+        goto done;
+    }
+    strtounicode(im_unicode_tmp, im_buf, n_im_digits);
+    p_im = im_unicode_tmp;
+#else
+    p_re = re_buf;
+    p_im = im_buf;
+#endif
+
+    /* Is a sign character present in the output?  If so, remember it
+       and skip it */
+    if (*p_re == '-') {
+        re_sign_char = *p_re;
+        ++p_re;
+        --n_re_digits;
+    }
+    if (*p_im == '-') {
+        im_sign_char = *p_im;
+        ++p_im;
+        --n_im_digits;
+    }
+
+    /* Determine if we have any "remainder" (after the digits, might include
+       decimal or exponent or both (or neither)) */
+    parse_number(p_re, n_re_digits, &n_re_remainder, &re_has_decimal);
+    parse_number(p_im, n_im_digits, &n_im_remainder, &im_has_decimal);
+
+    /* Determine the grouping, separator, and decimal point, if any. */
+    get_locale_info(format->type == 'n' ? LT_CURRENT_LOCALE :
+                    (format->thousands_separators ?
+                     LT_DEFAULT_LOCALE :
+                     LT_NO_LOCALE),
+                    &locale);
+
+    /* Turn off any padding. We'll do it later after we've composed
+       the numbers without padding. */
+    tmp_format.fill_char = '\0';
+    tmp_format.align = '<';
+    tmp_format.width = -1;
+
+    /* Calculate how much memory we'll need. */
+    n_re_total = calc_number_widths(&re_spec, 0, re_sign_char, p_re,
+                                    n_re_digits, n_re_remainder,
+                                    re_has_decimal, &locale, &tmp_format);
+
+    /* Same formatting, but always include a sign, unless the real part is
+     * going to be omitted, in which case we use whatever sign convention was
+     * requested by the original format. */
+    if (!skip_re)
+        tmp_format.sign = '+';
+    n_im_total = calc_number_widths(&im_spec, 0, im_sign_char, p_im,
+                                    n_im_digits, n_im_remainder,
+                                    im_has_decimal, &locale, &tmp_format);
+
+    if (skip_re)
+        n_re_total = 0;
+
+    /* Add 1 for the 'j', and optionally 2 for parens. */
+    calc_padding(n_re_total + n_im_total + 1 + add_parens * 2,
+                 format->width, format->align, &lpad, &rpad, &total);
+
+    result = STRINGLIB_NEW(NULL, total);
+    if (result == NULL)
+        goto done;
+
+    /* Populate the memory. First, the padding. */
+    p = fill_padding(STRINGLIB_STR(result),
+                     n_re_total + n_im_total + 1 + add_parens * 2,
+                     format->fill_char, lpad, rpad);
+
+    if (add_parens)
+        *p++ = '(';
+
+    if (!skip_re) {
+        fill_number(p, &re_spec, p_re, n_re_digits, NULL, 0, &locale, 0);
+        p += n_re_total;
+    }
+    fill_number(p, &im_spec, p_im, n_im_digits, NULL, 0, &locale, 0);
+    p += n_im_total;
+    *p++ = 'j';
+
+    if (add_parens)
+        *p++ = ')';
+
+done:
+    PyMem_Free(re_buf);
+    PyMem_Free(im_buf);
+#if STRINGLIB_IS_UNICODE
+    PyMem_Free(re_unicode_tmp);
+    PyMem_Free(im_unicode_tmp);
+#endif
+    return result;
+}
+#endif /* FORMAT_COMPLEX */
+
+/************************************************************************/
+/*********** built in formatters ****************************************/
+/************************************************************************/
+PyObject *
+FORMAT_STRING(PyObject *obj,
+              STRINGLIB_CHAR *format_spec,
+              Py_ssize_t format_spec_len)
+{
+    InternalFormatSpec format;
+    PyObject *result = NULL;
+
+    /* check for the special case of zero length format spec, make
+       it equivalent to str(obj) */
+    if (format_spec_len == 0) {
+        result = STRINGLIB_TOSTR(obj);
+        goto done;
+    }
+
+    /* parse the format_spec */
+    if (!parse_internal_render_format_spec(format_spec, format_spec_len,
+                                           &format, 's', '<'))
+        goto done;
+
+    /* type conversion? */
+    switch (format.type) {
+    case 's':
+        /* no type conversion needed, already a string.  do the formatting */
+        result = format_string_internal(obj, &format);
+        break;
+    default:
+        /* unknown */
+        unknown_presentation_type(format.type, obj->ob_type->tp_name);
+        goto done;
+    }
+
+done:
+    return result;
+}
+
+#if defined FORMAT_LONG || defined FORMAT_INT
+static PyObject*
+format_int_or_long(PyObject* obj,
+                   STRINGLIB_CHAR *format_spec,
+                   Py_ssize_t format_spec_len,
+                   IntOrLongToString tostring)
+{
+    PyObject *result = NULL;
+    PyObject *tmp = NULL;
+    InternalFormatSpec format;
+
+    /* check for the special case of zero length format spec, make
+       it equivalent to str(obj) */
+    if (format_spec_len == 0) {
+        result = STRINGLIB_TOSTR(obj);
+        goto done;
+    }
+
+    /* parse the format_spec */
+    if (!parse_internal_render_format_spec(format_spec,
+                                           format_spec_len,
+                                           &format, 'd', '>'))
+        goto done;
+
+    /* type conversion? */
+    switch (format.type) {
+    case 'b':
+    case 'c':
+    case 'd':
+    case 'o':
+    case 'x':
+    case 'X':
+    case 'n':
+        /* no type conversion needed, already an int (or long).  do
+           the formatting */
+            result = format_int_or_long_internal(obj, &format, tostring);
+        break;
+
+    case 'e':
+    case 'E':
+    case 'f':
+    case 'F':
+    case 'g':
+    case 'G':
+    case '%':
+        /* convert to float */
+        tmp = PyNumber_Float(obj);
+        if (tmp == NULL)
+            goto done;
+        result = format_float_internal(tmp, &format);
+        break;
+
+    default:
+        /* unknown */
+        unknown_presentation_type(format.type, obj->ob_type->tp_name);
+        goto done;
+    }
+
+done:
+    Py_XDECREF(tmp);
+    return result;
+}
+#endif /* FORMAT_LONG || defined FORMAT_INT */
+
+#ifdef FORMAT_LONG
+/* Need to define long_format as a function that will convert a long
+   to a string.  In 3.0, _PyLong_Format has the correct signature.  In
+   2.x, we need to fudge a few parameters */
+#if PY_VERSION_HEX >= 0x03000000
+#define long_format _PyLong_Format
+#else
+static PyObject*
+long_format(PyObject* value, int base)
+{
+    /* Convert to base, don't add trailing 'L', and use the new octal
+       format. We already know this is a long object */
+    assert(PyLong_Check(value));
+    /* convert to base, don't add 'L', and use the new octal format */
+    return _PyLong_Format(value, base, 0, 1);
+}
+#endif
+
+PyObject *
+FORMAT_LONG(PyObject *obj,
+            STRINGLIB_CHAR *format_spec,
+            Py_ssize_t format_spec_len)
+{
+    return format_int_or_long(obj, format_spec, format_spec_len,
+                              long_format);
+}
+#endif /* FORMAT_LONG */
+
+#ifdef FORMAT_INT
+/* this is only used for 2.x, not 3.0 */
+static PyObject*
+int_format(PyObject* value, int base)
+{
+    /* Convert to base, and use the new octal format. We already
+       know this is an int object */
+    assert(PyInt_Check(value));
+    return _PyInt_Format((PyIntObject*)value, base, 1);
+}
+
+PyObject *
+FORMAT_INT(PyObject *obj,
+           STRINGLIB_CHAR *format_spec,
+           Py_ssize_t format_spec_len)
+{
+    return format_int_or_long(obj, format_spec, format_spec_len,
+                              int_format);
+}
+#endif /* FORMAT_INT */
+
+#ifdef FORMAT_FLOAT
+PyObject *
+FORMAT_FLOAT(PyObject *obj,
+             STRINGLIB_CHAR *format_spec,
+             Py_ssize_t format_spec_len)
+{
+    PyObject *result = NULL;
+    InternalFormatSpec format;
+
+    /* check for the special case of zero length format spec, make
+       it equivalent to str(obj) */
+    if (format_spec_len == 0) {
+        result = STRINGLIB_TOSTR(obj);
+        goto done;
+    }
+
+    /* parse the format_spec */
+    if (!parse_internal_render_format_spec(format_spec,
+                                           format_spec_len,
+                                           &format, '\0', '>'))
+        goto done;
+
+    /* type conversion? */
+    switch (format.type) {
+    case '\0': /* No format code: like 'g', but with at least one decimal. */
+    case 'e':
+    case 'E':
+    case 'f':
+    case 'F':
+    case 'g':
+    case 'G':
+    case 'n':
+    case '%':
+        /* no conversion, already a float.  do the formatting */
+        result = format_float_internal(obj, &format);
+        break;
+
+    default:
+        /* unknown */
+        unknown_presentation_type(format.type, obj->ob_type->tp_name);
+        goto done;
+    }
+
+done:
+    return result;
+}
+#endif /* FORMAT_FLOAT */
+
+#ifdef FORMAT_COMPLEX
+PyObject *
+FORMAT_COMPLEX(PyObject *obj,
+               STRINGLIB_CHAR *format_spec,
+               Py_ssize_t format_spec_len)
+{
+    PyObject *result = NULL;
+    InternalFormatSpec format;
+
+    /* check for the special case of zero length format spec, make
+       it equivalent to str(obj) */
+    if (format_spec_len == 0) {
+        result = STRINGLIB_TOSTR(obj);
+        goto done;
+    }
+
+    /* parse the format_spec */
+    if (!parse_internal_render_format_spec(format_spec,
+                                           format_spec_len,
+                                           &format, '\0', '>'))
+        goto done;
+
+    /* type conversion? */
+    switch (format.type) {
+    case '\0': /* No format code: like 'g', but with at least one decimal. */
+    case 'e':
+    case 'E':
+    case 'f':
+    case 'F':
+    case 'g':
+    case 'G':
+    case 'n':
+        /* no conversion, already a complex.  do the formatting */
+        result = format_complex_internal(obj, &format);
+        break;
+
+    default:
+        /* unknown */
+        unknown_presentation_type(format.type, obj->ob_type->tp_name);
+        goto done;
+    }
+
+done:
+    return result;
+}
+#endif /* FORMAT_COMPLEX */

lib_python/2.7_Objects/stringlib/localeutil.h

+/* stringlib: locale related helpers implementation */
+
+#ifndef STRINGLIB_LOCALEUTIL_H
+#define STRINGLIB_LOCALEUTIL_H
+
+#include <locale.h>
+
+#define MAX(x, y) ((x) < (y) ? (y) : (x))
+#define MIN(x, y) ((x) < (y) ? (x) : (y))
+
+typedef struct {
+    const char *grouping;
+    char previous;
+    Py_ssize_t i; /* Where we're currently pointing in grouping. */
+} GroupGenerator;
+
+static void
+_GroupGenerator_init(GroupGenerator *self, const char *grouping)
+{
+    self->grouping = grouping;
+    self->i = 0;
+    self->previous = 0;
+}
+
+/* Returns the next grouping, or 0 to signify end. */
+static Py_ssize_t
+_GroupGenerator_next(GroupGenerator *self)
+{
+    /* Note that we don't really do much error checking here. If a
+       grouping string contains just CHAR_MAX, for example, then just
+       terminate the generator. That shouldn't happen, but at least we
+       fail gracefully. */
+    switch (self->grouping[self->i]) {
+    case 0:
+        return self->previous;
+    case CHAR_MAX:
+        /* Stop the generator. */
+        return 0;
+    default: {
+        char ch = self->grouping[self->i];
+        self->previous = ch;
+        self->i++;
+        return (Py_ssize_t)ch;
+    }
+    }
+}
+
+/* Fill in some digits, leading zeros, and thousands separator. All
+   are optional, depending on when we're called. */
+static void
+fill(STRINGLIB_CHAR **digits_end, STRINGLIB_CHAR **buffer_end,
+     Py_ssize_t n_chars, Py_ssize_t n_zeros, const char* thousands_sep,
+     Py_ssize_t thousands_sep_len)
+{
+#if STRINGLIB_IS_UNICODE
+    Py_ssize_t i;
+#endif
+
+    if (thousands_sep) {
+        *buffer_end -= thousands_sep_len;
+
+        /* Copy the thousands_sep chars into the buffer. */
+#if STRINGLIB_IS_UNICODE
+        /* Convert from the char's of the thousands_sep from
+           the locale into unicode. */
+        for (i = 0; i < thousands_sep_len; ++i)
+            (*buffer_end)[i] = thousands_sep[i];
+#else
+        /* No conversion, just memcpy the thousands_sep. */
+        memcpy(*buffer_end, thousands_sep, thousands_sep_len);
+#endif
+    }
+
+    *buffer_end -= n_chars;
+    *digits_end -= n_chars;
+    memcpy(*buffer_end, *digits_end, n_chars * sizeof(STRINGLIB_CHAR));
+
+    *buffer_end -= n_zeros;
+    STRINGLIB_FILL(*buffer_end, '0', n_zeros);
+}
+
+/**
+ * _Py_InsertThousandsGrouping:
+ * @buffer: A pointer to the start of a string.
+ * @n_buffer: Number of characters in @buffer.
+ * @digits: A pointer to the digits we're reading from. If count
+ *          is non-NULL, this is unused.
+ * @n_digits: The number of digits in the string, in which we want
+ *            to put the grouping chars.
+ * @min_width: The minimum width of the digits in the output string.
+ *             Output will be zero-padded on the left to fill.
+ * @grouping: see definition in localeconv().
+ * @thousands_sep: see definition in localeconv().
+ *
+ * There are 2 modes: counting and filling. If @buffer is NULL,
+ *  we are in counting mode, else filling mode.
+ * If counting, the required buffer size is returned.
+ * If filling, we know the buffer will be large enough, so we don't
+ *  need to pass in the buffer size.
+ * Inserts thousand grouping characters (as defined by grouping and
+ *  thousands_sep) into the string between buffer and buffer+n_digits.
+ *
+ * Return value: 0 on error, else 1.  Note that no error can occur if
+ *  count is non-NULL.
+ *
+ * This name won't be used, the includer of this file should define
+ *  it to be the actual function name, based on unicode or string.
+ *
+ * As closely as possible, this code mimics the logic in decimal.py's
+    _insert_thousands_sep().
+ **/
+Py_ssize_t
+_Py_InsertThousandsGrouping(STRINGLIB_CHAR *buffer,
+                            Py_ssize_t n_buffer,
+                            STRINGLIB_CHAR *digits,
+                            Py_ssize_t n_digits,
+                            Py_ssize_t min_width,
+                            const char *grouping,
+                            const char *thousands_sep)
+{
+    Py_ssize_t count = 0;
+    Py_ssize_t n_zeros;
+    int loop_broken = 0;
+    int use_separator = 0; /* First time through, don't append the
+                              separator. They only go between
+                              groups. */
+    STRINGLIB_CHAR *buffer_end = NULL;
+    STRINGLIB_CHAR *digits_end = NULL;
+    Py_ssize_t l;
+    Py_ssize_t n_chars;
+    Py_ssize_t thousands_sep_len = strlen(thousands_sep);
+    Py_ssize_t remaining = n_digits; /* Number of chars remaining to
+                                        be looked at */
+    /* A generator that returns all of the grouping widths, until it
+       returns 0. */
+    GroupGenerator groupgen;
+    _GroupGenerator_init(&groupgen, grouping);
+
+    if (buffer) {
+        buffer_end = buffer + n_buffer;
+        digits_end = digits + n_digits;
+    }
+
+    while ((l = _GroupGenerator_next(&groupgen)) > 0) {
+        l = MIN(l, MAX(MAX(remaining, min_width), 1));
+        n_zeros = MAX(0, l - remaining);
+        n_chars = MAX(0, MIN(remaining, l));
+
+        /* Use n_zero zero's and n_chars chars */
+
+        /* Count only, don't do anything. */
+        count += (use_separator ? thousands_sep_len : 0) + n_zeros + n_chars;
+
+        if (buffer) {
+            /* Copy into the output buffer. */
+            fill(&digits_end, &buffer_end, n_chars, n_zeros,
+                 use_separator ? thousands_sep : NULL, thousands_sep_len);
+        }
+
+        /* Use a separator next time. */
+        use_separator = 1;
+
+        remaining -= n_chars;
+        min_width -= l;
+
+        if (remaining <= 0 && min_width <= 0) {
+            loop_broken = 1;
+            break;
+        }
+        min_width -= thousands_sep_len;
+    }
+    if (!loop_broken) {
+        /* We left the loop without using a break statement. */
+
+        l = MAX(MAX(remaining, min_width), 1);
+        n_zeros = MAX(0, l - remaining);
+        n_chars = MAX(0, MIN(remaining, l));
+
+        /* Use n_zero zero's and n_chars chars */
+        count += (use_separator ? thousands_sep_len : 0) + n_zeros + n_chars;
+        if (buffer) {
+            /* Copy into the output buffer. */
+            fill(&digits_end, &buffer_end, n_chars, n_zeros,
+                 use_separator ? thousands_sep : NULL, thousands_sep_len);
+        }
+    }
+    return count;
+}
+
+/**
+ * _Py_InsertThousandsGroupingLocale:
+ * @buffer: A pointer to the start of a string.
+ * @n_digits: The number of digits in the string, in which we want
+ *            to put the grouping chars.
+ *
+ * Reads thee current locale and calls _Py_InsertThousandsGrouping().
+ **/
+Py_ssize_t
+_Py_InsertThousandsGroupingLocale(STRINGLIB_CHAR *buffer,
+                                  Py_ssize_t n_buffer,
+                                  STRINGLIB_CHAR *digits,
+                                  Py_ssize_t n_digits,
+                                  Py_ssize_t min_width)
+{
+        struct lconv *locale_data = localeconv();
+        const char *grouping = locale_data->grouping;
+        const char *thousands_sep = locale_data->thousands_sep;
+
+        return _Py_InsertThousandsGrouping(buffer, n_buffer, digits, n_digits,
+                                           min_width, grouping, thousands_sep);
+}
+#endif /* STRINGLIB_LOCALEUTIL_H */

lib_python/2.7_Objects/stringlib/partition.h

+/* stringlib: partition implementation */
+
+#ifndef STRINGLIB_PARTITION_H
+#define STRINGLIB_PARTITION_H
+
+#ifndef STRINGLIB_FASTSEARCH_H
+#error must include "stringlib/fastsearch.h" before including this module
+#endif
+
+Py_LOCAL_INLINE(PyObject*)
+stringlib_partition(PyObject* str_obj,
+                    const STRINGLIB_CHAR* str, Py_ssize_t str_len,
+                    PyObject* sep_obj,
+                    const STRINGLIB_CHAR* sep, Py_ssize_t sep_len)
+{
+    PyObject* out;
+    Py_ssize_t pos;
+
+    if (sep_len == 0) {
+        PyErr_SetString(PyExc_ValueError, "empty separator");
+        return NULL;
+    }
+
+    out = PyTuple_New(3);
+    if (!out)
+        return NULL;
+
+    pos = fastsearch(str, str_len, sep, sep_len, -1, FAST_SEARCH);
+
+    if (pos < 0) {
+#if STRINGLIB_MUTABLE
+        PyTuple_SET_ITEM(out, 0, STRINGLIB_NEW(str, str_len));
+        PyTuple_SET_ITEM(out, 1, STRINGLIB_NEW(NULL, 0));
+        PyTuple_SET_ITEM(out, 2, STRINGLIB_NEW(NULL, 0));
+#else
+        Py_INCREF(str_obj);
+        PyTuple_SET_ITEM(out, 0, (PyObject*) str_obj);
+        Py_INCREF(STRINGLIB_EMPTY);
+        PyTuple_SET_ITEM(out, 1, (PyObject*) STRINGLIB_EMPTY);
+        Py_INCREF(STRINGLIB_EMPTY);
+        PyTuple_SET_ITEM(out, 2, (PyObject*) STRINGLIB_EMPTY);
+#endif
+        return out;
+    }
+
+    PyTuple_SET_ITEM(out, 0, STRINGLIB_NEW(str, pos));
+    Py_INCREF(sep_obj);
+    PyTuple_SET_ITEM(out, 1, sep_obj);
+    pos += sep_len;
+    PyTuple_SET_ITEM(out, 2, STRINGLIB_NEW(str + pos, str_len - pos));
+
+    if (PyErr_Occurred()) {
+        Py_DECREF(out);
+        return NULL;
+    }
+
+    return out;
+}
+
+Py_LOCAL_INLINE(PyObject*)
+stringlib_rpartition(PyObject* str_obj,
+                     const STRINGLIB_CHAR* str, Py_ssize_t str_len,
+                     PyObject* sep_obj,
+                     const STRINGLIB_CHAR* sep, Py_ssize_t sep_len)
+{
+    PyObject* out;
+    Py_ssize_t pos;
+
+    if (sep_len == 0) {
+        PyErr_SetString(PyExc_ValueError, "empty separator");
+        return NULL;
+    }
+
+    out = PyTuple_New(3);
+    if (!out)
+        return NULL;
+
+    pos = fastsearch(str, str_len, sep, sep_len, -1, FAST_RSEARCH);
+
+    if (pos < 0) {
+#if STRINGLIB_MUTABLE
+        PyTuple_SET_ITEM(out, 0, STRINGLIB_NEW(NULL, 0));
+        PyTuple_SET_ITEM(out, 1, STRINGLIB_NEW(NULL, 0));
+        PyTuple_SET_ITEM(out, 2, STRINGLIB_NEW(str, str_len));
+#else
+        Py_INCREF(STRINGLIB_EMPTY);
+        PyTuple_SET_ITEM(out, 0, (PyObject*) STRINGLIB_EMPTY);
+        Py_INCREF(STRINGLIB_EMPTY);
+        PyTuple_SET_ITEM(out, 1, (PyObject*) STRINGLIB_EMPTY);
+        Py_INCREF(str_obj);
+        PyTuple_SET_ITEM(out, 2, (PyObject*) str_obj);
+#endif
+        return out;
+    }
+
+    PyTuple_SET_ITEM(out, 0, STRINGLIB_NEW(str, pos));
+    Py_INCREF(sep_obj);
+    PyTuple_SET_ITEM(out, 1, sep_obj);
+    pos += sep_len;
+    PyTuple_SET_ITEM(out, 2, STRINGLIB_NEW(str + pos, str_len - pos));
+
+    if (PyErr_Occurred()) {
+        Py_DECREF(out);
+        return NULL;
+    }
+
+    return out;
+}
+
+#endif

lib_python/2.7_Objects/stringlib/split.h

+/* stringlib: split implementation */
+
+#ifndef STRINGLIB_SPLIT_H
+#define STRINGLIB_SPLIT_H
+
+#ifndef STRINGLIB_FASTSEARCH_H
+#error must include "stringlib/fastsearch.h" before including this module
+#endif
+
+/* Overallocate the initial list to reduce the number of reallocs for small
+   split sizes.  Eg, "A A A A A A A A A A".split() (10 elements) has three
+   resizes, to sizes 4, 8, then 16.  Most observed string splits are for human
+   text (roughly 11 words per line) and field delimited data (usually 1-10
+   fields).  For large strings the split algorithms are bandwidth limited
+   so increasing the preallocation likely will not improve things.*/
+
+#define MAX_PREALLOC 12
+
+/* 5 splits gives 6 elements */
+#define PREALLOC_SIZE(maxsplit) \
+    (maxsplit >= MAX_PREALLOC ? MAX_PREALLOC : maxsplit+1)
+
+#define SPLIT_APPEND(data, left, right)         \
+    sub = STRINGLIB_NEW((data) + (left),        \
+                        (right) - (left));      \
+    if (sub == NULL)                            \
+        goto onError;                           \
+    if (PyList_Append(list, sub)) {             \
+        Py_DECREF(sub);                         \
+        goto onError;                           \
+    }                                           \
+    else                                        \
+        Py_DECREF(sub);
+
+#define SPLIT_ADD(data, left, right) {          \
+    sub = STRINGLIB_NEW((data) + (left),        \
+                        (right) - (left));      \
+    if (sub == NULL)                            \
+        goto onError;                           \
+    if (count < MAX_PREALLOC) {                 \
+        PyList_SET_ITEM(list, count, sub);      \
+    } else {                                    \
+        if (PyList_Append(list, sub)) {         \
+            Py_DECREF(sub);                     \
+            goto onError;                       \
+        }                                       \
+        else                                    \
+            Py_DECREF(sub);                     \
+    }                                           \
+    count++; }
+
+
+/* Always force the list to the expected size. */
+#define FIX_PREALLOC_SIZE(list) Py_SIZE(list) = count
+
+Py_LOCAL_INLINE(PyObject *)
+stringlib_split_whitespace(PyObject* str_obj,
+                           const STRINGLIB_CHAR* str, Py_ssize_t str_len,
+                           Py_ssize_t maxcount)
+{
+    Py_ssize_t i, j, count=0;
+    PyObject *list = PyList_New(PREALLOC_SIZE(maxcount));
+    PyObject *sub;
+
+    if (list == NULL)
+        return NULL;
+
+    i = j = 0;
+    while (maxcount-- > 0) {
+        while (i < str_len && STRINGLIB_ISSPACE(str[i]))
+            i++;
+        if (i == str_len) break;
+        j = i; i++;
+        while (i < str_len && !STRINGLIB_ISSPACE(str[i]))
+            i++;
+#ifndef STRINGLIB_MUTABLE
+        if (j == 0 && i == str_len && STRINGLIB_CHECK_EXACT(str_obj)) {
+            /* No whitespace in str_obj, so just use it as list[0] */
+            Py_INCREF(str_obj);
+            PyList_SET_ITEM(list, 0, (PyObject *)str_obj);
+            count++;
+            break;
+        }
+#endif
+        SPLIT_ADD(str, j, i);
+    }
+
+    if (i < str_len) {
+        /* Only occurs when maxcount was reached */
+        /* Skip any remaining whitespace and copy to end of string */
+        while (i < str_len && STRINGLIB_ISSPACE(str[i]))
+            i++;
+        if (i != str_len)
+            SPLIT_ADD(str, i, str_len);
+    }
+    FIX_PREALLOC_SIZE(list);
+    return list;
+
+  onError:
+    Py_DECREF(list);
+    return NULL;
+}
+
+Py_LOCAL_INLINE(PyObject *)
+stringlib_split_char(PyObject* str_obj,
+                     const STRINGLIB_CHAR* str, Py_ssize_t str_len,
+                     const STRINGLIB_CHAR ch,
+                     Py_ssize_t maxcount)
+{
+    Py_ssize_t i, j, count=0;
+    PyObject *list = PyList_New(PREALLOC_SIZE(maxcount));
+    PyObject *sub;
+
+    if (list == NULL)
+        return NULL;
+
+    i = j = 0;
+    while ((j < str_len) && (maxcount-- > 0)) {
+        for(; j < str_len; j++) {
+            /* I found that using memchr makes no difference */
+            if (str[j] == ch) {
+                SPLIT_ADD(str, i, j);
+                i = j = j + 1;
+                break;
+            }
+        }
+    }
+#ifndef STRINGLIB_MUTABLE
+    if (count == 0 && STRINGLIB_CHECK_EXACT(str_obj)) {
+        /* ch not in str_obj, so just use str_obj as list[0] */
+        Py_INCREF(str_obj);
+        PyList_SET_ITEM(list, 0, (PyObject *)str_obj);
+        count++;
+    } else
+#endif
+    if (i <= str_len) {
+        SPLIT_ADD(str, i, str_len);
+    }
+    FIX_PREALLOC_SIZE(list);
+    return list;
+
+  onError:
+    Py_DECREF(list);
+    return NULL;
+}
+
+Py_LOCAL_INLINE(PyObject *)
+stringlib_split(PyObject* str_obj,
+                const STRINGLIB_CHAR* str, Py_ssize_t str_len,
+                const STRINGLIB_CHAR* sep, Py_ssize_t sep_len,
+                Py_ssize_t maxcount)
+{
+    Py_ssize_t i, j, pos, count=0;
+    PyObject *list, *sub;
+
+    if (sep_len == 0) {
+        PyErr_SetString(PyExc_ValueError, "empty separator");
+        return NULL;
+    }
+    else if (sep_len == 1)
+        return stringlib_split_char(str_obj, str, str_len, sep[0], maxcount);
+
+    list = PyList_New(PREALLOC_SIZE(maxcount));
+    if (list == NULL)
+        return NULL;
+
+    i = j = 0;
+    while (maxcount-- > 0) {
+        pos = fastsearch(str+i, str_len-i, sep, sep_len, -1, FAST_SEARCH);
+        if (pos < 0)
+            break;
+        j = i + pos;
+        SPLIT_ADD(str, i, j);
+        i = j + sep_len;
+    }
+#ifndef STRINGLIB_MUTABLE
+    if (count == 0 && STRINGLIB_CHECK_EXACT(str_obj)) {
+        /* No match in str_obj, so just use it as list[0] */
+        Py_INCREF(str_obj);
+        PyList_SET_ITEM(list, 0, (PyObject *)str_obj);
+        count++;
+    } else
+#endif
+    {
+        SPLIT_ADD(str, i, str_len);
+    }
+    FIX_PREALLOC_SIZE(list);
+    return list;
+
+  onError:
+    Py_DECREF(list);
+    return NULL;
+}
+
+Py_LOCAL_INLINE(PyObject *)
+stringlib_rsplit_whitespace(PyObject* str_obj,
+                            const STRINGLIB_CHAR* str, Py_ssize_t str_len,
+                            Py_ssize_t maxcount)
+{
+    Py_ssize_t i, j, count=0;
+    PyObject *list = PyList_New(PREALLOC_SIZE(maxcount));
+    PyObject *sub;
+
+    if (list == NULL)
+        return NULL;
+
+    i = j = str_len - 1;
+    while (maxcount-- > 0) {
+        while (i >= 0 && STRINGLIB_ISSPACE(str[i]))
+            i--;
+        if (i < 0) break;
+        j = i; i--;
+        while (i >= 0 && !STRINGLIB_ISSPACE(str[i]))
+            i--;
+#ifndef STRINGLIB_MUTABLE
+        if (j == str_len - 1 && i < 0 && STRINGLIB_CHECK_EXACT(str_obj)) {
+            /* No whitespace in str_obj, so just use it as list[0] */
+            Py_INCREF(str_obj);
+            PyList_SET_ITEM(list, 0, (PyObject *)str_obj);
+            count++;
+            break;
+        }
+#endif
+        SPLIT_ADD(str, i + 1, j + 1);
+    }
+
+    if (i >= 0) {
+        /* Only occurs when maxcount was reached */
+        /* Skip any remaining whitespace and copy to beginning of string */
+        while (i >= 0 && STRINGLIB_ISSPACE(str[i]))
+            i--;
+        if (i >= 0)
+            SPLIT_ADD(str, 0, i + 1);
+    }
+    FIX_PREALLOC_SIZE(list);
+    if (PyList_Reverse(list) < 0)
+        goto onError;
+    return list;
+
+  onError:
+    Py_DECREF(list);
+    return NULL;
+}
+
+Py_LOCAL_INLINE(PyObject *)
+stringlib_rsplit_char(PyObject* str_obj,
+                      const STRINGLIB_CHAR* str, Py_ssize_t str_len,
+                      const STRINGLIB_CHAR ch,
+                      Py_ssize_t maxcount)
+{
+    Py_ssize_t i, j, count=0;
+    PyObject *list = PyList_New(PREALLOC_SIZE(maxcount));
+    PyObject *sub;
+
+    if (list == NULL)
+        return NULL;
+
+    i = j = str_len - 1;
+    while ((i >= 0) && (maxcount-- > 0)) {
+        for(; i >= 0; i--) {
+            if (str[i] == ch) {
+                SPLIT_ADD(str, i + 1, j + 1);
+                j = i = i - 1;
+                break;
+            }
+        }
+    }
+#ifndef STRINGLIB_MUTABLE
+    if (count == 0 && STRINGLIB_CHECK_EXACT(str_obj)) {
+        /* ch not in str_obj, so just use str_obj as list[0] */
+        Py_INCREF(str_obj);
+        PyList_SET_ITEM(list, 0, (PyObject *)str_obj);
+        count++;
+    } else
+#endif
+    if (j >= -1) {
+        SPLIT_ADD(str, 0, j + 1);
+    }
+    FIX_PREALLOC_SIZE(list);
+    if (PyList_Reverse(list) < 0)
+        goto onError;
+    return list;
+
+  onError:
+    Py_DECREF(list);
+    return NULL;
+}
+
+Py_LOCAL_INLINE(PyObject *)
+stringlib_rsplit(PyObject* str_obj,
+                 const STRINGLIB_CHAR* str, Py_ssize_t str_len,
+                 const STRINGLIB_CHAR* sep, Py_ssize_t sep_len,
+                 Py_ssize_t maxcount)
+{
+    Py_ssize_t j, pos, count=0;
+    PyObject *list, *sub;
+
+    if (sep_len == 0) {
+        PyErr_SetString(PyExc_ValueError, "empty separator");
+        return NULL;
+    }
+    else if (sep_len == 1)
+        return stringlib_rsplit_char(str_obj, str, str_len, sep[0], maxcount);
+
+    list = PyList_New(PREALLOC_SIZE(maxcount));
+    if (list == NULL)
+        return NULL;
+
+    j = str_len;
+    while (maxcount-- > 0) {
+        pos = fastsearch(str, j, sep, sep_len, -1, FAST_RSEARCH);
+        if (pos < 0)
+            break;
+        SPLIT_ADD(str, pos + sep_len, j);
+        j = pos;
+    }
+#ifndef STRINGLIB_MUTABLE
+    if (count == 0 && STRINGLIB_CHECK_EXACT(str_obj)) {
+        /* No match in str_obj, so just use it as list[0] */
+        Py_INCREF(str_obj);
+        PyList_SET_ITEM(list, 0, (PyObject *)str_obj);
+        count++;
+    } else
+#endif
+    {
+        SPLIT_ADD(str, 0, j);
+    }
+    FIX_PREALLOC_SIZE(list);
+    if (PyList_Reverse(list) < 0)
+        goto onError;
+    return list;
+
+  onError:
+    Py_DECREF(list);
+    return NULL;
+}
+
+Py_LOCAL_INLINE(PyObject *)
+stringlib_splitlines(PyObject* str_obj,
+                     const STRINGLIB_CHAR* str, Py_ssize_t str_len,
+                     int keepends)
+{
+    /* This does not use the preallocated list because splitlines is
+       usually run with hundreds of newlines.  The overhead of
+       switching between PyList_SET_ITEM and append causes about a
+       2-3% slowdown for that common case.  A smarter implementation
+       could move the if check out, so the SET_ITEMs are done first
+       and the appends only done when the prealloc buffer is full.
+       That's too much work for little gain.*/
+
+    register Py_ssize_t i;
+    register Py_ssize_t j;
+    PyObject *list = PyList_New(0);
+    PyObject *sub;
+
+    if (list == NULL)
+        return NULL;
+
+    for (i = j = 0; i < str_len; ) {
+        Py_ssize_t eol;
+
+        /* Find a line and append it */
+        while (i < str_len && !STRINGLIB_ISLINEBREAK(str[i]))
+            i++;
+
+        /* Skip the line break reading CRLF as one line break */
+        eol = i;
+        if (i < str_len) {
+            if (str[i] == '\r' && i + 1 < str_len && str[i+1] == '\n')
+                i += 2;
+            else
+                i++;
+            if (keepends)
+                eol = i;
+        }
+#ifndef STRINGLIB_MUTABLE
+        if (j == 0 && eol == str_len && STRINGLIB_CHECK_EXACT(str_obj)) {
+            /* No linebreak in str_obj, so just use it as list[0] */
+            if (PyList_Append(list, str_obj))
+                goto onError;
+            break;
+        }
+#endif
+        SPLIT_APPEND(str, j, eol);
+        j = i;
+    }
+    return list;
+
+  onError:
+    Py_DECREF(list);
+    return NULL;
+}
+
+#endif

lib_python/2.7_Objects/stringlib/string_format.h

+// This file is originally from CPython 2.7, with modifications for Pyston
+
+/*
+    string_format.h -- implementation of string.format().
+
+    It uses the Objects/stringlib conventions, so that it can be
+    compiled for both unicode and string objects.
+*/
+
+
+/* Defines for Python 2.6 compatibility */
+#if PY_VERSION_HEX < 0x03000000
+#define PyLong_FromSsize_t _PyLong_FromSsize_t
+#endif
+
+/* Defines for more efficiently reallocating the string buffer */
+#define INITIAL_SIZE_INCREMENT 100
+#define SIZE_MULTIPLIER 2
+#define MAX_SIZE_INCREMENT  3200
+
+
+/************************************************************************/
+/***********   Global data structures and forward declarations  *********/
+/************************************************************************/
+
+/*
+   A SubString consists of the characters between two string or
+   unicode pointers.
+*/
+typedef struct {
+    STRINGLIB_CHAR *ptr;
+    STRINGLIB_CHAR *end;
+} SubString;
+
+
+typedef enum {
+    ANS_INIT,
+    ANS_AUTO,
+    ANS_MANUAL
+} AutoNumberState;   /* Keep track if we're auto-numbering fields */
+
+/* Keeps track of our auto-numbering state, and which number field we're on */
+typedef struct {
+    AutoNumberState an_state;
+    int an_field_number;
+} AutoNumber;
+
+
+/* forward declaration for recursion */
+static PyObject *
+build_string(SubString *input, PyObject *args, PyObject *kwargs,
+             int recursion_depth, AutoNumber *auto_number);
+
+
+
+/************************************************************************/
+/**************************  Utility  functions  ************************/
+/************************************************************************/
+
+static void
+AutoNumber_Init(AutoNumber *auto_number)
+{
+    auto_number->an_state = ANS_INIT;
+    auto_number->an_field_number = 0;
+}
+
+/* fill in a SubString from a pointer and length */
+Py_LOCAL_INLINE(void)
+SubString_init(SubString *str, STRINGLIB_CHAR *p, Py_ssize_t len)
+{
+    str->ptr = p;
+    if (p == NULL)
+        str->end = NULL;
+    else
+        str->end = str->ptr + len;
+}
+
+/* return a new string.  if str->ptr is NULL, return None */
+Py_LOCAL_INLINE(PyObject *)
+SubString_new_object(SubString *str)
+{
+    if (str->ptr == NULL) {
+        Py_INCREF(Py_None);
+        return Py_None;
+    }
+    return STRINGLIB_NEW(str->ptr, str->end - str->ptr);
+}
+
+/* return a new string.  if str->ptr is NULL, return None */
+Py_LOCAL_INLINE(PyObject *)
+SubString_new_object_or_empty(SubString *str)
+{
+    if (str->ptr == NULL) {
+        return STRINGLIB_NEW(NULL, 0);
+    }
+    return STRINGLIB_NEW(str->ptr, str->end - str->ptr);
+}
+
+/* Return 1 if an error has been detected switching between automatic
+   field numbering and manual field specification, else return 0. Set
+   ValueError on error. */
+static int
+autonumber_state_error(AutoNumberState state, int field_name_is_empty)
+{
+    if (state == ANS_MANUAL) {
+        if (field_name_is_empty) {
+            PyErr_SetString(PyExc_ValueError, "cannot switch from "
+                            "manual field specification to "
+                            "automatic field numbering");
+            return 1;
+        }
+    }
+    else {
+        if (!field_name_is_empty) {
+            PyErr_SetString(PyExc_ValueError, "cannot switch from "
+                            "automatic field numbering to "
+                            "manual field specification");
+            return 1;
+        }
+    }
+    return 0;
+}
+
+
+/************************************************************************/
+/***********    Output string management functions       ****************/
+/************************************************************************/
+
+typedef struct {
+    STRINGLIB_CHAR *ptr;
+    STRINGLIB_CHAR *end;
+    PyObject *obj;
+    Py_ssize_t size_increment;
+} OutputString;
+
+/* initialize an OutputString object, reserving size characters */
+static int
+output_initialize(OutputString *output, Py_ssize_t size)
+{
+    output->obj = STRINGLIB_NEW(NULL, size);
+    if (output->obj == NULL)
+        return 0;
+
+    output->ptr = STRINGLIB_STR(output->obj);
+    output->end = STRINGLIB_LEN(output->obj) + output->ptr;
+    output->size_increment = INITIAL_SIZE_INCREMENT;
+
+    return 1;
+}
+
+/*
+    output_extend reallocates the output string buffer.
+    It returns a status:  0 for a failed reallocation,
+    1 for success.
+*/
+
+static int
+output_extend(OutputString *output, Py_ssize_t count)
+{
+    STRINGLIB_CHAR *startptr = STRINGLIB_STR(output->obj);
+    Py_ssize_t curlen = output->ptr - startptr;
+    Py_ssize_t maxlen = curlen + count + output->size_increment;
+
+    if (STRINGLIB_RESIZE(&output->obj, maxlen) < 0)
+        return 0;
+    startptr = STRINGLIB_STR(output->obj);
+    output->ptr = startptr + curlen;
+    output->end = startptr + maxlen;
+    if (output->size_increment < MAX_SIZE_INCREMENT)
+        output->size_increment *= SIZE_MULTIPLIER;
+    return 1;
+}
+
+/*
+    output_data dumps characters into our output string
+    buffer.
+
+    In some cases, it has to reallocate the string.
+
+    It returns a status:  0 for a failed reallocation,
+    1 for success.
+*/
+static int
+output_data(OutputString *output, const STRINGLIB_CHAR *s, Py_ssize_t count)
+{
+    if ((count > output->end - output->ptr) && !output_extend(output, count))
+        return 0;
+    memcpy(output->ptr, s, count * sizeof(STRINGLIB_CHAR));
+    output->ptr += count;
+    return 1;
+}
+
+/************************************************************************/
+/***********  Format string parsing -- integers and identifiers *********/
+/************************************************************************/
+
+static Py_ssize_t
+get_integer(const SubString *str)
+{
+    Py_ssize_t accumulator = 0;
+    Py_ssize_t digitval;
+    STRINGLIB_CHAR *p;
+
+    /* empty string is an error */
+    if (str->ptr >= str->end)
+        return -1;
+
+    for (p = str->ptr; p < str->end; p++) {
+        digitval = STRINGLIB_TODECIMAL(*p);
+        if (digitval < 0)
+            return -1;
+        /*
+           Detect possible overflow before it happens:
+
+              accumulator * 10 + digitval > PY_SSIZE_T_MAX if and only if
+              accumulator > (PY_SSIZE_T_MAX - digitval) / 10.
+        */
+        if (accumulator > (PY_SSIZE_T_MAX - digitval) / 10) {
+            PyErr_Format(PyExc_ValueError,
+                         "Too many decimal digits in format string");
+            return -1;
+        }
+        accumulator = accumulator * 10 + digitval;
+    }
+    return accumulator;
+}
+
+/************************************************************************/
+/******** Functions to get field objects and specification strings ******/
+/************************************************************************/
+
+/* do the equivalent of obj.name */
+static PyObject *
+getattr(PyObject *obj, SubString *name)
+{
+    PyObject *newobj;
+    PyObject *str = SubString_new_object(name);
+    if (str == NULL)
+        return NULL;
+    newobj = PyObject_GetAttr(obj, str);
+    Py_DECREF(str);
+    return newobj;
+}
+
+/* do the equivalent of obj[idx], where obj is a sequence */
+static PyObject *
+getitem_sequence(PyObject *obj, Py_ssize_t idx)
+{
+    return PySequence_GetItem(obj, idx);
+}
+
+/* do the equivalent of obj[idx], where obj is not a sequence */
+static PyObject *
+getitem_idx(PyObject *obj, Py_ssize_t idx)
+{
+    PyObject *newobj;
+    PyObject *idx_obj = PyLong_FromSsize_t(idx);
+    if (idx_obj == NULL)
+        return NULL;
+    newobj = PyObject_GetItem(obj, idx_obj);
+    Py_DECREF(idx_obj);
+    return newobj;
+}
+
+/* do the equivalent of obj[name] */
+static PyObject *
+getitem_str(PyObject *obj, SubString *name)
+{
+    PyObject *newobj;
+    PyObject *str = SubString_new_object(name);
+    if (str == NULL)
+        return NULL;
+    newobj = PyObject_GetItem(obj, str);
+    Py_DECREF(str);
+    return newobj;
+}
+
+typedef struct {
+    /* the entire string we're parsing.  we assume that someone else
+       is managing its lifetime, and that it will exist for the
+       lifetime of the iterator.  can be empty */
+    SubString str;
+
+    /* pointer to where we are inside field_name */
+    STRINGLIB_CHAR *ptr;
+} FieldNameIterator;
+
+
+static int
+FieldNameIterator_init(FieldNameIterator *self, STRINGLIB_CHAR *ptr,
+                       Py_ssize_t len)
+{
+    SubString_init(&self->str, ptr, len);
+    self->ptr = self->str.ptr;
+    return 1;
+}
+
+static int
+_FieldNameIterator_attr(FieldNameIterator *self, SubString *name)
+{
+    STRINGLIB_CHAR c;
+
+    name->ptr = self->ptr;
+
+    /* return everything until '.' or '[' */
+    while (self->ptr < self->str.end) {
+        switch (c = *self->ptr++) {
+        case '[':
+        case '.':
+            /* backup so that we this character will be seen next time */
+            self->ptr--;
+            break;
+        default:
+            continue;
+        }
+        break;
+    }
+    /* end of string is okay */
+    name->end = self->ptr;
+    return 1;
+}
+
+static int
+_FieldNameIterator_item(FieldNameIterator *self, SubString *name)
+{
+    int bracket_seen = 0;
+    STRINGLIB_CHAR c;
+
+    name->ptr = self->ptr;
+
+    /* return everything until ']' */
+    while (self->ptr < self->str.end) {
+        switch (c = *self->ptr++) {
+        case ']':
+            bracket_seen = 1;
+            break;
+        default:
+            continue;
+        }
+        break;
+    }
+    /* make sure we ended with a ']' */
+    if (!bracket_seen) {
+        PyErr_SetString(PyExc_ValueError, "Missing ']' in format string");
+        return 0;
+    }
+
+    /* end of string is okay */
+    /* don't include the ']' */
+    name->end = self->ptr-1;
+    return 1;
+}
+
+/* returns 0 on error, 1 on non-error termination, and 2 if it returns a value */
+static int
+FieldNameIterator_next(FieldNameIterator *self, int *is_attribute,
+                       Py_ssize_t *name_idx, SubString *name)
+{
+    /* check at end of input */
+    if (self->ptr >= self->str.end)
+        return 1;
+
+    switch (*self->ptr++) {
+    case '.':
+        *is_attribute = 1;
+        if (_FieldNameIterator_attr(self, name) == 0)
+            return 0;
+        *name_idx = -1;
+        break;
+    case '[':
+        *is_attribute = 0;
+        if (_FieldNameIterator_item(self, name) == 0)
+            return 0;
+        *name_idx = get_integer(name);
+        if (*name_idx == -1 && PyErr_Occurred())
+            return 0;
+        break;
+    default:
+        /* Invalid character follows ']' */
+        PyErr_SetString(PyExc_ValueError, "Only '.' or '[' may "
+                        "follow ']' in format field specifier");
+        return 0;
+    }
+
+    /* empty string is an error */
+    if (name->ptr == name->end) {
+        PyErr_SetString(PyExc_ValueError, "Empty attribute in format string");
+        return 0;
+    }
+
+    return 2;
+}
+
+
+/* input: field_name
+   output: 'first' points to the part before the first '[' or '.'
+           'first_idx' is -1 if 'first' is not an integer, otherwise
+                       it's the value of first converted to an integer
+           'rest' is an iterator to return the rest
+*/
+static int
+field_name_split(STRINGLIB_CHAR *ptr, Py_ssize_t len, SubString *first,
+                 Py_ssize_t *first_idx, FieldNameIterator *rest,
+                 AutoNumber *auto_number)
+{
+    STRINGLIB_CHAR c;
+    STRINGLIB_CHAR *p = ptr;
+    STRINGLIB_CHAR *end = ptr + len;
+    int field_name_is_empty;
+    int using_numeric_index;
+
+    /* find the part up until the first '.' or '[' */
+    while (p < end) {
+        switch (c = *p++) {
+        case '[':
+        case '.':
+            /* backup so that we this character is available to the
+               "rest" iterator */
+            p--;
+            break;
+        default:
+            continue;
+        }
+        break;
+    }
+
+    /* set up the return values */
+    SubString_init(first, ptr, p - ptr);
+    FieldNameIterator_init(rest, p, end - p);
+
+    /* see if "first" is an integer, in which case it's used as an index */
+    *first_idx = get_integer(first);
+    if (*first_idx == -1 && PyErr_Occurred())
+        return 0;
+
+    field_name_is_empty = first->ptr >= first->end;
+
+    /* If the field name is omitted or if we have a numeric index
+       specified, then we're doing numeric indexing into args. */
+    using_numeric_index = field_name_is_empty || *first_idx != -1;
+
+    /* We always get here exactly one time for each field we're
+       processing. And we get here in field order (counting by left
+       braces). So this is the perfect place to handle automatic field
+       numbering if the field name is omitted. */
+
+    /* Check if we need to do the auto-numbering. It's not needed if
+       we're called from string.Format routines, because it's handled
+       in that class by itself. */
+    if (auto_number) {
+        /* Initialize our auto numbering state if this is the first
+           time we're either auto-numbering or manually numbering. */
+        if (auto_number->an_state == ANS_INIT && using_numeric_index)
+            auto_number->an_state = field_name_is_empty ?
+                ANS_AUTO : ANS_MANUAL;
+
+        /* Make sure our state is consistent with what we're doing
+           this time through. Only check if we're using a numeric
+           index. */
+        if (using_numeric_index)
+            if (autonumber_state_error(auto_number->an_state,
+                                       field_name_is_empty))
+                return 0;
+        /* Zero length field means we want to do auto-numbering of the
+           fields. */
+        if (field_name_is_empty)
+            *first_idx = (auto_number->an_field_number)++;
+    }
+
+    return 1;
+}
+
+
+/*
+    get_field_object returns the object inside {}, before the
+    format_spec.  It handles getindex and getattr lookups and consumes
+    the entire input string.
+*/
+static PyObject *
+get_field_object(SubString *input, PyObject *args, PyObject *kwargs,
+                 AutoNumber *auto_number)
+{
+    PyObject *obj = NULL;
+    int ok;
+    int is_attribute;
+    SubString name;
+    SubString first;
+    Py_ssize_t index;
+    FieldNameIterator rest;
+
+    if (!field_name_split(input->ptr, input->end - input->ptr, &first,
+                          &index, &rest, auto_number)) {
+        goto error;
+    }
+
+    if (index == -1) {
+        /* look up in kwargs */
+        PyObject *key = SubString_new_object(&first);
+        if (key == NULL)
+            goto error;
+        if ((kwargs == NULL) || (obj = PyDict_GetItem(kwargs, key)) == NULL) {
+            PyErr_SetObject(PyExc_KeyError, key);
+            Py_DECREF(key);
+            goto error;
+        }
+        Py_DECREF(key);
+        Py_INCREF(obj);
+    }
+    else {
+        /* look up in args */
+        obj = PySequence_GetItem(args, index);
+        if (obj == NULL)
+            goto error;
+    }
+
+    /* iterate over the rest of the field_name */
+    while ((ok = FieldNameIterator_next(&rest, &is_attribute, &index,
+                                        &name)) == 2) {
+        PyObject *tmp;
+
+        if (is_attribute)
+            /* getattr lookup "." */
+            tmp = getattr(obj, &name);
+        else
+            /* getitem lookup "[]" */
+            if (index == -1)
+                tmp = getitem_str(obj, &name);
+            else
+                if (PySequence_Check(obj))
+                    tmp = getitem_sequence(obj, index);
+                else
+                    /* not a sequence */
+                    tmp = getitem_idx(obj, index);
+        if (tmp == NULL)
+            goto error;
+
+        /* assign to obj */
+        Py_DECREF(obj);
+        obj = tmp;
+    }
+    /* end of iterator, this is the non-error case */
+    if (ok == 1)
+        return obj;
+error:
+    Py_XDECREF(obj);
+    return NULL;
+}
+
+/************************************************************************/
+/*****************  Field rendering functions  **************************/
+/************************************************************************/
+
+/*
+    render_field() is the main function in this section.  It takes the
+    field object and field specification string generated by
+    get_field_and_spec, and renders the field into the output string.
+
+    render_field calls fieldobj.__format__(format_spec) method, and
+    appends to the output.
+*/
+static int
+render_field(PyObject *fieldobj, SubString *format_spec, OutputString *output)
+{
+    int ok = 0;
+    PyObject *result = NULL;
+    PyObject *format_spec_object = NULL;
+    PyObject *(*formatter)(PyObject *, STRINGLIB_CHAR *, Py_ssize_t) = NULL;
+    STRINGLIB_CHAR* format_spec_start = format_spec->ptr ?
+            format_spec->ptr : NULL;
+    Py_ssize_t format_spec_len = format_spec->ptr ?
+            format_spec->end - format_spec->ptr : 0;
+
+    /* If we know the type exactly, skip the lookup of __format__ and just
+       call the formatter directly. */
+#if STRINGLIB_IS_UNICODE
+    if (PyUnicode_CheckExact(fieldobj))
+        formatter = _PyUnicode_FormatAdvanced;
+    /* Unfortunately, there's a problem with checking for int, long,
+       and float here.  If we're being included as unicode, their
+       formatters expect string format_spec args.  For now, just skip
+       this optimization for unicode.  This could be fixed, but it's a
+       hassle. */
+#else
+    if (PyString_CheckExact(fieldobj))
+        formatter = _PyBytes_FormatAdvanced;
+    else if (PyInt_CheckExact(fieldobj))
+        formatter =_PyInt_FormatAdvanced;
+    else if (PyLong_CheckExact(fieldobj))
+        formatter =_PyLong_FormatAdvanced;
+    else if (PyFloat_CheckExact(fieldobj))
+        formatter = _PyFloat_FormatAdvanced;
+#endif
+
+    if (formatter) {
+        /* we know exactly which formatter will be called when __format__ is
+           looked up, so call it directly, instead. */
+        result = formatter(fieldobj, format_spec_start, format_spec_len);
+    }
+    else {
+        /* We need to create an object out of the pointers we have, because
+           __format__ takes a string/unicode object for format_spec. */
+        format_spec_object = STRINGLIB_NEW(format_spec_start,
+                                           format_spec_len);
+        if (format_spec_object == NULL)
+            goto done;
+
+        result = PyObject_Format(fieldobj, format_spec_object);
+    }
+    if (result == NULL)
+        goto done;
+
+#if PY_VERSION_HEX >= 0x03000000
+    assert(PyUnicode_Check(result));
+#else
+    assert(PyString_Check(result) || PyUnicode_Check(result));
+
+    /* Convert result to our type.  We could be str, and result could
+       be unicode */
+    {
+        PyObject *tmp = STRINGLIB_TOSTR(result);
+        if (tmp == NULL)
+            goto done;
+        Py_DECREF(result);
+        result = tmp;
+    }
+#endif
+
+    ok = output_data(output,
+                     STRINGLIB_STR(result), STRINGLIB_LEN(result));
+done:
+    Py_XDECREF(format_spec_object);
+    Py_XDECREF(result);
+    return ok;
+}
+
+static int
+parse_field(SubString *str, SubString *field_name, SubString *format_spec,
+            STRINGLIB_CHAR *conversion)
+{
+    /* Note this function works if the field name is zero length,
+       which is good.  Zero length field names are handled later, in
+       field_name_split. */
+
+    STRINGLIB_CHAR c = 0;
+
+    /* initialize these, as they may be empty */
+    *conversion = '\0';
+    SubString_init(format_spec, NULL, 0);
+
+    /* Search for the field name.  it's terminated by the end of
+       the string, or a ':' or '!' */
+    field_name->ptr = str->ptr;
+    while (str->ptr < str->end) {
+        switch (c = *(str->ptr++)) {
+        case ':':
+        case '!':
+            break;
+        default:
+            continue;
+        }
+        break;
+    }
+
+    if (c == '!' || c == ':') {
+        /* we have a format specifier and/or a conversion */
+        /* don't include the last character */
+        field_name->end = str->ptr-1;
+
+        /* the format specifier is the rest of the string */
+        format_spec->ptr = str->ptr;
+        format_spec->end = str->end;
+
+        /* see if there's a conversion specifier */
+        if (c == '!') {
+            /* there must be another character present */
+            if (format_spec->ptr >= format_spec->end) {
+                PyErr_SetString(PyExc_ValueError,
+                                "end of format while looking for conversion "
+                                "specifier");
+                return 0;
+            }
+            *conversion = *(format_spec->ptr++);
+
+            /* if there is another character, it must be a colon */
+            if (format_spec->ptr < format_spec->end) {
+                c = *(format_spec->ptr++);
+                if (c != ':') {
+                    PyErr_SetString(PyExc_ValueError,
+                                    "expected ':' after format specifier");
+                    return 0;
+                }
+            }
+        }
+    }
+    else
+        /* end of string, there's no format_spec or conversion */
+        field_name->end = str->ptr;
+
+    return 1;
+}
+
+/************************************************************************/
+/******* Output string allocation and escape-to-markup processing  ******/
+/************************************************************************/
+
+/* MarkupIterator breaks the string into pieces of either literal
+   text, or things inside {} that need to be marked up.  it is
+   designed to make it easy to wrap a Python iterator around it, for
+   use with the Formatter class */
+
+typedef struct {
+    SubString str;
+} MarkupIterator;
+
+static int
+MarkupIterator_init(MarkupIterator *self, STRINGLIB_CHAR *ptr, Py_ssize_t len)
+{
+    SubString_init(&self->str, ptr, len);
+    return 1;
+}
+
+/* returns 0 on error, 1 on non-error termination, and 2 if it got a
+   string (or something to be expanded) */
+static int
+MarkupIterator_next(MarkupIterator *self, SubString *literal,
+                    int *field_present, SubString *field_name,
+                    SubString *format_spec, STRINGLIB_CHAR *conversion,
+                    int *format_spec_needs_expanding)
+{
+    int at_end;
+    STRINGLIB_CHAR c = 0;
+    STRINGLIB_CHAR *start;
+    int count;
+    Py_ssize_t len;
+    int markup_follows = 0;
+
+    /* initialize all of the output variables */
+    SubString_init(literal, NULL, 0);
+    SubString_init(field_name, NULL, 0);
+    SubString_init(format_spec, NULL, 0);
+    *conversion = '\0';
+    *format_spec_needs_expanding = 0;
+    *field_present = 0;
+
+    /* No more input, end of iterator.  This is the normal exit
+       path. */
+    if (self->str.ptr >= self->str.end)
+        return 1;
+
+    start = self->str.ptr;
+
+    /* First read any literal text. Read until the end of string, an
+       escaped '{' or '}', or an unescaped '{'.  In order to never
+       allocate memory and so I can just pass pointers around, if
+       there's an escaped '{' or '}' then we'll return the literal
+       including the brace, but no format object.  The next time
+       through, we'll return the rest of the literal, skipping past
+       the second consecutive brace. */
+    while (self->str.ptr < self->str.end) {
+        switch (c = *(self->str.ptr++)) {
+        case '{':
+        case '}':
+            markup_follows = 1;
+            break;
+        default:
+            continue;
+        }
+        break;
+    }
+
+    at_end = self->str.ptr >= self->str.end;
+    len = self->str.ptr - start;
+
+    if ((c == '}') && (at_end || (c != *self->str.ptr))) {
+        PyErr_SetString(PyExc_ValueError, "Single '}' encountered "
+                        "in format string");
+        return 0;
+    }
+    if (at_end && c == '{') {
+        PyErr_SetString(PyExc_ValueError, "Single '{' encountered "
+                        "in format string");
+        return 0;
+    }
+    if (!at_end) {
+        if (c == *self->str.ptr) {
+            /* escaped } or {, skip it in the input.  there is no
+               markup object following us, just this literal text */
+            self->str.ptr++;
+            markup_follows = 0;
+        }
+        else
+            len--;
+    }
+
+    /* record the literal text */
+    literal->ptr = start;
+    literal->end = start + len;
+
+    if (!markup_follows)
+        return 2;
+
+    /* this is markup, find the end of the string by counting nested
+       braces.  note that this prohibits escaped braces, so that
+       format_specs cannot have braces in them. */
+    *field_present = 1;
+    count = 1;
+
+    start = self->str.ptr;
+
+    /* we know we can't have a zero length string, so don't worry
+       about that case */
+    while (self->str.ptr < self->str.end) {
+        switch (c = *(self->str.ptr++)) {
+        case '{':
+            /* the format spec needs to be recursively expanded.
+               this is an optimization, and not strictly needed */
+            *format_spec_needs_expanding = 1;
+            count++;
+            break;
+        case '}':
+            count--;
+            if (count <= 0) {
+                /* we're done.  parse and get out */
+                SubString s;
+
+                SubString_init(&s, start, self->str.ptr - 1 - start);
+                if (parse_field(&s, field_name, format_spec, conversion) == 0)
+                    return 0;
+
+                /* success */
+                return 2;
+            }
+            break;
+        }
+    }
+
+    /* end of string while searching for matching '}' */
+    PyErr_SetString(PyExc_ValueError, "unmatched '{' in format");
+    return 0;
+}
+
+
+/* do the !r or !s conversion on obj */
+static PyObject *
+do_conversion(PyObject *obj, STRINGLIB_CHAR conversion)
+{
+    /* XXX in pre-3.0, do we need to convert this to unicode, since it
+       might have returned a string? */
+    switch (conversion) {
+    case 'r':
+        return PyObject_Repr(obj);
+    case 's':
+        return STRINGLIB_TOSTR(obj);
+    default:
+        if (conversion > 32 && conversion < 127) {
+                /* It's the ASCII subrange; casting to char is safe
+                   (assuming the execution character set is an ASCII
+                   superset). */
+                PyErr_Format(PyExc_ValueError,
+                     "Unknown conversion specifier %c",
+                     (char)conversion);
+        } else
+                PyErr_Format(PyExc_ValueError,
+                     "Unknown conversion specifier \\x%x",
+                     (unsigned int)conversion);
+        return NULL;
+    }
+}
+
+/* given:
+
+   {field_name!conversion:format_spec}
+
+   compute the result and write it to output.
+   format_spec_needs_expanding is an optimization.  if it's false,
+   just output the string directly, otherwise recursively expand the
+   format_spec string.
+
+   field_name is allowed to be zero length, in which case we
+   are doing auto field numbering.
+*/
+
+static int
+output_markup(SubString *field_name, SubString *format_spec,
+              int format_spec_needs_expanding, STRINGLIB_CHAR conversion,
+              OutputString *output, PyObject *args, PyObject *kwargs,
+              int recursion_depth, AutoNumber *auto_number)
+{
+    PyObject *tmp = NULL;
+    PyObject *fieldobj = NULL;
+    SubString expanded_format_spec;
+    SubString *actual_format_spec;
+    int result = 0;
+
+    /* convert field_name to an object */
+    fieldobj = get_field_object(field_name, args, kwargs, auto_number);
+    if (fieldobj == NULL)
+        goto done;
+
+    if (conversion != '\0') {
+        tmp = do_conversion(fieldobj, conversion);
+        if (tmp == NULL)
+            goto done;
+
+        /* do the assignment, transferring ownership: fieldobj = tmp */
+        Py_DECREF(fieldobj);
+        fieldobj = tmp;
+        tmp = NULL;
+    }
+
+    /* if needed, recurively compute the format_spec */
+    if (format_spec_needs_expanding) {
+        tmp = build_string(format_spec, args, kwargs, recursion_depth-1,
+                           auto_number);
+        if (tmp == NULL)
+            goto done;
+
+        /* note that in the case we're expanding the format string,
+           tmp must be kept around until after the call to
+           render_field. */
+        SubString_init(&expanded_format_spec,
+                       STRINGLIB_STR(tmp), STRINGLIB_LEN(tmp));
+        actual_format_spec = &expanded_format_spec;
+    }
+    else
+        actual_format_spec = format_spec;
+
+    if (render_field(fieldobj, actual_format_spec, output) == 0)
+        goto done;
+
+    result = 1;
+
+done:
+    Py_XDECREF(fieldobj);
+    Py_XDECREF(tmp);
+
+    return result;
+}
+
+/*
+    do_markup is the top-level loop for the format() method.  It
+    searches through the format string for escapes to markup codes, and
+    calls other functions to move non-markup text to the output,
+    and to perform the markup to the output.
+*/
+static int
+do_markup(SubString *input, PyObject *args, PyObject *kwargs,
+          OutputString *output, int recursion_depth, AutoNumber *auto_number)
+{
+    MarkupIterator iter;
+    int format_spec_needs_expanding;
+    int result;
+    int field_present;
+    SubString literal;
+    SubString field_name;
+    SubString format_spec;
+    STRINGLIB_CHAR conversion;
+
+    MarkupIterator_init(&iter, input->ptr, input->end - input->ptr);
+    while ((result = MarkupIterator_next(&iter, &literal, &field_present,
+                                         &field_name, &format_spec,
+                                         &conversion,
+                                         &format_spec_needs_expanding)) == 2) {
+        if (!output_data(output, literal.ptr, literal.end - literal.ptr))
+            return 0;
+        if (field_present)
+            if (!output_markup(&field_name, &format_spec,
+                               format_spec_needs_expanding, conversion, output,
+                               args, kwargs, recursion_depth, auto_number))
+                return 0;
+    }
+    return result;
+}
+
+
+/*
+    build_string allocates the output string and then
+    calls do_markup to do the heavy lifting.
+*/
+static PyObject *
+build_string(SubString *input, PyObject *args, PyObject *kwargs,
+             int recursion_depth, AutoNumber *auto_number)
+{
+    OutputString output;
+    PyObject *result = NULL;
+    Py_ssize_t count;
+
+    output.obj = NULL; /* needed so cleanup code always works */
+
+    /* check the recursion level */
+    if (recursion_depth <= 0) {
+        PyErr_SetString(PyExc_ValueError,
+                        "Max string recursion exceeded");
+        goto done;
+    }
+
+    /* initial size is the length of the format string, plus the size
+       increment.  seems like a reasonable default */
+    if (!output_initialize(&output,
+                           input->end - input->ptr +
+                           INITIAL_SIZE_INCREMENT))
+        goto done;
+
+    if (!do_markup(input, args, kwargs, &output, recursion_depth,
+                   auto_number)) {
+        goto done;
+    }
+
+    count = output.ptr - STRINGLIB_STR(output.obj);
+    if (STRINGLIB_RESIZE(&output.obj, count) < 0) {
+        goto done;
+    }
+
+    /* transfer ownership to result */
+    result = output.obj;
+    output.obj = NULL;
+
+done:
+    Py_XDECREF(output.obj);
+    return result;
+}
+
+/************************************************************************/
+/*********** main routine ***********************************************/
+/************************************************************************/
+
+/* this is the main entry point */
+// Pyston change: changed to non-static
+/* static */ PyObject *
+do_string_format(PyObject *self, PyObject *args, PyObject *kwargs)
+{
+    SubString input;
+
+    /* PEP 3101 says only 2 levels, so that
+       "{0:{1}}".format('abc', 's')            # works
+       "{0:{1:{2}}}".format('abc', 's', '')    # fails
+    */
+    int recursion_depth = 2;
+
+    AutoNumber auto_number;
+
+    AutoNumber_Init(&auto_number);
+    SubString_init(&input, STRINGLIB_STR(self), STRINGLIB_LEN(self));
+    return build_string(&input, args, kwargs, recursion_depth, &auto_number);
+}
+
+
+
+/************************************************************************/
+/*********** formatteriterator ******************************************/
+/************************************************************************/
+
+/* This is used to implement string.Formatter.vparse().  It exists so
+   Formatter can share code with the built in unicode.format() method.
+   It's really just a wrapper around MarkupIterator that is callable
+   from Python. */
+
+typedef struct {
+    PyObject_HEAD
+
+    STRINGLIB_OBJECT *str;
+
+    MarkupIterator it_markup;
+} formatteriterobject;
+
+static void
+formatteriter_dealloc(formatteriterobject *it)
+{
+    Py_XDECREF(it->str);
+    PyObject_FREE(it);
+}
+
+/* returns a tuple:
+   (literal, field_name, format_spec, conversion)
+
+   literal is any literal text to output.  might be zero length
+   field_name is the string before the ':'.  might be None
+   format_spec is the string after the ':'.  mibht be None
+   conversion is either None, or the string after the '!'
+*/
+static PyObject *
+formatteriter_next(formatteriterobject *it)
+{
+    SubString literal;
+    SubString field_name;
+    SubString format_spec;
+    STRINGLIB_CHAR conversion;
+    int format_spec_needs_expanding;
+    int field_present;
+    int result = MarkupIterator_next(&it->it_markup, &literal, &field_present,
+                                     &field_name, &format_spec, &conversion,
+                                     &format_spec_needs_expanding);
+
+    /* all of the SubString objects point into it->str, so no
+       memory management needs to be done on them */
+    assert(0 <= result && result <= 2);
+    if (result == 0 || result == 1)
+        /* if 0, error has already been set, if 1, iterator is empty */
+        return NULL;
+    else {
+        PyObject *literal_str = NULL;
+        PyObject *field_name_str = NULL;
+        PyObject *format_spec_str = NULL;
+        PyObject *conversion_str = NULL;
+        PyObject *tuple = NULL;
+
+        literal_str = SubString_new_object(&literal);
+        if (literal_str == NULL)
+            goto done;
+
+        field_name_str = SubString_new_object(&field_name);
+        if (field_name_str == NULL)
+            goto done;
+
+        /* if field_name is non-zero length, return a string for
+           format_spec (even if zero length), else return None */
+        format_spec_str = (field_present ?
+                           SubString_new_object_or_empty :
+                           SubString_new_object)(&format_spec);
+        if (format_spec_str == NULL)
+            goto done;
+
+        /* if the conversion is not specified, return a None,
+           otherwise create a one length string with the conversion
+           character */
+        if (conversion == '\0') {
+            conversion_str = Py_None;
+            Py_INCREF(conversion_str);
+        }
+        else
+            conversion_str = STRINGLIB_NEW(&conversion, 1);
+        if (conversion_str == NULL)
+            goto done;
+
+        tuple = PyTuple_Pack(4, literal_str, field_name_str, format_spec_str,
+                             conversion_str);
+    done:
+        Py_XDECREF(literal_str);
+        Py_XDECREF(field_name_str);
+        Py_XDECREF(format_spec_str);
+        Py_XDECREF(conversion_str);
+        return tuple;
+    }
+}
+
+static PyMethodDef formatteriter_methods[] = {
+    {NULL,              NULL}           /* sentinel */
+};
+
+static PyTypeObject PyFormatterIter_Type = {
+    // Pyston change:
+    PyVarObject_HEAD_INIT(NULL /* was &PyType_Type */, 0)
+    "formatteriterator",                /* tp_name */
+    sizeof(formatteriterobject),        /* tp_basicsize */
+    0,                                  /* tp_itemsize */
+    /* methods */
+    (destructor)formatteriter_dealloc,  /* tp_dealloc */
+    0,                                  /* tp_print */
+    0,                                  /* tp_getattr */
+    0,                                  /* tp_setattr */
+    0,                                  /* tp_compare */
+    0,                                  /* tp_repr */
+    0,                                  /* tp_as_number */
+    0,                                  /* tp_as_sequence */
+    0,                                  /* tp_as_mapping */
+    0,                                  /* tp_hash */
+    0,                                  /* tp_call */
+    0,                                  /* tp_str */
+    PyObject_GenericGetAttr,            /* tp_getattro */
+    0,                                  /* tp_setattro */
+    0,                                  /* tp_as_buffer */
+    Py_TPFLAGS_DEFAULT,                 /* tp_flags */
+    0,                                  /* tp_doc */
+    0,                                  /* tp_traverse */
+    0,                                  /* tp_clear */
+    0,                                  /* tp_richcompare */
+    0,                                  /* tp_weaklistoffset */
+    PyObject_SelfIter,                  /* tp_iter */
+    (iternextfunc)formatteriter_next,   /* tp_iternext */
+    formatteriter_methods,              /* tp_methods */
+    0,
+};
+
+/* unicode_formatter_parser is used to implement
+   string.Formatter.vformat.  it parses a string and returns tuples
+   describing the parsed elements.  It's a wrapper around
+   stringlib/string_format.h's MarkupIterator */
+static PyObject *
+formatter_parser(STRINGLIB_OBJECT *self)
+{
+    formatteriterobject *it;
+
+    it = PyObject_New(formatteriterobject, &PyFormatterIter_Type);
+    if (it == NULL)
+        return NULL;
+
+    /* take ownership, give the object to the iterator */
+    Py_INCREF(self);
+    it->str = self;
+
+    /* initialize the contained MarkupIterator */
+    MarkupIterator_init(&it->it_markup,
+                        STRINGLIB_STR(self),
+                        STRINGLIB_LEN(self));
+
+    return (PyObject *)it;
+}
+
+
+/************************************************************************/
+/*********** fieldnameiterator ******************************************/
+/************************************************************************/
+
+
+/* This is used to implement string.Formatter.vparse().  It parses the
+   field name into attribute and item values.  It's a Python-callable
+   wrapper around FieldNameIterator */
+
+typedef struct {
+    PyObject_HEAD
+
+    STRINGLIB_OBJECT *str;
+
+    FieldNameIterator it_field;
+} fieldnameiterobject;
+
+static void
+fieldnameiter_dealloc(fieldnameiterobject *it)
+{
+    Py_XDECREF(it->str);
+    PyObject_FREE(it);
+}
+
+/* returns a tuple:
+   (is_attr, value)
+   is_attr is true if we used attribute syntax (e.g., '.foo')
+              false if we used index syntax (e.g., '[foo]')
+   value is an integer or string
+*/
+static PyObject *
+fieldnameiter_next(fieldnameiterobject *it)
+{
+    int result;
+    int is_attr;
+    Py_ssize_t idx;
+    SubString name;
+
+    result = FieldNameIterator_next(&it->it_field, &is_attr,
+                                    &idx, &name);
+    if (result == 0 || result == 1)
+        /* if 0, error has already been set, if 1, iterator is empty */
+        return NULL;
+    else {
+        PyObject* result = NULL;
+        PyObject* is_attr_obj = NULL;
+        PyObject* obj = NULL;
+
+        is_attr_obj = PyBool_FromLong(is_attr);
+        if (is_attr_obj == NULL)
+            goto done;
+
+        /* either an integer or a string */
+        if (idx != -1)
+            obj = PyLong_FromSsize_t(idx);
+        else
+            obj = SubString_new_object(&name);
+        if (obj == NULL)
+            goto done;
+
+        /* return a tuple of values */
+        result = PyTuple_Pack(2, is_attr_obj, obj);
+
+    done:
+        Py_XDECREF(is_attr_obj);
+        Py_XDECREF(obj);
+        return result;
+    }
+}
+
+static PyMethodDef fieldnameiter_methods[] = {
+    {NULL,              NULL}           /* sentinel */
+};
+
+static PyTypeObject PyFieldNameIter_Type = {
+    // Pyston change:
+    PyVarObject_HEAD_INIT(NULL /* was &PyType_Type */, 0)
+    "fieldnameiterator",                /* tp_name */
+    sizeof(fieldnameiterobject),        /* tp_basicsize */
+    0,                                  /* tp_itemsize */
+    /* methods */
+    (destructor)fieldnameiter_dealloc,  /* tp_dealloc */
+    0,                                  /* tp_print */
+    0,                                  /* tp_getattr */
+    0,                                  /* tp_setattr */
+    0,                                  /* tp_compare */
+    0,                                  /* tp_repr */
+    0,                                  /* tp_as_number */
+    0,                                  /* tp_as_sequence */
+    0,                                  /* tp_as_mapping */
+    0,                                  /* tp_hash */
+    0,                                  /* tp_call */
+    0,                                  /* tp_str */
+    PyObject_GenericGetAttr,            /* tp_getattro */
+    0,                                  /* tp_setattro */
+    0,                                  /* tp_as_buffer */
+    Py_TPFLAGS_DEFAULT,                 /* tp_flags */
+    0,                                  /* tp_doc */
+    0,                                  /* tp_traverse */
+    0,                                  /* tp_clear */
+    0,                                  /* tp_richcompare */
+    0,                                  /* tp_weaklistoffset */
+    PyObject_SelfIter,                  /* tp_iter */
+    (iternextfunc)fieldnameiter_next,   /* tp_iternext */
+    fieldnameiter_methods,              /* tp_methods */
+    0};
+
+/* unicode_formatter_field_name_split is used to implement
+   string.Formatter.vformat.  it takes an PEP 3101 "field name", and
+   returns a tuple of (first, rest): "first", the part before the
+   first '.' or '['; and "rest", an iterator for the rest of the field
+   name.  it's a wrapper around stringlib/string_format.h's
+   field_name_split.  The iterator it returns is a
+   FieldNameIterator */
+static PyObject *
+formatter_field_name_split(STRINGLIB_OBJECT *self)
+{
+    SubString first;
+    Py_ssize_t first_idx;
+    fieldnameiterobject *it;
+
+    PyObject *first_obj = NULL;
+    PyObject *result = NULL;
+
+    it = PyObject_New(fieldnameiterobject, &PyFieldNameIter_Type);
+    if (it == NULL)
+        return NULL;
+
+    /* take ownership, give the object to the iterator.  this is
+       just to keep the field_name alive */
+    Py_INCREF(self);
+    it->str = self;
+
+    /* Pass in auto_number = NULL. We'll return an empty string for
+       first_obj in that case. */
+    if (!field_name_split(STRINGLIB_STR(self),
+                          STRINGLIB_LEN(self),
+                          &first, &first_idx, &it->it_field, NULL))
+        goto done;
+
+    /* first becomes an integer, if possible; else a string */
+    if (first_idx != -1)
+        first_obj = PyLong_FromSsize_t(first_idx);
+    else
+        /* convert "first" into a string object */
+        first_obj = SubString_new_object(&first);
+    if (first_obj == NULL)
+        goto done;
+
+    /* return a tuple of values */
+    result = PyTuple_Pack(2, first_obj, it);
+
+done:
+    Py_XDECREF(it);
+    Py_XDECREF(first_obj);
+    return result;
+}

lib_python/2.7_Objects/stringlib/stringdefs.h

+#ifndef STRINGLIB_STRINGDEFS_H
+#define STRINGLIB_STRINGDEFS_H
+
+/* this is sort of a hack.  there's at least one place (formatting
+   floats) where some stringlib code takes a different path if it's
+   compiled as unicode. */
+#define STRINGLIB_IS_UNICODE     0
+
+#define STRINGLIB_OBJECT         PyStringObject
+#define STRINGLIB_CHAR           char
+#define STRINGLIB_TYPE_NAME      "string"
+#define STRINGLIB_PARSE_CODE     "S"
+#define STRINGLIB_EMPTY          nullstring
+#define STRINGLIB_ISSPACE        Py_ISSPACE
+#define STRINGLIB_ISLINEBREAK(x) ((x == '\n') || (x == '\r'))
+#define STRINGLIB_ISDECIMAL(x)   ((x >= '0') && (x <= '9'))
+#define STRINGLIB_TODECIMAL(x)   (STRINGLIB_ISDECIMAL(x) ? (x - '0') : -1)
+#define STRINGLIB_TOUPPER        Py_TOUPPER
+#define STRINGLIB_TOLOWER        Py_TOLOWER
+#define STRINGLIB_FILL           memset
+#define STRINGLIB_STR            PyString_AS_STRING
+#define STRINGLIB_LEN            PyString_GET_SIZE
+#define STRINGLIB_NEW            PyString_FromStringAndSize
+#define STRINGLIB_RESIZE         _PyString_Resize
+#define STRINGLIB_CHECK          PyString_Check
+#define STRINGLIB_CHECK_EXACT    PyString_CheckExact
+#define STRINGLIB_TOSTR          PyObject_Str
+#define STRINGLIB_GROUPING       _PyString_InsertThousandsGrouping
+#define STRINGLIB_GROUPING_LOCALE _PyString_InsertThousandsGroupingLocale
+
+#define STRINGLIB_WANT_CONTAINS_OBJ 1
+
+#endif /* !STRINGLIB_STRINGDEFS_H */

lib_python/2.7_Objects/stringlib/transmogrify.h

+/* NOTE: this API is -ONLY- for use with single byte character strings. */
+/* Do not use it with Unicode. */
+
+/* the more complicated methods.  parts of these should be pulled out into the
+   shared code in bytes_methods.c to cut down on duplicate code bloat.  */
+
+PyDoc_STRVAR(expandtabs__doc__,
+"B.expandtabs([tabsize]) -> copy of B\n\
+\n\
+Return a copy of B where all tab characters are expanded using spaces.\n\
+If tabsize is not given, a tab size of 8 characters is assumed.");
+
+static PyObject*
+stringlib_expandtabs(PyObject *self, PyObject *args)
+{
+    const char *e, *p;
+    char *q;
+    Py_ssize_t i, j;
+    PyObject *u;
+    int tabsize = 8;
+    
+    if (!PyArg_ParseTuple(args, "|i:expandtabs", &tabsize))
+        return NULL;
+    
+    /* First pass: determine size of output string */
+    i = j = 0;
+    e = STRINGLIB_STR(self) + STRINGLIB_LEN(self);
+    for (p = STRINGLIB_STR(self); p < e; p++) {
+        if (*p == '\t') {
+            if (tabsize > 0) {
+                Py_ssize_t incr = tabsize - (j % tabsize);
+                if (j > PY_SSIZE_T_MAX - incr)
+                    goto overflow;
+                j += incr;
+            }
+        }
+        else {
+            if (j > PY_SSIZE_T_MAX - 1)
+                goto overflow;
+            j++;
+            if (*p == '\n' || *p == '\r') {
+                if (i > PY_SSIZE_T_MAX - j)
+                    goto overflow;
+                i += j;
+                j = 0;
+            }
+        }
+    }
+    
+    if (i > PY_SSIZE_T_MAX - j)
+        goto overflow;
+    
+    /* Second pass: create output string and fill it */
+    u = STRINGLIB_NEW(NULL, i + j);
+    if (!u)
+        return NULL;
+    
+    j = 0;
+    q = STRINGLIB_STR(u);
+    
+    for (p = STRINGLIB_STR(self); p < e; p++) {
+        if (*p == '\t') {
+            if (tabsize > 0) {
+                i = tabsize - (j % tabsize);
+                j += i;
+                while (i--)
+                    *q++ = ' ';
+            }
+        }
+        else {
+            j++;
+            *q++ = *p;
+            if (*p == '\n' || *p == '\r')
+                j = 0;
+        }
+    }
+
+    return u;
+  overflow:
+    PyErr_SetString(PyExc_OverflowError, "result too long");
+    return NULL;
+}
+
+Py_LOCAL_INLINE(PyObject *)
+pad(PyObject *self, Py_ssize_t left, Py_ssize_t right, char fill)
+{
+    PyObject *u;
+
+    if (left < 0)
+        left = 0;
+    if (right < 0)
+        right = 0;
+
+    if (left == 0 && right == 0 && STRINGLIB_CHECK_EXACT(self)) {
+#if STRINGLIB_MUTABLE
+        /* We're defined as returning a copy;  If the object is mutable
+         * that means we must make an identical copy. */
+        return STRINGLIB_NEW(STRINGLIB_STR(self), STRINGLIB_LEN(self));
+#else
+        Py_INCREF(self);
+        return (PyObject *)self;
+#endif /* STRINGLIB_MUTABLE */
+    }
+
+    u = STRINGLIB_NEW(NULL,
+				   left + STRINGLIB_LEN(self) + right);
+    if (u) {
+        if (left)
+            memset(STRINGLIB_STR(u), fill, left);
+        Py_MEMCPY(STRINGLIB_STR(u) + left,
+	       STRINGLIB_STR(self),
+	       STRINGLIB_LEN(self));
+        if (right)
+            memset(STRINGLIB_STR(u) + left + STRINGLIB_LEN(self),
+		   fill, right);
+    }
+
+    return u;
+}
+
+PyDoc_STRVAR(ljust__doc__,
+"B.ljust(width[, fillchar]) -> copy of B\n"
+"\n"
+"Return B left justified in a string of length width. Padding is\n"
+"done using the specified fill character (default is a space).");
+
+static PyObject *
+stringlib_ljust(PyObject *self, PyObject *args)
+{
+    Py_ssize_t width;
+    char fillchar = ' ';
+
+    if (!PyArg_ParseTuple(args, "n|c:ljust", &width, &fillchar))
+        return NULL;
+
+    if (STRINGLIB_LEN(self) >= width && STRINGLIB_CHECK_EXACT(self)) {
+#if STRINGLIB_MUTABLE
+        /* We're defined as returning a copy;  If the object is mutable
+         * that means we must make an identical copy. */
+        return STRINGLIB_NEW(STRINGLIB_STR(self), STRINGLIB_LEN(self));
+#else
+        Py_INCREF(self);
+        return (PyObject*) self;
+#endif
+    }
+
+    return pad(self, 0, width - STRINGLIB_LEN(self), fillchar);
+}
+
+
+PyDoc_STRVAR(rjust__doc__,
+"B.rjust(width[, fillchar]) -> copy of B\n"
+"\n"
+"Return B right justified in a string of length width. Padding is\n"
+"done using the specified fill character (default is a space)");
+
+static PyObject *
+stringlib_rjust(PyObject *self, PyObject *args)
+{
+    Py_ssize_t width;
+    char fillchar = ' ';
+
+    if (!PyArg_ParseTuple(args, "n|c:rjust", &width, &fillchar))
+        return NULL;
+
+    if (STRINGLIB_LEN(self) >= width && STRINGLIB_CHECK_EXACT(self)) {
+#if STRINGLIB_MUTABLE
+        /* We're defined as returning a copy;  If the object is mutable
+         * that means we must make an identical copy. */
+        return STRINGLIB_NEW(STRINGLIB_STR(self), STRINGLIB_LEN(self));
+#else
+        Py_INCREF(self);
+        return (PyObject*) self;
+#endif
+    }
+
+    return pad(self, width - STRINGLIB_LEN(self), 0, fillchar);
+}
+
+
+PyDoc_STRVAR(center__doc__,
+"B.center(width[, fillchar]) -> copy of B\n"
+"\n"
+"Return B centered in a string of length width.  Padding is\n"
+"done using the specified fill character (default is a space).");
+
+static PyObject *
+stringlib_center(PyObject *self, PyObject *args)
+{
+    Py_ssize_t marg, left;
+    Py_ssize_t width;
+    char fillchar = ' ';
+
+    if (!PyArg_ParseTuple(args, "n|c:center", &width, &fillchar))
+        return NULL;
+
+    if (STRINGLIB_LEN(self) >= width && STRINGLIB_CHECK_EXACT(self)) {
+#if STRINGLIB_MUTABLE
+        /* We're defined as returning a copy;  If the object is mutable
+         * that means we must make an identical copy. */
+        return STRINGLIB_NEW(STRINGLIB_STR(self), STRINGLIB_LEN(self));
+#else
+        Py_INCREF(self);
+        return (PyObject*) self;
+#endif
+    }
+
+    marg = width - STRINGLIB_LEN(self);
+    left = marg / 2 + (marg & width & 1);
+
+    return pad(self, left, marg - left, fillchar);
+}
+
+PyDoc_STRVAR(zfill__doc__,
+"B.zfill(width) -> copy of B\n"
+"\n"
+"Pad a numeric string B with zeros on the left, to fill a field\n"
+"of the specified width.  B is never truncated.");
+
+static PyObject *
+stringlib_zfill(PyObject *self, PyObject *args)
+{
+    Py_ssize_t fill;
+    PyObject *s;
+    char *p;
+    Py_ssize_t width;
+
+    if (!PyArg_ParseTuple(args, "n:zfill", &width))
+        return NULL;
+
+    if (STRINGLIB_LEN(self) >= width) {
+        if (STRINGLIB_CHECK_EXACT(self)) {
+#if STRINGLIB_MUTABLE
+            /* We're defined as returning a copy;  If the object is mutable
+             * that means we must make an identical copy. */
+            return STRINGLIB_NEW(STRINGLIB_STR(self), STRINGLIB_LEN(self));
+#else
+            Py_INCREF(self);
+            return (PyObject*) self;
+#endif
+        }
+        else
+            return STRINGLIB_NEW(
+                STRINGLIB_STR(self),
+                STRINGLIB_LEN(self)
+            );
+    }
+
+    fill = width - STRINGLIB_LEN(self);
+
+    s = pad(self, fill, 0, '0');
+
+    if (s == NULL)
+        return NULL;
+
+    p = STRINGLIB_STR(s);
+    if (p[fill] == '+' || p[fill] == '-') {
+        /* move sign to beginning of string */
+        p[0] = p[fill];
+        p[fill] = '0';
+    }
+
+    return (PyObject*) s;
+}

lib_python/2.7_Objects/stringlib/unicodedefs.h

+#ifndef STRINGLIB_UNICODEDEFS_H
+#define STRINGLIB_UNICODEDEFS_H
+
+/* this is sort of a hack.  there's at least one place (formatting
+   floats) where some stringlib code takes a different path if it's
+   compiled as unicode. */
+#define STRINGLIB_IS_UNICODE     1
+
+#define STRINGLIB_OBJECT         PyUnicodeObject
+#define STRINGLIB_CHAR           Py_UNICODE
+#define STRINGLIB_TYPE_NAME      "unicode"
+#define STRINGLIB_PARSE_CODE     "U"
+#define STRINGLIB_EMPTY          unicode_empty
+#define STRINGLIB_ISSPACE        Py_UNICODE_ISSPACE
+#define STRINGLIB_ISLINEBREAK    BLOOM_LINEBREAK
+#define STRINGLIB_ISDECIMAL      Py_UNICODE_ISDECIMAL
+#define STRINGLIB_TODECIMAL      Py_UNICODE_TODECIMAL
+#define STRINGLIB_TOUPPER        Py_UNICODE_TOUPPER
+#define STRINGLIB_TOLOWER        Py_UNICODE_TOLOWER
+#define STRINGLIB_FILL           Py_UNICODE_FILL
+#define STRINGLIB_STR            PyUnicode_AS_UNICODE
+#define STRINGLIB_LEN            PyUnicode_GET_SIZE
+#define STRINGLIB_NEW            PyUnicode_FromUnicode
+#define STRINGLIB_RESIZE         PyUnicode_Resize
+#define STRINGLIB_CHECK          PyUnicode_Check
+#define STRINGLIB_CHECK_EXACT    PyUnicode_CheckExact
+#define STRINGLIB_GROUPING       _PyUnicode_InsertThousandsGrouping
+
+#if PY_VERSION_HEX < 0x03000000
+#define STRINGLIB_TOSTR          PyObject_Unicode
+#else
+#define STRINGLIB_TOSTR          PyObject_Str
+#endif
+
+#define STRINGLIB_WANT_CONTAINS_OBJ 1
+
+#endif /* !STRINGLIB_UNICODEDEFS_H */

lib_python/2.7_Objects/stringobject.c

+// Pyston change: this is just a shim to import stuff from stringlib/
+
+#include "Python.h"
+
+#include "stringlib/stringdefs.h"
+#include "stringlib/string_format.h"
+
+#define _Py_InsertThousandsGrouping _PyString_InsertThousandsGrouping
+#include "stringlib/localeutil.h"

lib_python/2.7_Python/dtoa.c

+// This file is originally from CPython 2.7, with modifications for Pyston
+
+/****************************************************************
+ *
+ * The author of this software is David M. Gay.
+ *
+ * Copyright (c) 1991, 2000, 2001 by Lucent Technologies.
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose without fee is hereby granted, provided that this entire notice
+ * is included in all copies of any software which is or includes a copy
+ * or modification of this software and in all copies of the supporting
+ * documentation for such software.
+ *
+ * THIS SOFTWARE IS BEING PROVIDED "AS IS", WITHOUT ANY EXPRESS OR IMPLIED
+ * WARRANTY.  IN PARTICULAR, NEITHER THE AUTHOR NOR LUCENT MAKES ANY
+ * REPRESENTATION OR WARRANTY OF ANY KIND CONCERNING THE MERCHANTABILITY
+ * OF THIS SOFTWARE OR ITS FITNESS FOR ANY PARTICULAR PURPOSE.
+ *
+ ***************************************************************/
+
+/****************************************************************
+ * This is dtoa.c by David M. Gay, downloaded from
+ * http://www.netlib.org/fp/dtoa.c on April 15, 2009 and modified for
+ * inclusion into the Python core by Mark E. T. Dickinson and Eric V. Smith.
+ *
+ * Please remember to check http://www.netlib.org/fp regularly (and especially
+ * before any Python release) for bugfixes and updates.
+ *
+ * The major modifications from Gay's original code are as follows:
+ *
+ *  0. The original code has been specialized to Python's needs by removing
+ *     many of the #ifdef'd sections.  In particular, code to support VAX and
+ *     IBM floating-point formats, hex NaNs, hex floats, locale-aware
+ *     treatment of the decimal point, and setting of the inexact flag have
+ *     been removed.
+ *
+ *  1. We use PyMem_Malloc and PyMem_Free in place of malloc and free.
+ *
+ *  2. The public functions strtod, dtoa and freedtoa all now have
+ *     a _Py_dg_ prefix.
+ *
+ *  3. Instead of assuming that PyMem_Malloc always succeeds, we thread
+ *     PyMem_Malloc failures through the code.  The functions
+ *
+ *       Balloc, multadd, s2b, i2b, mult, pow5mult, lshift, diff, d2b
+ *
+ *     of return type *Bigint all return NULL to indicate a malloc failure.
+ *     Similarly, rv_alloc and nrv_alloc (return type char *) return NULL on
+ *     failure.  bigcomp now has return type int (it used to be void) and
+ *     returns -1 on failure and 0 otherwise.  _Py_dg_dtoa returns NULL
+ *     on failure.  _Py_dg_strtod indicates failure due to malloc failure
+ *     by returning -1.0, setting errno=ENOMEM and *se to s00.
+ *
+ *  4. The static variable dtoa_result has been removed.  Callers of
+ *     _Py_dg_dtoa are expected to call _Py_dg_freedtoa to free
+ *     the memory allocated by _Py_dg_dtoa.
+ *
+ *  5. The code has been reformatted to better fit with Python's
+ *     C style guide (PEP 7).
+ *
+ *  6. A bug in the memory allocation has been fixed: to avoid FREEing memory
+ *     that hasn't been MALLOC'ed, private_mem should only be used when k <=
+ *     Kmax.
+ *
+ *  7. _Py_dg_strtod has been modified so that it doesn't accept strings with
+ *     leading whitespace.
+ *
+ ***************************************************************/
+
+/* Please send bug reports for the original dtoa.c code to David M. Gay (dmg
+ * at acm dot org, with " at " changed at "@" and " dot " changed to ".").
+ * Please report bugs for this modified version using the Python issue tracker
+ * (http://bugs.python.org). */
+
+/* On a machine with IEEE extended-precision registers, it is
+ * necessary to specify double-precision (53-bit) rounding precision
+ * before invoking strtod or dtoa.  If the machine uses (the equivalent
+ * of) Intel 80x87 arithmetic, the call
+ *      _control87(PC_53, MCW_PC);
+ * does this with many compilers.  Whether this or another call is
+ * appropriate depends on the compiler; for this to work, it may be
+ * necessary to #include "float.h" or another system-dependent header
+ * file.
+ */
+
+/* strtod for IEEE-, VAX-, and IBM-arithmetic machines.
+ *
+ * This strtod returns a nearest machine number to the input decimal
+ * string (or sets errno to ERANGE).  With IEEE arithmetic, ties are
+ * broken by the IEEE round-even rule.  Otherwise ties are broken by
+ * biased rounding (add half and chop).
+ *
+ * Inspired loosely by William D. Clinger's paper "How to Read Floating
+ * Point Numbers Accurately" [Proc. ACM SIGPLAN '90, pp. 92-101].
+ *
+ * Modifications:
+ *
+ *      1. We only require IEEE, IBM, or VAX double-precision
+ *              arithmetic (not IEEE double-extended).
+ *      2. We get by with floating-point arithmetic in a case that
+ *              Clinger missed -- when we're computing d * 10^n
+ *              for a small integer d and the integer n is not too
+ *              much larger than 22 (the maximum integer k for which
+ *              we can represent 10^k exactly), we may be able to
+ *              compute (d*10^k) * 10^(e-k) with just one roundoff.
+ *      3. Rather than a bit-at-a-time adjustment of the binary
+ *              result in the hard case, we use floating-point
+ *              arithmetic to determine the adjustment to within
+ *              one bit; only in really hard cases do we need to
+ *              compute a second residual.
+ *      4. Because of 3., we don't need a large table of powers of 10
+ *              for ten-to-e (just some small tables, e.g. of 10^k
+ *              for 0 <= k <= 22).
+ */
+
+/* Linking of Python's #defines to Gay's #defines starts here. */
+
+#include "Python.h"
+
+/* if PY_NO_SHORT_FLOAT_REPR is defined, then don't even try to compile
+   the following code */
+#ifndef PY_NO_SHORT_FLOAT_REPR
+
+#include "float.h"
+
+#define MALLOC PyMem_Malloc
+#define FREE PyMem_Free
+
+/* This code should also work for ARM mixed-endian format on little-endian
+   machines, where doubles have byte order 45670123 (in increasing address
+   order, 0 being the least significant byte). */
+#ifdef DOUBLE_IS_LITTLE_ENDIAN_IEEE754
+#  define IEEE_8087
+#endif
+#if defined(DOUBLE_IS_BIG_ENDIAN_IEEE754) ||  \
+  defined(DOUBLE_IS_ARM_MIXED_ENDIAN_IEEE754)
+#  define IEEE_MC68k
+#endif
+#if defined(IEEE_8087) + defined(IEEE_MC68k) != 1
+#error "Exactly one of IEEE_8087 or IEEE_MC68k should be defined."
+#endif
+
+/* The code below assumes that the endianness of integers matches the
+   endianness of the two 32-bit words of a double.  Check this. */
+#if defined(WORDS_BIGENDIAN) && (defined(DOUBLE_IS_LITTLE_ENDIAN_IEEE754) || \
+                                 defined(DOUBLE_IS_ARM_MIXED_ENDIAN_IEEE754))
+#error "doubles and ints have incompatible endianness"
+#endif
+
+#if !defined(WORDS_BIGENDIAN) && defined(DOUBLE_IS_BIG_ENDIAN_IEEE754)
+#error "doubles and ints have incompatible endianness"
+#endif
+
+
+#if defined(HAVE_UINT32_T) && defined(HAVE_INT32_T)
+typedef PY_UINT32_T ULong;
+typedef PY_INT32_T Long;
+#else
+#error "Failed to find an exact-width 32-bit integer type"
+#endif
+
+#if defined(HAVE_UINT64_T)
+#define ULLong PY_UINT64_T
+#else
+#undef ULLong
+#endif
+
+#undef DEBUG
+#ifdef Py_DEBUG
+#define DEBUG
+#endif
+
+/* End Python #define linking */
+
+#ifdef DEBUG
+#define Bug(x) {fprintf(stderr, "%s\n", x); exit(1);}
+#endif
+
+#ifndef PRIVATE_MEM
+#define PRIVATE_MEM 2304
+#endif
+#define PRIVATE_mem ((PRIVATE_MEM+sizeof(double)-1)/sizeof(double))
+static double private_mem[PRIVATE_mem], *pmem_next = private_mem;
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+typedef union { double d; ULong L[2]; } U;
+
+#ifdef IEEE_8087
+#define word0(x) (x)->L[1]
+#define word1(x) (x)->L[0]
+#else
+#define word0(x) (x)->L[0]
+#define word1(x) (x)->L[1]
+#endif
+#define dval(x) (x)->d
+
+#ifndef STRTOD_DIGLIM
+#define STRTOD_DIGLIM 40
+#endif
+
+/* maximum permitted exponent value for strtod; exponents larger than
+   MAX_ABS_EXP in absolute value get truncated to +-MAX_ABS_EXP.  MAX_ABS_EXP
+   should fit into an int. */
+#ifndef MAX_ABS_EXP
+#define MAX_ABS_EXP 1100000000U
+#endif
+/* Bound on length of pieces of input strings in _Py_dg_strtod; specifically,
+   this is used to bound the total number of digits ignoring leading zeros and
+   the number of digits that follow the decimal point.  Ideally, MAX_DIGITS
+   should satisfy MAX_DIGITS + 400 < MAX_ABS_EXP; that ensures that the
+   exponent clipping in _Py_dg_strtod can't affect the value of the output. */
+#ifndef MAX_DIGITS
+#define MAX_DIGITS 1000000000U
+#endif
+
+/* Guard against trying to use the above values on unusual platforms with ints
+ * of width less than 32 bits. */
+#if MAX_ABS_EXP > INT_MAX
+#error "MAX_ABS_EXP should fit in an int"
+#endif
+#if MAX_DIGITS > INT_MAX
+#error "MAX_DIGITS should fit in an int"
+#endif
+
+/* The following definition of Storeinc is appropriate for MIPS processors.
+ * An alternative that might be better on some machines is
+ * #define Storeinc(a,b,c) (*a++ = b << 16 | c & 0xffff)
+ */
+#if defined(IEEE_8087)
+#define Storeinc(a,b,c) (((unsigned short *)a)[1] = (unsigned short)b,  \
+                         ((unsigned short *)a)[0] = (unsigned short)c, a++)
+#else
+#define Storeinc(a,b,c) (((unsigned short *)a)[0] = (unsigned short)b,  \
+                         ((unsigned short *)a)[1] = (unsigned short)c, a++)
+#endif
+
+/* #define P DBL_MANT_DIG */
+/* Ten_pmax = floor(P*log(2)/log(5)) */
+/* Bletch = (highest power of 2 < DBL_MAX_10_EXP) / 16 */
+/* Quick_max = floor((P-1)*log(FLT_RADIX)/log(10) - 1) */
+/* Int_max = floor(P*log(FLT_RADIX)/log(10) - 1) */
+
+#define Exp_shift  20
+#define Exp_shift1 20
+#define Exp_msk1    0x100000
+#define Exp_msk11   0x100000
+#define Exp_mask  0x7ff00000
+#define P 53
+#define Nbits 53
+#define Bias 1023
+#define Emax 1023
+#define Emin (-1022)
+#define Etiny (-1074)  /* smallest denormal is 2**Etiny */
+#define Exp_1  0x3ff00000
+#define Exp_11 0x3ff00000
+#define Ebits 11
+#define Frac_mask  0xfffff
+#define Frac_mask1 0xfffff
+#define Ten_pmax 22
+#define Bletch 0x10
+#define Bndry_mask  0xfffff
+#define Bndry_mask1 0xfffff
+#define Sign_bit 0x80000000
+#define Log2P 1
+#define Tiny0 0
+#define Tiny1 1
+#define Quick_max 14
+#define Int_max 14
+
+#ifndef Flt_Rounds
+#ifdef FLT_ROUNDS
+#define Flt_Rounds FLT_ROUNDS
+#else
+#define Flt_Rounds 1
+#endif
+#endif /*Flt_Rounds*/
+
+#define Rounding Flt_Rounds
+
+#define Big0 (Frac_mask1 | Exp_msk1*(DBL_MAX_EXP+Bias-1))
+#define Big1 0xffffffff
+
+/* struct BCinfo is used to pass information from _Py_dg_strtod to bigcomp */
+
+typedef struct BCinfo BCinfo;
+struct
+BCinfo {
+    int e0, nd, nd0, scale;
+};
+
+#define FFFFFFFF 0xffffffffUL
+
+#define Kmax 7
+
+/* struct Bigint is used to represent arbitrary-precision integers.  These
+   integers are stored in sign-magnitude format, with the magnitude stored as
+   an array of base 2**32 digits.  Bigints are always normalized: if x is a
+   Bigint then x->wds >= 1, and either x->wds == 1 or x[wds-1] is nonzero.
+
+   The Bigint fields are as follows:
+
+     - next is a header used by Balloc and Bfree to keep track of lists
+         of freed Bigints;  it's also used for the linked list of
+         powers of 5 of the form 5**2**i used by pow5mult.
+     - k indicates which pool this Bigint was allocated from
+     - maxwds is the maximum number of words space was allocated for
+       (usually maxwds == 2**k)
+     - sign is 1 for negative Bigints, 0 for positive.  The sign is unused
+       (ignored on inputs, set to 0 on outputs) in almost all operations
+       involving Bigints: a notable exception is the diff function, which
+       ignores signs on inputs but sets the sign of the output correctly.
+     - wds is the actual number of significant words
+     - x contains the vector of words (digits) for this Bigint, from least
+       significant (x[0]) to most significant (x[wds-1]).
+*/
+
+struct
+Bigint {
+    struct Bigint *next;
+    int k, maxwds, sign, wds;
+    ULong x[1];
+};
+
+typedef struct Bigint Bigint;
+
+#ifndef Py_USING_MEMORY_DEBUGGER
+
+/* Memory management: memory is allocated from, and returned to, Kmax+1 pools
+   of memory, where pool k (0 <= k <= Kmax) is for Bigints b with b->maxwds ==
+   1 << k.  These pools are maintained as linked lists, with freelist[k]
+   pointing to the head of the list for pool k.
+
+   On allocation, if there's no free slot in the appropriate pool, MALLOC is
+   called to get more memory.  This memory is not returned to the system until
+   Python quits.  There's also a private memory pool that's allocated from
+   in preference to using MALLOC.
+
+   For Bigints with more than (1 << Kmax) digits (which implies at least 1233
+   decimal digits), memory is directly allocated using MALLOC, and freed using
+   FREE.
+
+   XXX: it would be easy to bypass this memory-management system and
+   translate each call to Balloc into a call to PyMem_Malloc, and each
+   Bfree to PyMem_Free.  Investigate whether this has any significant
+   performance on impact. */
+
+static Bigint *freelist[Kmax+1];
+
+/* Allocate space for a Bigint with up to 1<<k digits */
+
+static Bigint *
+Balloc(int k)
+{
+    int x;
+    Bigint *rv;
+    unsigned int len;
+
+    if (k <= Kmax && (rv = freelist[k]))
+        freelist[k] = rv->next;
+    else {
+        x = 1 << k;
+        len = (sizeof(Bigint) + (x-1)*sizeof(ULong) + sizeof(double) - 1)
+            /sizeof(double);
+        if (k <= Kmax && pmem_next - private_mem + len <= PRIVATE_mem) {
+            rv = (Bigint*)pmem_next;
+            pmem_next += len;
+        }
+        else {
+            rv = (Bigint*)MALLOC(len*sizeof(double));
+            if (rv == NULL)
+                return NULL;
+        }
+        rv->k = k;
+        rv->maxwds = x;
+    }
+    rv->sign = rv->wds = 0;
+    return rv;
+}
+
+/* Free a Bigint allocated with Balloc */
+
+static void
+Bfree(Bigint *v)
+{
+    if (v) {
+        if (v->k > Kmax)
+            FREE((void*)v);
+        else {
+            v->next = freelist[v->k];
+            freelist[v->k] = v;
+        }
+    }
+}
+
+#else
+
+/* Alternative versions of Balloc and Bfree that use PyMem_Malloc and
+   PyMem_Free directly in place of the custom memory allocation scheme above.
+   These are provided for the benefit of memory debugging tools like
+   Valgrind. */
+
+/* Allocate space for a Bigint with up to 1<<k digits */
+
+static Bigint *
+Balloc(int k)
+{
+    int x;
+    Bigint *rv;
+    unsigned int len;
+
+    x = 1 << k;
+    len = (sizeof(Bigint) + (x-1)*sizeof(ULong) + sizeof(double) - 1)
+        /sizeof(double);
+
+    rv = (Bigint*)MALLOC(len*sizeof(double));
+    if (rv == NULL)
+        return NULL;
+
+    rv->k = k;
+    rv->maxwds = x;
+    rv->sign = rv->wds = 0;
+    return rv;
+}
+
+/* Free a Bigint allocated with Balloc */
+
+static void
+Bfree(Bigint *v)
+{
+    if (v) {
+        FREE((void*)v);
+    }
+}
+
+#endif /* Py_USING_MEMORY_DEBUGGER */
+
+#define Bcopy(x,y) memcpy((char *)&x->sign, (char *)&y->sign,   \
+                          y->wds*sizeof(Long) + 2*sizeof(int))
+
+/* Multiply a Bigint b by m and add a.  Either modifies b in place and returns
+   a pointer to the modified b, or Bfrees b and returns a pointer to a copy.
+   On failure, return NULL.  In this case, b will have been already freed. */
+
+static Bigint *
+multadd(Bigint *b, int m, int a)       /* multiply by m and add a */
+{
+    int i, wds;
+#ifdef ULLong
+    ULong *x;
+    ULLong carry, y;
+#else
+    ULong carry, *x, y;
+    ULong xi, z;
+#endif
+    Bigint *b1;
+
+    wds = b->wds;
+    x = b->x;
+    i = 0;
+    carry = a;
+    do {
+#ifdef ULLong
+        y = *x * (ULLong)m + carry;
+        carry = y >> 32;
+        *x++ = (ULong)(y & FFFFFFFF);
+#else
+        xi = *x;
+        y = (xi & 0xffff) * m + carry;
+        z = (xi >> 16) * m + (y >> 16);
+        carry = z >> 16;
+        *x++ = (z << 16) + (y & 0xffff);
+#endif
+    }
+    while(++i < wds);
+    if (carry) {
+        if (wds >= b->maxwds) {
+            b1 = Balloc(b->k+1);
+            if (b1 == NULL){
+                Bfree(b);
+                return NULL;
+            }
+            Bcopy(b1, b);
+            Bfree(b);
+            b = b1;
+        }
+        b->x[wds++] = (ULong)carry;
+        b->wds = wds;
+    }
+    return b;
+}
+
+/* convert a string s containing nd decimal digits (possibly containing a
+   decimal separator at position nd0, which is ignored) to a Bigint.  This
+   function carries on where the parsing code in _Py_dg_strtod leaves off: on
+   entry, y9 contains the result of converting the first 9 digits.  Returns
+   NULL on failure. */
+
+static Bigint *
+s2b(const char *s, int nd0, int nd, ULong y9)
+{
+    Bigint *b;
+    int i, k;
+    Long x, y;
+
+    x = (nd + 8) / 9;
+    for(k = 0, y = 1; x > y; y <<= 1, k++) ;
+    b = Balloc(k);
+    if (b == NULL)
+        return NULL;
+    b->x[0] = y9;
+    b->wds = 1;
+
+    if (nd <= 9)
+      return b;
+
+    s += 9;
+    for (i = 9; i < nd0; i++) {
+        b = multadd(b, 10, *s++ - '0');
+        if (b == NULL)
+            return NULL;
+    }
+    s++;
+    for(; i < nd; i++) {
+        b = multadd(b, 10, *s++ - '0');
+        if (b == NULL)
+            return NULL;
+    }
+    return b;
+}
+
+/* count leading 0 bits in the 32-bit integer x. */
+
+static int
+hi0bits(ULong x)
+{
+    int k = 0;
+
+    if (!(x & 0xffff0000)) {
+        k = 16;
+        x <<= 16;
+    }
+    if (!(x & 0xff000000)) {
+        k += 8;
+        x <<= 8;
+    }
+    if (!(x & 0xf0000000)) {
+        k += 4;
+        x <<= 4;
+    }
+    if (!(x & 0xc0000000)) {
+        k += 2;
+        x <<= 2;
+    }
+    if (!(x & 0x80000000)) {
+        k++;
+        if (!(x & 0x40000000))
+            return 32;
+    }
+    return k;
+}
+
+/* count trailing 0 bits in the 32-bit integer y, and shift y right by that
+   number of bits. */
+
+static int
+lo0bits(ULong *y)
+{
+    int k;
+    ULong x = *y;
+
+    if (x & 7) {
+        if (x & 1)
+            return 0;
+        if (x & 2) {
+            *y = x >> 1;
+            return 1;
+        }
+        *y = x >> 2;
+        return 2;
+    }
+    k = 0;
+    if (!(x & 0xffff)) {
+        k = 16;
+        x >>= 16;
+    }
+    if (!(x & 0xff)) {
+        k += 8;
+        x >>= 8;
+    }
+    if (!(x & 0xf)) {
+        k += 4;
+        x >>= 4;
+    }
+    if (!(x & 0x3)) {
+        k += 2;
+        x >>= 2;
+    }
+    if (!(x & 1)) {
+        k++;
+        x >>= 1;
+        if (!x)
+            return 32;
+    }
+    *y = x;
+    return k;
+}
+
+/* convert a small nonnegative integer to a Bigint */
+
+static Bigint *
+i2b(int i)
+{
+    Bigint *b;
+
+    b = Balloc(1);
+    if (b == NULL)
+        return NULL;
+    b->x[0] = i;
+    b->wds = 1;
+    return b;
+}
+
+/* multiply two Bigints.  Returns a new Bigint, or NULL on failure.  Ignores
+   the signs of a and b. */
+
+static Bigint *
+mult(Bigint *a, Bigint *b)
+{
+    Bigint *c;
+    int k, wa, wb, wc;
+    ULong *x, *xa, *xae, *xb, *xbe, *xc, *xc0;
+    ULong y;
+#ifdef ULLong
+    ULLong carry, z;
+#else
+    ULong carry, z;
+    ULong z2;
+#endif
+
+    if ((!a->x[0] && a->wds == 1) || (!b->x[0] && b->wds == 1)) {
+        c = Balloc(0);
+        if (c == NULL)
+            return NULL;
+        c->wds = 1;
+        c->x[0] = 0;
+        return c;
+    }
+
+    if (a->wds < b->wds) {
+        c = a;
+        a = b;
+        b = c;
+    }
+    k = a->k;
+    wa = a->wds;
+    wb = b->wds;
+    wc = wa + wb;
+    if (wc > a->maxwds)
+        k++;
+    c = Balloc(k);
+    if (c == NULL)
+        return NULL;
+    for(x = c->x, xa = x + wc; x < xa; x++)
+        *x = 0;
+    xa = a->x;
+    xae = xa + wa;
+    xb = b->x;
+    xbe = xb + wb;
+    xc0 = c->x;
+#ifdef ULLong
+    for(; xb < xbe; xc0++) {
+        if ((y = *xb++)) {
+            x = xa;
+            xc = xc0;
+            carry = 0;
+            do {
+                z = *x++ * (ULLong)y + *xc + carry;
+                carry = z >> 32;
+                *xc++ = (ULong)(z & FFFFFFFF);
+            }
+            while(x < xae);
+            *xc = (ULong)carry;
+        }
+    }
+#else
+    for(; xb < xbe; xb++, xc0++) {
+        if (y = *xb & 0xffff) {
+            x = xa;
+            xc = xc0;
+            carry = 0;
+            do {
+                z = (*x & 0xffff) * y + (*xc & 0xffff) + carry;
+                carry = z >> 16;
+                z2 = (*x++ >> 16) * y + (*xc >> 16) + carry;
+                carry = z2 >> 16;
+                Storeinc(xc, z2, z);
+            }
+            while(x < xae);
+            *xc = carry;
+        }
+        if (y = *xb >> 16) {
+            x = xa;
+            xc = xc0;
+            carry = 0;
+            z2 = *xc;
+            do {
+                z = (*x & 0xffff) * y + (*xc >> 16) + carry;
+                carry = z >> 16;
+                Storeinc(xc, z, z2);
+                z2 = (*x++ >> 16) * y + (*xc & 0xffff) + carry;
+                carry = z2 >> 16;
+            }
+            while(x < xae);
+            *xc = z2;
+        }
+    }
+#endif
+    for(xc0 = c->x, xc = xc0 + wc; wc > 0 && !*--xc; --wc) ;
+    c->wds = wc;
+    return c;
+}
+
+#ifndef Py_USING_MEMORY_DEBUGGER
+
+/* p5s is a linked list of powers of 5 of the form 5**(2**i), i >= 2 */
+
+static Bigint *p5s;
+
+/* multiply the Bigint b by 5**k.  Returns a pointer to the result, or NULL on
+   failure; if the returned pointer is distinct from b then the original
+   Bigint b will have been Bfree'd.   Ignores the sign of b. */
+
+static Bigint *
+pow5mult(Bigint *b, int k)
+{
+    Bigint *b1, *p5, *p51;
+    int i;
+    static int p05[3] = { 5, 25, 125 };
+
+    if ((i = k & 3)) {
+        b = multadd(b, p05[i-1], 0);
+        if (b == NULL)
+            return NULL;
+    }
+
+    if (!(k >>= 2))
+        return b;
+    p5 = p5s;
+    if (!p5) {
+        /* first time */
+        p5 = i2b(625);
+        if (p5 == NULL) {
+            Bfree(b);
+            return NULL;
+        }
+        p5s = p5;
+        p5->next = 0;
+    }
+    for(;;) {
+        if (k & 1) {
+            b1 = mult(b, p5);
+            Bfree(b);
+            b = b1;
+            if (b == NULL)
+                return NULL;
+        }
+        if (!(k >>= 1))
+            break;
+        p51 = p5->next;
+        if (!p51) {
+            p51 = mult(p5,p5);
+            if (p51 == NULL) {
+                Bfree(b);
+                return NULL;
+            }
+            p51->next = 0;
+            p5->next = p51;
+        }
+        p5 = p51;
+    }
+    return b;
+}
+
+#else
+
+/* Version of pow5mult that doesn't cache powers of 5. Provided for
+   the benefit of memory debugging tools like Valgrind. */
+
+static Bigint *
+pow5mult(Bigint *b, int k)
+{
+    Bigint *b1, *p5, *p51;
+    int i;
+    static int p05[3] = { 5, 25, 125 };
+
+    if ((i = k & 3)) {
+        b = multadd(b, p05[i-1], 0);
+        if (b == NULL)
+            return NULL;
+    }
+
+    if (!(k >>= 2))
+        return b;
+    p5 = i2b(625);
+    if (p5 == NULL) {
+        Bfree(b);
+        return NULL;
+    }
+
+    for(;;) {
+        if (k & 1) {
+            b1 = mult(b, p5);
+            Bfree(b);
+            b = b1;
+            if (b == NULL) {
+                Bfree(p5);
+                return NULL;
+            }
+        }
+        if (!(k >>= 1))
+            break;
+        p51 = mult(p5, p5);
+        Bfree(p5);
+        p5 = p51;
+        if (p5 == NULL) {
+            Bfree(b);
+            return NULL;
+        }
+    }
+    Bfree(p5);
+    return b;
+}
+
+#endif /* Py_USING_MEMORY_DEBUGGER */
+
+/* shift a Bigint b left by k bits.  Return a pointer to the shifted result,
+   or NULL on failure.  If the returned pointer is distinct from b then the
+   original b will have been Bfree'd.   Ignores the sign of b. */
+
+static Bigint *
+lshift(Bigint *b, int k)
+{
+    int i, k1, n, n1;
+    Bigint *b1;
+    ULong *x, *x1, *xe, z;
+
+    if (!k || (!b->x[0] && b->wds == 1))
+        return b;
+
+    n = k >> 5;
+    k1 = b->k;
+    n1 = n + b->wds + 1;
+    for(i = b->maxwds; n1 > i; i <<= 1)
+        k1++;
+    b1 = Balloc(k1);
+    if (b1 == NULL) {
+        Bfree(b);
+        return NULL;
+    }
+    x1 = b1->x;
+    for(i = 0; i < n; i++)
+        *x1++ = 0;
+    x = b->x;
+    xe = x + b->wds;
+    if (k &= 0x1f) {
+        k1 = 32 - k;
+        z = 0;
+        do {
+            *x1++ = *x << k | z;
+            z = *x++ >> k1;
+        }
+        while(x < xe);
+        if ((*x1 = z))
+            ++n1;
+    }
+    else do
+             *x1++ = *x++;
+        while(x < xe);
+    b1->wds = n1 - 1;
+    Bfree(b);
+    return b1;
+}
+
+/* Do a three-way compare of a and b, returning -1 if a < b, 0 if a == b and
+   1 if a > b.  Ignores signs of a and b. */
+
+static int
+cmp(Bigint *a, Bigint *b)
+{
+    ULong *xa, *xa0, *xb, *xb0;
+    int i, j;
+
+    i = a->wds;
+    j = b->wds;
+#ifdef DEBUG
+    if (i > 1 && !a->x[i-1])
+        Bug("cmp called with a->x[a->wds-1] == 0");
+    if (j > 1 && !b->x[j-1])
+        Bug("cmp called with b->x[b->wds-1] == 0");
+#endif
+    if (i -= j)
+        return i;
+    xa0 = a->x;
+    xa = xa0 + j;
+    xb0 = b->x;
+    xb = xb0 + j;
+    for(;;) {
+        if (*--xa != *--xb)
+            return *xa < *xb ? -1 : 1;
+        if (xa <= xa0)
+            break;
+    }
+    return 0;
+}
+
+/* Take the difference of Bigints a and b, returning a new Bigint.  Returns
+   NULL on failure.  The signs of a and b are ignored, but the sign of the
+   result is set appropriately. */
+
+static Bigint *
+diff(Bigint *a, Bigint *b)
+{
+    Bigint *c;
+    int i, wa, wb;
+    ULong *xa, *xae, *xb, *xbe, *xc;
+#ifdef ULLong
+    ULLong borrow, y;
+#else
+    ULong borrow, y;
+    ULong z;
+#endif
+
+    i = cmp(a,b);
+    if (!i) {
+        c = Balloc(0);
+        if (c == NULL)
+            return NULL;
+        c->wds = 1;
+        c->x[0] = 0;
+        return c;
+    }
+    if (i < 0) {
+        c = a;
+        a = b;
+        b = c;
+        i = 1;
+    }
+    else
+        i = 0;
+    c = Balloc(a->k);
+    if (c == NULL)
+        return NULL;
+    c->sign = i;
+    wa = a->wds;
+    xa = a->x;
+    xae = xa + wa;
+    wb = b->wds;
+    xb = b->x;
+    xbe = xb + wb;
+    xc = c->x;
+    borrow = 0;
+#ifdef ULLong
+    do {
+        y = (ULLong)*xa++ - *xb++ - borrow;
+        borrow = y >> 32 & (ULong)1;
+        *xc++ = (ULong)(y & FFFFFFFF);
+    }
+    while(xb < xbe);
+    while(xa < xae) {
+        y = *xa++ - borrow;
+        borrow = y >> 32 & (ULong)1;
+        *xc++ = (ULong)(y & FFFFFFFF);
+    }
+#else
+    do {
+        y = (*xa & 0xffff) - (*xb & 0xffff) - borrow;
+        borrow = (y & 0x10000) >> 16;
+        z = (*xa++ >> 16) - (*xb++ >> 16) - borrow;
+        borrow = (z & 0x10000) >> 16;
+        Storeinc(xc, z, y);
+    }
+    while(xb < xbe);
+    while(xa < xae) {
+        y = (*xa & 0xffff) - borrow;
+        borrow = (y & 0x10000) >> 16;
+        z = (*xa++ >> 16) - borrow;
+        borrow = (z & 0x10000) >> 16;
+        Storeinc(xc, z, y);
+    }
+#endif
+    while(!*--xc)
+        wa--;
+    c->wds = wa;
+    return c;
+}
+
+/* Given a positive normal double x, return the difference between x and the
+   next double up.  Doesn't give correct results for subnormals. */
+
+static double
+ulp(U *x)
+{
+    Long L;
+    U u;
+
+    L = (word0(x) & Exp_mask) - (P-1)*Exp_msk1;
+    word0(&u) = L;
+    word1(&u) = 0;
+    return dval(&u);
+}
+
+/* Convert a Bigint to a double plus an exponent */
+
+static double
+b2d(Bigint *a, int *e)
+{
+    ULong *xa, *xa0, w, y, z;
+    int k;
+    U d;
+
+    xa0 = a->x;
+    xa = xa0 + a->wds;
+    y = *--xa;
+#ifdef DEBUG
+    if (!y) Bug("zero y in b2d");
+#endif
+    k = hi0bits(y);
+    *e = 32 - k;
+    if (k < Ebits) {
+        word0(&d) = Exp_1 | y >> (Ebits - k);
+        w = xa > xa0 ? *--xa : 0;
+        word1(&d) = y << ((32-Ebits) + k) | w >> (Ebits - k);
+        goto ret_d;
+    }
+    z = xa > xa0 ? *--xa : 0;
+    if (k -= Ebits) {
+        word0(&d) = Exp_1 | y << k | z >> (32 - k);
+        y = xa > xa0 ? *--xa : 0;
+        word1(&d) = z << k | y >> (32 - k);
+    }
+    else {
+        word0(&d) = Exp_1 | y;
+        word1(&d) = z;
+    }
+  ret_d:
+    return dval(&d);
+}
+
+/* Convert a scaled double to a Bigint plus an exponent.  Similar to d2b,
+   except that it accepts the scale parameter used in _Py_dg_strtod (which
+   should be either 0 or 2*P), and the normalization for the return value is
+   different (see below).  On input, d should be finite and nonnegative, and d
+   / 2**scale should be exactly representable as an IEEE 754 double.
+
+   Returns a Bigint b and an integer e such that
+
+     dval(d) / 2**scale = b * 2**e.
+
+   Unlike d2b, b is not necessarily odd: b and e are normalized so
+   that either 2**(P-1) <= b < 2**P and e >= Etiny, or b < 2**P
+   and e == Etiny.  This applies equally to an input of 0.0: in that
+   case the return values are b = 0 and e = Etiny.
+
+   The above normalization ensures that for all possible inputs d,
+   2**e gives ulp(d/2**scale).
+
+   Returns NULL on failure.
+*/
+
+static Bigint *
+sd2b(U *d, int scale, int *e)
+{
+    Bigint *b;
+
+    b = Balloc(1);
+    if (b == NULL)
+        return NULL;
+    
+    /* First construct b and e assuming that scale == 0. */
+    b->wds = 2;
+    b->x[0] = word1(d);
+    b->x[1] = word0(d) & Frac_mask;
+    *e = Etiny - 1 + (int)((word0(d) & Exp_mask) >> Exp_shift);
+    if (*e < Etiny)
+        *e = Etiny;
+    else
+        b->x[1] |= Exp_msk1;
+
+    /* Now adjust for scale, provided that b != 0. */
+    if (scale && (b->x[0] || b->x[1])) {
+        *e -= scale;
+        if (*e < Etiny) {
+            scale = Etiny - *e;
+            *e = Etiny;
+            /* We can't shift more than P-1 bits without shifting out a 1. */
+            assert(0 < scale && scale <= P - 1);
+            if (scale >= 32) {
+                /* The bits shifted out should all be zero. */
+                assert(b->x[0] == 0);
+                b->x[0] = b->x[1];
+                b->x[1] = 0;
+                scale -= 32;
+            }
+            if (scale) {
+                /* The bits shifted out should all be zero. */
+                assert(b->x[0] << (32 - scale) == 0);
+                b->x[0] = (b->x[0] >> scale) | (b->x[1] << (32 - scale));
+                b->x[1] >>= scale;
+            }
+        }
+    }
+    /* Ensure b is normalized. */
+    if (!b->x[1])
+        b->wds = 1;
+
+    return b;
+}
+
+/* Convert a double to a Bigint plus an exponent.  Return NULL on failure.
+
+   Given a finite nonzero double d, return an odd Bigint b and exponent *e
+   such that fabs(d) = b * 2**e.  On return, *bbits gives the number of
+   significant bits of b; that is, 2**(*bbits-1) <= b < 2**(*bbits).
+
+   If d is zero, then b == 0, *e == -1010, *bbits = 0.
+ */
+
+static Bigint *
+d2b(U *d, int *e, int *bits)
+{
+    Bigint *b;
+    int de, k;
+    ULong *x, y, z;
+    int i;
+
+    b = Balloc(1);
+    if (b == NULL)
+        return NULL;
+    x = b->x;
+
+    z = word0(d) & Frac_mask;
+    word0(d) &= 0x7fffffff;   /* clear sign bit, which we ignore */
+    if ((de = (int)(word0(d) >> Exp_shift)))
+        z |= Exp_msk1;
+    if ((y = word1(d))) {
+        if ((k = lo0bits(&y))) {
+            x[0] = y | z << (32 - k);
+            z >>= k;
+        }
+        else
+            x[0] = y;
+        i =
+            b->wds = (x[1] = z) ? 2 : 1;
+    }
+    else {
+        k = lo0bits(&z);
+        x[0] = z;
+        i =
+            b->wds = 1;
+        k += 32;
+    }
+    if (de) {
+        *e = de - Bias - (P-1) + k;
+        *bits = P - k;
+    }
+    else {
+        *e = de - Bias - (P-1) + 1 + k;
+        *bits = 32*i - hi0bits(x[i-1]);
+    }
+    return b;
+}
+
+/* Compute the ratio of two Bigints, as a double.  The result may have an
+   error of up to 2.5 ulps. */
+
+static double
+ratio(Bigint *a, Bigint *b)
+{
+    U da, db;
+    int k, ka, kb;
+
+    dval(&da) = b2d(a, &ka);
+    dval(&db) = b2d(b, &kb);
+    k = ka - kb + 32*(a->wds - b->wds);
+    if (k > 0)
+        word0(&da) += k*Exp_msk1;
+    else {
+        k = -k;
+        word0(&db) += k*Exp_msk1;
+    }
+    return dval(&da) / dval(&db);
+}
+
+static const double
+tens[] = {
+    1e0, 1e1, 1e2, 1e3, 1e4, 1e5, 1e6, 1e7, 1e8, 1e9,
+    1e10, 1e11, 1e12, 1e13, 1e14, 1e15, 1e16, 1e17, 1e18, 1e19,
+    1e20, 1e21, 1e22
+};
+
+static const double
+bigtens[] = { 1e16, 1e32, 1e64, 1e128, 1e256 };
+static const double tinytens[] = { 1e-16, 1e-32, 1e-64, 1e-128,
+                                   9007199254740992.*9007199254740992.e-256
+                                   /* = 2^106 * 1e-256 */
+};
+/* The factor of 2^53 in tinytens[4] helps us avoid setting the underflow */
+/* flag unnecessarily.  It leads to a song and dance at the end of strtod. */
+#define Scale_Bit 0x10
+#define n_bigtens 5
+
+#define ULbits 32
+#define kshift 5
+#define kmask 31
+
+
+static int
+dshift(Bigint *b, int p2)
+{
+    int rv = hi0bits(b->x[b->wds-1]) - 4;
+    if (p2 > 0)
+        rv -= p2;
+    return rv & kmask;
+}
+
+/* special case of Bigint division.  The quotient is always in the range 0 <=
+   quotient < 10, and on entry the divisor S is normalized so that its top 4
+   bits (28--31) are zero and bit 27 is set. */
+
+static int
+quorem(Bigint *b, Bigint *S)
+{
+    int n;
+    ULong *bx, *bxe, q, *sx, *sxe;
+#ifdef ULLong
+    ULLong borrow, carry, y, ys;
+#else
+    ULong borrow, carry, y, ys;
+    ULong si, z, zs;
+#endif
+
+    n = S->wds;
+#ifdef DEBUG
+    /*debug*/ if (b->wds > n)
+        /*debug*/       Bug("oversize b in quorem");
+#endif
+    if (b->wds < n)
+        return 0;
+    sx = S->x;
+    sxe = sx + --n;
+    bx = b->x;
+    bxe = bx + n;
+    q = *bxe / (*sxe + 1);      /* ensure q <= true quotient */
+#ifdef DEBUG
+    /*debug*/ if (q > 9)
+        /*debug*/       Bug("oversized quotient in quorem");
+#endif
+    if (q) {
+        borrow = 0;
+        carry = 0;
+        do {
+#ifdef ULLong
+            ys = *sx++ * (ULLong)q + carry;
+            carry = ys >> 32;
+            y = *bx - (ys & FFFFFFFF) - borrow;
+            borrow = y >> 32 & (ULong)1;
+            *bx++ = (ULong)(y & FFFFFFFF);
+#else
+            si = *sx++;
+            ys = (si & 0xffff) * q + carry;
+            zs = (si >> 16) * q + (ys >> 16);
+            carry = zs >> 16;
+            y = (*bx & 0xffff) - (ys & 0xffff) - borrow;
+            borrow = (y & 0x10000) >> 16;
+            z = (*bx >> 16) - (zs & 0xffff) - borrow;
+            borrow = (z & 0x10000) >> 16;
+            Storeinc(bx, z, y);
+#endif
+        }
+        while(sx <= sxe);
+        if (!*bxe) {
+            bx = b->x;
+            while(--bxe > bx && !*bxe)
+                --n;
+            b->wds = n;
+        }
+    }
+    if (cmp(b, S) >= 0) {
+        q++;
+        borrow = 0;
+        carry = 0;
+        bx = b->x;
+        sx = S->x;
+        do {
+#ifdef ULLong
+            ys = *sx++ + carry;
+            carry = ys >> 32;
+            y = *bx - (ys & FFFFFFFF) - borrow;
+            borrow = y >> 32 & (ULong)1;
+            *bx++ = (ULong)(y & FFFFFFFF);
+#else
+            si = *sx++;
+            ys = (si & 0xffff) + carry;
+            zs = (si >> 16) + (ys >> 16);
+            carry = zs >> 16;
+            y = (*bx & 0xffff) - (ys & 0xffff) - borrow;
+            borrow = (y & 0x10000) >> 16;
+            z = (*bx >> 16) - (zs & 0xffff) - borrow;
+            borrow = (z & 0x10000) >> 16;
+            Storeinc(bx, z, y);
+#endif
+        }
+        while(sx <= sxe);
+        bx = b->x;
+        bxe = bx + n;
+        if (!*bxe) {
+            while(--bxe > bx && !*bxe)
+                --n;
+            b->wds = n;
+        }
+    }
+    return q;
+}
+
+/* sulp(x) is a version of ulp(x) that takes bc.scale into account.
+
+   Assuming that x is finite and nonnegative (positive zero is fine
+   here) and x / 2^bc.scale is exactly representable as a double,
+   sulp(x) is equivalent to 2^bc.scale * ulp(x / 2^bc.scale). */
+
+static double
+sulp(U *x, BCinfo *bc)
+{
+    U u;
+
+    if (bc->scale && 2*P + 1 > (int)((word0(x) & Exp_mask) >> Exp_shift)) {
+        /* rv/2^bc->scale is subnormal */
+        word0(&u) = (P+2)*Exp_msk1;
+        word1(&u) = 0;
+        return u.d;
+    }
+    else {
+        assert(word0(x) || word1(x)); /* x != 0.0 */
+        return ulp(x);
+    }
+}
+
+/* The bigcomp function handles some hard cases for strtod, for inputs
+   with more than STRTOD_DIGLIM digits.  It's called once an initial
+   estimate for the double corresponding to the input string has
+   already been obtained by the code in _Py_dg_strtod.
+
+   The bigcomp function is only called after _Py_dg_strtod has found a
+   double value rv such that either rv or rv + 1ulp represents the
+   correctly rounded value corresponding to the original string.  It
+   determines which of these two values is the correct one by
+   computing the decimal digits of rv + 0.5ulp and comparing them with
+   the corresponding digits of s0.
+
+   In the following, write dv for the absolute value of the number represented
+   by the input string.
+
+   Inputs:
+
+     s0 points to the first significant digit of the input string.
+
+     rv is a (possibly scaled) estimate for the closest double value to the
+        value represented by the original input to _Py_dg_strtod.  If
+        bc->scale is nonzero, then rv/2^(bc->scale) is the approximation to
+        the input value.
+
+     bc is a struct containing information gathered during the parsing and
+        estimation steps of _Py_dg_strtod.  Description of fields follows:
+
+        bc->e0 gives the exponent of the input value, such that dv = (integer
+           given by the bd->nd digits of s0) * 10**e0
+
+        bc->nd gives the total number of significant digits of s0.  It will
+           be at least 1.
+
+        bc->nd0 gives the number of significant digits of s0 before the
+           decimal separator.  If there's no decimal separator, bc->nd0 ==
+           bc->nd.
+
+        bc->scale is the value used to scale rv to avoid doing arithmetic with
+           subnormal values.  It's either 0 or 2*P (=106).
+
+   Outputs:
+
+     On successful exit, rv/2^(bc->scale) is the closest double to dv.
+
+     Returns 0 on success, -1 on failure (e.g., due to a failed malloc call). */
+
+static int
+bigcomp(U *rv, const char *s0, BCinfo *bc)
+{
+    Bigint *b, *d;
+    int b2, d2, dd, i, nd, nd0, odd, p2, p5;
+
+    nd = bc->nd;
+    nd0 = bc->nd0;
+    p5 = nd + bc->e0;
+    b = sd2b(rv, bc->scale, &p2);
+    if (b == NULL)
+        return -1;
+
+    /* record whether the lsb of rv/2^(bc->scale) is odd:  in the exact halfway
+       case, this is used for round to even. */
+    odd = b->x[0] & 1;
+
+    /* left shift b by 1 bit and or a 1 into the least significant bit;
+       this gives us b * 2**p2 = rv/2^(bc->scale) + 0.5 ulp. */
+    b = lshift(b, 1);
+    if (b == NULL)
+        return -1;
+    b->x[0] |= 1;
+    p2--;
+
+    p2 -= p5;
+    d = i2b(1);
+    if (d == NULL) {
+        Bfree(b);
+        return -1;
+    }
+    /* Arrange for convenient computation of quotients:
+     * shift left if necessary so divisor has 4 leading 0 bits.
+     */
+    if (p5 > 0) {
+        d = pow5mult(d, p5);
+        if (d == NULL) {
+            Bfree(b);
+            return -1;
+        }
+    }
+    else if (p5 < 0) {
+        b = pow5mult(b, -p5);
+        if (b == NULL) {
+            Bfree(d);
+            return -1;
+        }
+    }
+    if (p2 > 0) {
+        b2 = p2;
+        d2 = 0;
+    }
+    else {
+        b2 = 0;
+        d2 = -p2;
+    }
+    i = dshift(d, d2);
+    if ((b2 += i) > 0) {
+        b = lshift(b, b2);
+        if (b == NULL) {
+            Bfree(d);
+            return -1;
+        }
+    }
+    if ((d2 += i) > 0) {
+        d = lshift(d, d2);
+        if (d == NULL) {
+            Bfree(b);
+            return -1;
+        }
+    }
+
+    /* Compare s0 with b/d: set dd to -1, 0, or 1 according as s0 < b/d, s0 ==
+     * b/d, or s0 > b/d.  Here the digits of s0 are thought of as representing
+     * a number in the range [0.1, 1). */
+    if (cmp(b, d) >= 0)
+        /* b/d >= 1 */
+        dd = -1;
+    else {
+        i = 0;
+        for(;;) {
+            b = multadd(b, 10, 0);
+            if (b == NULL) {
+                Bfree(d);
+                return -1;
+            }
+            dd = s0[i < nd0 ? i : i+1] - '0' - quorem(b, d);
+            i++;
+
+            if (dd)
+                break;
+            if (!b->x[0] && b->wds == 1) {
+                /* b/d == 0 */
+                dd = i < nd;
+                break;
+            }
+            if (!(i < nd)) {
+                /* b/d != 0, but digits of s0 exhausted */
+                dd = -1;
+                break;
+            }
+        }
+    }
+    Bfree(b);
+    Bfree(d);
+    if (dd > 0 || (dd == 0 && odd))
+        dval(rv) += sulp(rv, bc);
+    return 0;
+}
+
+double
+_Py_dg_strtod(const char *s00, char **se)
+{
+    int bb2, bb5, bbe, bd2, bd5, bs2, c, dsign, e, e1, error;
+    int esign, i, j, k, lz, nd, nd0, odd, sign;
+    const char *s, *s0, *s1;
+    double aadj, aadj1;
+    U aadj2, adj, rv, rv0;
+    ULong y, z, abs_exp;
+    Long L;
+    BCinfo bc;
+    Bigint *bb, *bb1, *bd, *bd0, *bs, *delta;
+    size_t ndigits, fraclen;
+
+    dval(&rv) = 0.;
+
+    /* Start parsing. */
+    c = *(s = s00);
+
+    /* Parse optional sign, if present. */
+    sign = 0;
+    switch (c) {
+    case '-':
+        sign = 1;
+        /* no break */
+    case '+':
+        c = *++s;
+    }
+
+    /* Skip leading zeros: lz is true iff there were leading zeros. */
+    s1 = s;
+    while (c == '0')
+        c = *++s;
+    lz = s != s1;
+
+    /* Point s0 at the first nonzero digit (if any).  fraclen will be the
+       number of digits between the decimal point and the end of the
+       digit string.  ndigits will be the total number of digits ignoring
+       leading zeros. */
+    s0 = s1 = s;
+    while ('0' <= c && c <= '9')
+        c = *++s;
+    ndigits = s - s1;
+    fraclen = 0;
+
+    /* Parse decimal point and following digits. */
+    if (c == '.') {
+        c = *++s;
+        if (!ndigits) {
+            s1 = s;
+            while (c == '0')
+                c = *++s;
+            lz = lz || s != s1;
+            fraclen += (s - s1);
+            s0 = s;
+        }
+        s1 = s;
+        while ('0' <= c && c <= '9')
+            c = *++s;
+        ndigits += s - s1;
+        fraclen += s - s1;
+    }
+
+    /* Now lz is true if and only if there were leading zero digits, and
+       ndigits gives the total number of digits ignoring leading zeros.  A
+       valid input must have at least one digit. */
+    if (!ndigits && !lz) {
+        if (se)
+            *se = (char *)s00;
+        goto parse_error;
+    }
+
+    /* Range check ndigits and fraclen to make sure that they, and values
+       computed with them, can safely fit in an int. */
+    if (ndigits > MAX_DIGITS || fraclen > MAX_DIGITS) {
+        if (se)
+            *se = (char *)s00;
+        goto parse_error;
+    }
+    nd = (int)ndigits;
+    nd0 = (int)ndigits - (int)fraclen;
+
+    /* Parse exponent. */
+    e = 0;
+    if (c == 'e' || c == 'E') {
+        s00 = s;
+        c = *++s;
+
+        /* Exponent sign. */
+        esign = 0;
+        switch (c) {
+        case '-':
+            esign = 1;
+            /* no break */
+        case '+':
+            c = *++s;
+        }
+
+        /* Skip zeros.  lz is true iff there are leading zeros. */
+        s1 = s;
+        while (c == '0')
+            c = *++s;
+        lz = s != s1;
+
+        /* Get absolute value of the exponent. */
+        s1 = s;
+        abs_exp = 0;
+        while ('0' <= c && c <= '9') {
+            abs_exp = 10*abs_exp + (c - '0');
+            c = *++s;
+        }
+
+        /* abs_exp will be correct modulo 2**32.  But 10**9 < 2**32, so if
+           there are at most 9 significant exponent digits then overflow is
+           impossible. */
+        if (s - s1 > 9 || abs_exp > MAX_ABS_EXP)
+            e = (int)MAX_ABS_EXP;
+        else
+            e = (int)abs_exp;
+        if (esign)
+            e = -e;
+
+        /* A valid exponent must have at least one digit. */
+        if (s == s1 && !lz)
+            s = s00;
+    }
+
+    /* Adjust exponent to take into account position of the point. */
+    e -= nd - nd0;
+    if (nd0 <= 0)
+        nd0 = nd;
+
+    /* Finished parsing.  Set se to indicate how far we parsed */
+    if (se)
+        *se = (char *)s;
+
+    /* If all digits were zero, exit with return value +-0.0.  Otherwise,
+       strip trailing zeros: scan back until we hit a nonzero digit. */
+    if (!nd)
+        goto ret;
+    for (i = nd; i > 0; ) {
+        --i;
+        if (s0[i < nd0 ? i : i+1] != '0') {
+            ++i;
+            break;
+        }
+    }
+    e += nd - i;
+    nd = i;
+    if (nd0 > nd)
+        nd0 = nd;
+
+    /* Summary of parsing results.  After parsing, and dealing with zero
+     * inputs, we have values s0, nd0, nd, e, sign, where:
+     *
+     *  - s0 points to the first significant digit of the input string
+     *
+     *  - nd is the total number of significant digits (here, and
+     *    below, 'significant digits' means the set of digits of the
+     *    significand of the input that remain after ignoring leading
+     *    and trailing zeros).
+     *
+     *  - nd0 indicates the position of the decimal point, if present; it
+     *    satisfies 1 <= nd0 <= nd.  The nd significant digits are in
+     *    s0[0:nd0] and s0[nd0+1:nd+1] using the usual Python half-open slice
+     *    notation.  (If nd0 < nd, then s0[nd0] contains a '.'  character; if
+     *    nd0 == nd, then s0[nd0] could be any non-digit character.)
+     *
+     *  - e is the adjusted exponent: the absolute value of the number
+     *    represented by the original input string is n * 10**e, where
+     *    n is the integer represented by the concatenation of
+     *    s0[0:nd0] and s0[nd0+1:nd+1]
+     *
+     *  - sign gives the sign of the input:  1 for negative, 0 for positive
+     *
+     *  - the first and last significant digits are nonzero
+     */
+
+    /* put first DBL_DIG+1 digits into integer y and z.
+     *
+     *  - y contains the value represented by the first min(9, nd)
+     *    significant digits
+     *
+     *  - if nd > 9, z contains the value represented by significant digits
+     *    with indices in [9, min(16, nd)).  So y * 10**(min(16, nd) - 9) + z
+     *    gives the value represented by the first min(16, nd) sig. digits.
+     */
+
+    bc.e0 = e1 = e;
+    y = z = 0;
+    for (i = 0; i < nd; i++) {
+        if (i < 9)
+            y = 10*y + s0[i < nd0 ? i : i+1] - '0';
+        else if (i < DBL_DIG+1)
+            z = 10*z + s0[i < nd0 ? i : i+1] - '0';
+        else
+            break;
+    }
+
+    k = nd < DBL_DIG + 1 ? nd : DBL_DIG + 1;
+    dval(&rv) = y;
+    if (k > 9) {
+        dval(&rv) = tens[k - 9] * dval(&rv) + z;
+    }
+    bd0 = 0;
+    if (nd <= DBL_DIG
+        && Flt_Rounds == 1
+        ) {
+        if (!e)
+            goto ret;
+        if (e > 0) {
+            if (e <= Ten_pmax) {
+                dval(&rv) *= tens[e];
+                goto ret;
+            }
+            i = DBL_DIG - nd;
+            if (e <= Ten_pmax + i) {
+                /* A fancier test would sometimes let us do
+                 * this for larger i values.
+                 */
+                e -= i;
+                dval(&rv) *= tens[i];
+                dval(&rv) *= tens[e];
+                goto ret;
+            }
+        }
+        else if (e >= -Ten_pmax) {
+            dval(&rv) /= tens[-e];
+            goto ret;
+        }
+    }
+    e1 += nd - k;
+
+    bc.scale = 0;
+
+    /* Get starting approximation = rv * 10**e1 */
+
+    if (e1 > 0) {
+        if ((i = e1 & 15))
+            dval(&rv) *= tens[i];
+        if (e1 &= ~15) {
+            if (e1 > DBL_MAX_10_EXP)
+                goto ovfl;
+            e1 >>= 4;
+            for(j = 0; e1 > 1; j++, e1 >>= 1)
+                if (e1 & 1)
+                    dval(&rv) *= bigtens[j];
+            /* The last multiplication could overflow. */
+            word0(&rv) -= P*Exp_msk1;
+            dval(&rv) *= bigtens[j];
+            if ((z = word0(&rv) & Exp_mask)
+                > Exp_msk1*(DBL_MAX_EXP+Bias-P))
+                goto ovfl;
+            if (z > Exp_msk1*(DBL_MAX_EXP+Bias-1-P)) {
+                /* set to largest number */
+                /* (Can't trust DBL_MAX) */
+                word0(&rv) = Big0;
+                word1(&rv) = Big1;
+            }
+            else
+                word0(&rv) += P*Exp_msk1;
+        }
+    }
+    else if (e1 < 0) {
+        /* The input decimal value lies in [10**e1, 10**(e1+16)).
+
+           If e1 <= -512, underflow immediately.
+           If e1 <= -256, set bc.scale to 2*P.
+
+           So for input value < 1e-256, bc.scale is always set;
+           for input value >= 1e-240, bc.scale is never set.
+           For input values in [1e-256, 1e-240), bc.scale may or may
+           not be set. */
+
+        e1 = -e1;
+        if ((i = e1 & 15))
+            dval(&rv) /= tens[i];
+        if (e1 >>= 4) {
+            if (e1 >= 1 << n_bigtens)
+                goto undfl;
+            if (e1 & Scale_Bit)
+                bc.scale = 2*P;
+            for(j = 0; e1 > 0; j++, e1 >>= 1)
+                if (e1 & 1)
+                    dval(&rv) *= tinytens[j];
+            if (bc.scale && (j = 2*P + 1 - ((word0(&rv) & Exp_mask)
+                                            >> Exp_shift)) > 0) {
+                /* scaled rv is denormal; clear j low bits */
+                if (j >= 32) {
+                    word1(&rv) = 0;
+                    if (j >= 53)
+                        word0(&rv) = (P+2)*Exp_msk1;
+                    else
+                        word0(&rv) &= 0xffffffff << (j-32);
+                }
+                else
+                    word1(&rv) &= 0xffffffff << j;
+            }
+            if (!dval(&rv))
+                goto undfl;
+        }
+    }
+
+    /* Now the hard part -- adjusting rv to the correct value.*/
+
+    /* Put digits into bd: true value = bd * 10^e */
+
+    bc.nd = nd;
+    bc.nd0 = nd0;       /* Only needed if nd > STRTOD_DIGLIM, but done here */
+                        /* to silence an erroneous warning about bc.nd0 */
+                        /* possibly not being initialized. */
+    if (nd > STRTOD_DIGLIM) {
+        /* ASSERT(STRTOD_DIGLIM >= 18); 18 == one more than the */
+        /* minimum number of decimal digits to distinguish double values */
+        /* in IEEE arithmetic. */
+
+        /* Truncate input to 18 significant digits, then discard any trailing
+           zeros on the result by updating nd, nd0, e and y suitably. (There's
+           no need to update z; it's not reused beyond this point.) */
+        for (i = 18; i > 0; ) {
+            /* scan back until we hit a nonzero digit.  significant digit 'i'
+            is s0[i] if i < nd0, s0[i+1] if i >= nd0. */
+            --i;
+            if (s0[i < nd0 ? i : i+1] != '0') {
+                ++i;
+                break;
+            }
+        }
+        e += nd - i;
+        nd = i;
+        if (nd0 > nd)
+            nd0 = nd;
+        if (nd < 9) { /* must recompute y */
+            y = 0;
+            for(i = 0; i < nd0; ++i)
+                y = 10*y + s0[i] - '0';
+            for(; i < nd; ++i)
+                y = 10*y + s0[i+1] - '0';
+        }
+    }
+    bd0 = s2b(s0, nd0, nd, y);
+    if (bd0 == NULL)
+        goto failed_malloc;
+
+    /* Notation for the comments below.  Write:
+
+         - dv for the absolute value of the number represented by the original
+           decimal input string.
+
+         - if we've truncated dv, write tdv for the truncated value.
+           Otherwise, set tdv == dv.
+
+         - srv for the quantity rv/2^bc.scale; so srv is the current binary
+           approximation to tdv (and dv).  It should be exactly representable
+           in an IEEE 754 double.
+    */
+
+    for(;;) {
+
+        /* This is the main correction loop for _Py_dg_strtod.
+
+           We've got a decimal value tdv, and a floating-point approximation
+           srv=rv/2^bc.scale to tdv.  The aim is to determine whether srv is
+           close enough (i.e., within 0.5 ulps) to tdv, and to compute a new
+           approximation if not.
+
+           To determine whether srv is close enough to tdv, compute integers
+           bd, bb and bs proportional to tdv, srv and 0.5 ulp(srv)
+           respectively, and then use integer arithmetic to determine whether
+           |tdv - srv| is less than, equal to, or greater than 0.5 ulp(srv).
+        */
+
+        bd = Balloc(bd0->k);
+        if (bd == NULL) {
+            Bfree(bd0);
+            goto failed_malloc;
+        }
+        Bcopy(bd, bd0);
+        bb = sd2b(&rv, bc.scale, &bbe);   /* srv = bb * 2^bbe */
+        if (bb == NULL) {
+            Bfree(bd);
+            Bfree(bd0);
+            goto failed_malloc;
+        }
+        /* Record whether lsb of bb is odd, in case we need this
+           for the round-to-even step later. */
+        odd = bb->x[0] & 1;
+
+        /* tdv = bd * 10**e;  srv = bb * 2**bbe */
+        bs = i2b(1);
+        if (bs == NULL) {
+            Bfree(bb);
+            Bfree(bd);
+            Bfree(bd0);
+            goto failed_malloc;
+        }
+
+        if (e >= 0) {
+            bb2 = bb5 = 0;
+            bd2 = bd5 = e;
+        }
+        else {
+            bb2 = bb5 = -e;
+            bd2 = bd5 = 0;
+        }
+        if (bbe >= 0)
+            bb2 += bbe;
+        else
+            bd2 -= bbe;
+        bs2 = bb2;
+        bb2++;
+        bd2++;
+
+        /* At this stage bd5 - bb5 == e == bd2 - bb2 + bbe, bb2 - bs2 == 1,
+           and bs == 1, so:
+
+              tdv == bd * 10**e = bd * 2**(bbe - bb2 + bd2) * 5**(bd5 - bb5)
+              srv == bb * 2**bbe = bb * 2**(bbe - bb2 + bb2)
+              0.5 ulp(srv) == 2**(bbe-1) = bs * 2**(bbe - bb2 + bs2)
+
+           It follows that:
+
+              M * tdv = bd * 2**bd2 * 5**bd5
+              M * srv = bb * 2**bb2 * 5**bb5
+              M * 0.5 ulp(srv) = bs * 2**bs2 * 5**bb5
+
+           for some constant M.  (Actually, M == 2**(bb2 - bbe) * 5**bb5, but
+           this fact is not needed below.)
+        */
+
+        /* Remove factor of 2**i, where i = min(bb2, bd2, bs2). */
+        i = bb2 < bd2 ? bb2 : bd2;
+        if (i > bs2)
+            i = bs2;
+        if (i > 0) {
+            bb2 -= i;
+            bd2 -= i;
+            bs2 -= i;
+        }
+
+        /* Scale bb, bd, bs by the appropriate powers of 2 and 5. */
+        if (bb5 > 0) {
+            bs = pow5mult(bs, bb5);
+            if (bs == NULL) {
+                Bfree(bb);
+                Bfree(bd);
+                Bfree(bd0);
+                goto failed_malloc;
+            }
+            bb1 = mult(bs, bb);
+            Bfree(bb);
+            bb = bb1;
+            if (bb == NULL) {
+                Bfree(bs);
+                Bfree(bd);
+                Bfree(bd0);
+                goto failed_malloc;
+            }
+        }
+        if (bb2 > 0) {
+            bb = lshift(bb, bb2);
+            if (bb == NULL) {
+                Bfree(bs);
+                Bfree(bd);
+                Bfree(bd0);
+                goto failed_malloc;
+            }
+        }
+        if (bd5 > 0) {
+            bd = pow5mult(bd, bd5);
+            if (bd == NULL) {
+                Bfree(bb);
+                Bfree(bs);
+                Bfree(bd0);
+                goto failed_malloc;
+            }
+        }
+        if (bd2 > 0) {
+            bd = lshift(bd, bd2);
+            if (bd == NULL) {
+                Bfree(bb);
+                Bfree(bs);
+                Bfree(bd0);
+                goto failed_malloc;
+            }
+        }
+        if (bs2 > 0) {
+            bs = lshift(bs, bs2);
+            if (bs == NULL) {
+                Bfree(bb);
+                Bfree(bd);
+                Bfree(bd0);
+                goto failed_malloc;
+            }
+        }
+
+        /* Now bd, bb and bs are scaled versions of tdv, srv and 0.5 ulp(srv),
+           respectively.  Compute the difference |tdv - srv|, and compare
+           with 0.5 ulp(srv). */
+
+        delta = diff(bb, bd);
+        if (delta == NULL) {
+            Bfree(bb);
+            Bfree(bs);
+            Bfree(bd);
+            Bfree(bd0);
+            goto failed_malloc;
+        }
+        dsign = delta->sign;
+        delta->sign = 0;
+        i = cmp(delta, bs);
+        if (bc.nd > nd && i <= 0) {
+            if (dsign)
+                break;  /* Must use bigcomp(). */
+
+            /* Here rv overestimates the truncated decimal value by at most
+               0.5 ulp(rv).  Hence rv either overestimates the true decimal
+               value by <= 0.5 ulp(rv), or underestimates it by some small
+               amount (< 0.1 ulp(rv)); either way, rv is within 0.5 ulps of
+               the true decimal value, so it's possible to exit.
+
+               Exception: if scaled rv is a normal exact power of 2, but not
+               DBL_MIN, then rv - 0.5 ulp(rv) takes us all the way down to the
+               next double, so the correctly rounded result is either rv - 0.5
+               ulp(rv) or rv; in this case, use bigcomp to distinguish. */
+
+            if (!word1(&rv) && !(word0(&rv) & Bndry_mask)) {
+                /* rv can't be 0, since it's an overestimate for some
+                   nonzero value.  So rv is a normal power of 2. */
+                j = (int)(word0(&rv) & Exp_mask) >> Exp_shift;
+                /* rv / 2^bc.scale = 2^(j - 1023 - bc.scale); use bigcomp if
+                   rv / 2^bc.scale >= 2^-1021. */
+                if (j - bc.scale >= 2) {
+                    dval(&rv) -= 0.5 * sulp(&rv, &bc);
+                    break; /* Use bigcomp. */
+                }
+            }
+
+            {
+                bc.nd = nd;
+                i = -1; /* Discarded digits make delta smaller. */
+            }
+        }
+
+        if (i < 0) {
+            /* Error is less than half an ulp -- check for
+             * special case of mantissa a power of two.
+             */
+            if (dsign || word1(&rv) || word0(&rv) & Bndry_mask
+                || (word0(&rv) & Exp_mask) <= (2*P+1)*Exp_msk1
+                ) {
+                break;
+            }
+            if (!delta->x[0] && delta->wds <= 1) {
+                /* exact result */
+                break;
+            }
+            delta = lshift(delta,Log2P);
+            if (delta == NULL) {
+                Bfree(bb);
+                Bfree(bs);
+                Bfree(bd);
+                Bfree(bd0);
+                goto failed_malloc;
+            }
+            if (cmp(delta, bs) > 0)
+                goto drop_down;
+            break;
+        }
+        if (i == 0) {
+            /* exactly half-way between */
+            if (dsign) {
+                if ((word0(&rv) & Bndry_mask1) == Bndry_mask1
+                    &&  word1(&rv) == (
+                        (bc.scale &&
+                         (y = word0(&rv) & Exp_mask) <= 2*P*Exp_msk1) ?
+                        (0xffffffff & (0xffffffff << (2*P+1-(y>>Exp_shift)))) :
+                        0xffffffff)) {
+                    /*boundary case -- increment exponent*/
+                    word0(&rv) = (word0(&rv) & Exp_mask)
+                        + Exp_msk1
+                        ;
+                    word1(&rv) = 0;
+                    dsign = 0;
+                    break;
+                }
+            }
+            else if (!(word0(&rv) & Bndry_mask) && !word1(&rv)) {
+              drop_down:
+                /* boundary case -- decrement exponent */
+                if (bc.scale) {
+                    L = word0(&rv) & Exp_mask;
+                    if (L <= (2*P+1)*Exp_msk1) {
+                        if (L > (P+2)*Exp_msk1)
+                            /* round even ==> */
+                            /* accept rv */
+                            break;
+                        /* rv = smallest denormal */
+                        if (bc.nd > nd)
+                            break;
+                        goto undfl;
+                    }
+                }
+                L = (word0(&rv) & Exp_mask) - Exp_msk1;
+                word0(&rv) = L | Bndry_mask1;
+                word1(&rv) = 0xffffffff;
+                break;
+            }
+            if (!odd)
+                break;
+            if (dsign)
+                dval(&rv) += sulp(&rv, &bc);
+            else {
+                dval(&rv) -= sulp(&rv, &bc);
+                if (!dval(&rv)) {
+                    if (bc.nd >nd)
+                        break;
+                    goto undfl;
+                }
+            }
+            dsign = 1 - dsign;
+            break;
+        }
+        if ((aadj = ratio(delta, bs)) <= 2.) {
+            if (dsign)
+                aadj = aadj1 = 1.;
+            else if (word1(&rv) || word0(&rv) & Bndry_mask) {
+                if (word1(&rv) == Tiny1 && !word0(&rv)) {
+                    if (bc.nd >nd)
+                        break;
+                    goto undfl;
+                }
+                aadj = 1.;
+                aadj1 = -1.;
+            }
+            else {
+                /* special case -- power of FLT_RADIX to be */
+                /* rounded down... */
+
+                if (aadj < 2./FLT_RADIX)
+                    aadj = 1./FLT_RADIX;
+                else
+                    aadj *= 0.5;
+                aadj1 = -aadj;
+            }
+        }
+        else {
+            aadj *= 0.5;
+            aadj1 = dsign ? aadj : -aadj;
+            if (Flt_Rounds == 0)
+                aadj1 += 0.5;
+        }
+        y = word0(&rv) & Exp_mask;
+
+        /* Check for overflow */
+
+        if (y == Exp_msk1*(DBL_MAX_EXP+Bias-1)) {
+            dval(&rv0) = dval(&rv);
+            word0(&rv) -= P*Exp_msk1;
+            adj.d = aadj1 * ulp(&rv);
+            dval(&rv) += adj.d;
+            if ((word0(&rv) & Exp_mask) >=
+                Exp_msk1*(DBL_MAX_EXP+Bias-P)) {
+                if (word0(&rv0) == Big0 && word1(&rv0) == Big1) {
+                    Bfree(bb);
+                    Bfree(bd);
+                    Bfree(bs);
+                    Bfree(bd0);
+                    Bfree(delta);
+                    goto ovfl;
+                }
+                word0(&rv) = Big0;
+                word1(&rv) = Big1;
+                goto cont;
+            }
+            else
+                word0(&rv) += P*Exp_msk1;
+        }
+        else {
+            if (bc.scale && y <= 2*P*Exp_msk1) {
+                if (aadj <= 0x7fffffff) {
+                    if ((z = (ULong)aadj) <= 0)
+                        z = 1;
+                    aadj = z;
+                    aadj1 = dsign ? aadj : -aadj;
+                }
+                dval(&aadj2) = aadj1;
+                word0(&aadj2) += (2*P+1)*Exp_msk1 - y;
+                aadj1 = dval(&aadj2);
+            }
+            adj.d = aadj1 * ulp(&rv);
+            dval(&rv) += adj.d;
+        }
+        z = word0(&rv) & Exp_mask;
+        if (bc.nd == nd) {
+            if (!bc.scale)
+                if (y == z) {
+                    /* Can we stop now? */
+                    L = (Long)aadj;
+                    aadj -= L;
+                    /* The tolerances below are conservative. */
+                    if (dsign || word1(&rv) || word0(&rv) & Bndry_mask) {
+                        if (aadj < .4999999 || aadj > .5000001)
+                            break;
+                    }
+                    else if (aadj < .4999999/FLT_RADIX)
+                        break;
+                }
+        }
+      cont:
+        Bfree(bb);
+        Bfree(bd);
+        Bfree(bs);
+        Bfree(delta);
+    }
+    Bfree(bb);
+    Bfree(bd);
+    Bfree(bs);
+    Bfree(bd0);
+    Bfree(delta);
+    if (bc.nd > nd) {
+        error = bigcomp(&rv, s0, &bc);
+        if (error)
+            goto failed_malloc;
+    }
+
+    if (bc.scale) {
+        word0(&rv0) = Exp_1 - 2*P*Exp_msk1;
+        word1(&rv0) = 0;
+        dval(&rv) *= dval(&rv0);
+    }
+
+  ret:
+    return sign ? -dval(&rv) : dval(&rv);
+
+  parse_error:
+    return 0.0;
+
+  failed_malloc:
+    errno = ENOMEM;
+    return -1.0;
+
+  undfl:
+    return sign ? -0.0 : 0.0;
+
+  ovfl:
+    errno = ERANGE;
+    /* Can't trust HUGE_VAL */
+    word0(&rv) = Exp_mask;
+    word1(&rv) = 0;
+    return sign ? -dval(&rv) : dval(&rv);
+
+}
+
+static char *
+rv_alloc(int i)
+{
+    int j, k, *r;
+
+    j = sizeof(ULong);
+    for(k = 0;
+        sizeof(Bigint) - sizeof(ULong) - sizeof(int) + j <= (unsigned)i;
+        j <<= 1)
+        k++;
+    r = (int*)Balloc(k);
+    if (r == NULL)
+        return NULL;
+    *r = k;
+    return (char *)(r+1);
+}
+
+static char *
+nrv_alloc(char *s, char **rve, int n)
+{
+    char *rv, *t;
+
+    rv = rv_alloc(n);
+    if (rv == NULL)
+        return NULL;
+    t = rv;
+    while((*t = *s++)) t++;
+    if (rve)
+        *rve = t;
+    return rv;
+}
+
+/* freedtoa(s) must be used to free values s returned by dtoa
+ * when MULTIPLE_THREADS is #defined.  It should be used in all cases,
+ * but for consistency with earlier versions of dtoa, it is optional
+ * when MULTIPLE_THREADS is not defined.
+ */
+
+void
+_Py_dg_freedtoa(char *s)
+{
+    Bigint *b = (Bigint *)((int *)s - 1);
+    b->maxwds = 1 << (b->k = *(int*)b);
+    Bfree(b);
+}
+
+/* dtoa for IEEE arithmetic (dmg): convert double to ASCII string.
+ *
+ * Inspired by "How to Print Floating-Point Numbers Accurately" by
+ * Guy L. Steele, Jr. and Jon L. White [Proc. ACM SIGPLAN '90, pp. 112-126].
+ *
+ * Modifications:
+ *      1. Rather than iterating, we use a simple numeric overestimate
+ *         to determine k = floor(log10(d)).  We scale relevant
+ *         quantities using O(log2(k)) rather than O(k) multiplications.
+ *      2. For some modes > 2 (corresponding to ecvt and fcvt), we don't
+ *         try to generate digits strictly left to right.  Instead, we
+ *         compute with fewer bits and propagate the carry if necessary
+ *         when rounding the final digit up.  This is often faster.
+ *      3. Under the assumption that input will be rounded nearest,
+ *         mode 0 renders 1e23 as 1e23 rather than 9.999999999999999e22.
+ *         That is, we allow equality in stopping tests when the
+ *         round-nearest rule will give the same floating-point value
+ *         as would satisfaction of the stopping test with strict
+ *         inequality.
+ *      4. We remove common factors of powers of 2 from relevant
+ *         quantities.
+ *      5. When converting floating-point integers less than 1e16,
+ *         we use floating-point arithmetic rather than resorting
+ *         to multiple-precision integers.
+ *      6. When asked to produce fewer than 15 digits, we first try
+ *         to get by with floating-point arithmetic; we resort to
+ *         multiple-precision integer arithmetic only if we cannot
+ *         guarantee that the floating-point calculation has given
+ *         the correctly rounded result.  For k requested digits and
+ *         "uniformly" distributed input, the probability is
+ *         something like 10^(k-15) that we must resort to the Long
+ *         calculation.
+ */
+
+/* Additional notes (METD): (1) returns NULL on failure.  (2) to avoid memory
+   leakage, a successful call to _Py_dg_dtoa should always be matched by a
+   call to _Py_dg_freedtoa. */
+
+char *
+_Py_dg_dtoa(double dd, int mode, int ndigits,
+            int *decpt, int *sign, char **rve)
+{
+    /*  Arguments ndigits, decpt, sign are similar to those
+        of ecvt and fcvt; trailing zeros are suppressed from
+        the returned string.  If not null, *rve is set to point
+        to the end of the return value.  If d is +-Infinity or NaN,
+        then *decpt is set to 9999.
+
+        mode:
+        0 ==> shortest string that yields d when read in
+        and rounded to nearest.
+        1 ==> like 0, but with Steele & White stopping rule;
+        e.g. with IEEE P754 arithmetic , mode 0 gives
+        1e23 whereas mode 1 gives 9.999999999999999e22.
+        2 ==> max(1,ndigits) significant digits.  This gives a
+        return value similar to that of ecvt, except
+        that trailing zeros are suppressed.
+        3 ==> through ndigits past the decimal point.  This
+        gives a return value similar to that from fcvt,
+        except that trailing zeros are suppressed, and
+        ndigits can be negative.
+        4,5 ==> similar to 2 and 3, respectively, but (in
+        round-nearest mode) with the tests of mode 0 to
+        possibly return a shorter string that rounds to d.
+        With IEEE arithmetic and compilation with
+        -DHonor_FLT_ROUNDS, modes 4 and 5 behave the same
+        as modes 2 and 3 when FLT_ROUNDS != 1.
+        6-9 ==> Debugging modes similar to mode - 4:  don't try
+        fast floating-point estimate (if applicable).
+
+        Values of mode other than 0-9 are treated as mode 0.
+
+        Sufficient space is allocated to the return value
+        to hold the suppressed trailing zeros.
+    */
+
+    int bbits, b2, b5, be, dig, i, ieps, ilim, ilim0, ilim1,
+        j, j1, k, k0, k_check, leftright, m2, m5, s2, s5,
+        spec_case, try_quick;
+    Long L;
+    int denorm;
+    ULong x;
+    Bigint *b, *b1, *delta, *mlo, *mhi, *S;
+    U d2, eps, u;
+    double ds;
+    char *s, *s0;
+
+    /* set pointers to NULL, to silence gcc compiler warnings and make
+       cleanup easier on error */
+    mlo = mhi = S = 0;
+    s0 = 0;
+
+    u.d = dd;
+    if (word0(&u) & Sign_bit) {
+        /* set sign for everything, including 0's and NaNs */
+        *sign = 1;
+        word0(&u) &= ~Sign_bit; /* clear sign bit */
+    }
+    else
+        *sign = 0;
+
+    /* quick return for Infinities, NaNs and zeros */
+    if ((word0(&u) & Exp_mask) == Exp_mask)
+    {
+        /* Infinity or NaN */
+        *decpt = 9999;
+        if (!word1(&u) && !(word0(&u) & 0xfffff))
+            return nrv_alloc("Infinity", rve, 8);
+        return nrv_alloc("NaN", rve, 3);
+    }
+    if (!dval(&u)) {
+        *decpt = 1;
+        return nrv_alloc("0", rve, 1);
+    }
+
+    /* compute k = floor(log10(d)).  The computation may leave k
+       one too large, but should never leave k too small. */
+    b = d2b(&u, &be, &bbits);
+    if (b == NULL)
+        goto failed_malloc;
+    if ((i = (int)(word0(&u) >> Exp_shift1 & (Exp_mask>>Exp_shift1)))) {
+        dval(&d2) = dval(&u);
+        word0(&d2) &= Frac_mask1;
+        word0(&d2) |= Exp_11;
+
+        /* log(x)       ~=~ log(1.5) + (x-1.5)/1.5
+         * log10(x)      =  log(x) / log(10)
+         *              ~=~ log(1.5)/log(10) + (x-1.5)/(1.5*log(10))
+         * log10(d) = (i-Bias)*log(2)/log(10) + log10(d2)
+         *
+         * This suggests computing an approximation k to log10(d) by
+         *
+         * k = (i - Bias)*0.301029995663981
+         *      + ( (d2-1.5)*0.289529654602168 + 0.176091259055681 );
+         *
+         * We want k to be too large rather than too small.
+         * The error in the first-order Taylor series approximation
+         * is in our favor, so we just round up the constant enough
+         * to compensate for any error in the multiplication of
+         * (i - Bias) by 0.301029995663981; since |i - Bias| <= 1077,
+         * and 1077 * 0.30103 * 2^-52 ~=~ 7.2e-14,
+         * adding 1e-13 to the constant term more than suffices.
+         * Hence we adjust the constant term to 0.1760912590558.
+         * (We could get a more accurate k by invoking log10,
+         *  but this is probably not worthwhile.)
+         */
+
+        i -= Bias;
+        denorm = 0;
+    }
+    else {
+        /* d is denormalized */
+
+        i = bbits + be + (Bias + (P-1) - 1);
+        x = i > 32  ? word0(&u) << (64 - i) | word1(&u) >> (i - 32)
+            : word1(&u) << (32 - i);
+        dval(&d2) = x;
+        word0(&d2) -= 31*Exp_msk1; /* adjust exponent */
+        i -= (Bias + (P-1) - 1) + 1;
+        denorm = 1;
+    }
+    ds = (dval(&d2)-1.5)*0.289529654602168 + 0.1760912590558 +
+        i*0.301029995663981;
+    k = (int)ds;
+    if (ds < 0. && ds != k)
+        k--;    /* want k = floor(ds) */
+    k_check = 1;
+    if (k >= 0 && k <= Ten_pmax) {
+        if (dval(&u) < tens[k])
+            k--;
+        k_check = 0;
+    }
+    j = bbits - i - 1;
+    if (j >= 0) {
+        b2 = 0;
+        s2 = j;
+    }
+    else {
+        b2 = -j;
+        s2 = 0;
+    }
+    if (k >= 0) {
+        b5 = 0;
+        s5 = k;
+        s2 += k;
+    }
+    else {
+        b2 -= k;
+        b5 = -k;
+        s5 = 0;
+    }
+    if (mode < 0 || mode > 9)
+        mode = 0;
+
+    try_quick = 1;
+
+    if (mode > 5) {
+        mode -= 4;
+        try_quick = 0;
+    }
+    leftright = 1;
+    ilim = ilim1 = -1;  /* Values for cases 0 and 1; done here to */
+    /* silence erroneous "gcc -Wall" warning. */
+    switch(mode) {
+    case 0:
+    case 1:
+        i = 18;
+        ndigits = 0;
+        break;
+    case 2:
+        leftright = 0;
+        /* no break */
+    case 4:
+        if (ndigits <= 0)
+            ndigits = 1;
+        ilim = ilim1 = i = ndigits;
+        break;
+    case 3:
+        leftright = 0;
+        /* no break */
+    case 5:
+        i = ndigits + k + 1;
+        ilim = i;
+        ilim1 = i - 1;
+        if (i <= 0)
+            i = 1;
+    }
+    s0 = rv_alloc(i);
+    if (s0 == NULL)
+        goto failed_malloc;
+    s = s0;
+
+
+    if (ilim >= 0 && ilim <= Quick_max && try_quick) {
+
+        /* Try to get by with floating-point arithmetic. */
+
+        i = 0;
+        dval(&d2) = dval(&u);
+        k0 = k;
+        ilim0 = ilim;
+        ieps = 2; /* conservative */
+        if (k > 0) {
+            ds = tens[k&0xf];
+            j = k >> 4;
+            if (j & Bletch) {
+                /* prevent overflows */
+                j &= Bletch - 1;
+                dval(&u) /= bigtens[n_bigtens-1];
+                ieps++;
+            }
+            for(; j; j >>= 1, i++)
+                if (j & 1) {
+                    ieps++;
+                    ds *= bigtens[i];
+                }
+            dval(&u) /= ds;
+        }
+        else if ((j1 = -k)) {
+            dval(&u) *= tens[j1 & 0xf];
+            for(j = j1 >> 4; j; j >>= 1, i++)
+                if (j & 1) {
+                    ieps++;
+                    dval(&u) *= bigtens[i];
+                }
+        }
+        if (k_check && dval(&u) < 1. && ilim > 0) {
+            if (ilim1 <= 0)
+                goto fast_failed;
+            ilim = ilim1;
+            k--;
+            dval(&u) *= 10.;
+            ieps++;
+        }
+        dval(&eps) = ieps*dval(&u) + 7.;
+        word0(&eps) -= (P-1)*Exp_msk1;
+        if (ilim == 0) {
+            S = mhi = 0;
+            dval(&u) -= 5.;
+            if (dval(&u) > dval(&eps))
+                goto one_digit;
+            if (dval(&u) < -dval(&eps))
+                goto no_digits;
+            goto fast_failed;
+        }
+        if (leftright) {
+            /* Use Steele & White method of only
+             * generating digits needed.
+             */
+            dval(&eps) = 0.5/tens[ilim-1] - dval(&eps);
+            for(i = 0;;) {
+                L = (Long)dval(&u);
+                dval(&u) -= L;
+                *s++ = '0' + (int)L;
+                if (dval(&u) < dval(&eps))
+                    goto ret1;
+                if (1. - dval(&u) < dval(&eps))
+                    goto bump_up;
+                if (++i >= ilim)
+                    break;
+                dval(&eps) *= 10.;
+                dval(&u) *= 10.;
+            }
+        }
+        else {
+            /* Generate ilim digits, then fix them up. */
+            dval(&eps) *= tens[ilim-1];
+            for(i = 1;; i++, dval(&u) *= 10.) {
+                L = (Long)(dval(&u));
+                if (!(dval(&u) -= L))
+                    ilim = i;
+                *s++ = '0' + (int)L;
+                if (i == ilim) {
+                    if (dval(&u) > 0.5 + dval(&eps))
+                        goto bump_up;
+                    else if (dval(&u) < 0.5 - dval(&eps)) {
+                        while(*--s == '0');
+                        s++;
+                        goto ret1;
+                    }
+                    break;
+                }
+            }
+        }
+      fast_failed:
+        s = s0;
+        dval(&u) = dval(&d2);
+        k = k0;
+        ilim = ilim0;
+    }
+
+    /* Do we have a "small" integer? */
+
+    if (be >= 0 && k <= Int_max) {
+        /* Yes. */
+        ds = tens[k];
+        if (ndigits < 0 && ilim <= 0) {
+            S = mhi = 0;
+            if (ilim < 0 || dval(&u) <= 5*ds)
+                goto no_digits;
+            goto one_digit;
+        }
+        for(i = 1;; i++, dval(&u) *= 10.) {
+            L = (Long)(dval(&u) / ds);
+            dval(&u) -= L*ds;
+            *s++ = '0' + (int)L;
+            if (!dval(&u)) {
+                break;
+            }
+            if (i == ilim) {
+                dval(&u) += dval(&u);
+                if (dval(&u) > ds || (dval(&u) == ds && L & 1)) {
+                  bump_up:
+                    while(*--s == '9')
+                        if (s == s0) {
+                            k++;
+                            *s = '0';
+                            break;
+                        }
+                    ++*s++;
+                }
+                break;
+            }
+        }
+        goto ret1;
+    }
+
+    m2 = b2;
+    m5 = b5;
+    if (leftright) {
+        i =
+            denorm ? be + (Bias + (P-1) - 1 + 1) :
+            1 + P - bbits;
+        b2 += i;
+        s2 += i;
+        mhi = i2b(1);
+        if (mhi == NULL)
+            goto failed_malloc;
+    }
+    if (m2 > 0 && s2 > 0) {
+        i = m2 < s2 ? m2 : s2;
+        b2 -= i;
+        m2 -= i;
+        s2 -= i;
+    }
+    if (b5 > 0) {
+        if (leftright) {
+            if (m5 > 0) {
+                mhi = pow5mult(mhi, m5);
+                if (mhi == NULL)
+                    goto failed_malloc;
+                b1 = mult(mhi, b);
+                Bfree(b);
+                b = b1;
+                if (b == NULL)
+                    goto failed_malloc;
+            }
+            if ((j = b5 - m5)) {
+                b = pow5mult(b, j);
+                if (b == NULL)
+                    goto failed_malloc;
+            }
+        }
+        else {
+            b = pow5mult(b, b5);
+            if (b == NULL)
+                goto failed_malloc;
+        }
+    }
+    S = i2b(1);
+    if (S == NULL)
+        goto failed_malloc;
+    if (s5 > 0) {
+        S = pow5mult(S, s5);
+        if (S == NULL)
+            goto failed_malloc;
+    }
+
+    /* Check for special case that d is a normalized power of 2. */
+
+    spec_case = 0;
+    if ((mode < 2 || leftright)
+        ) {
+        if (!word1(&u) && !(word0(&u) & Bndry_mask)
+            && word0(&u) & (Exp_mask & ~Exp_msk1)
+            ) {
+            /* The special case */
+            b2 += Log2P;
+            s2 += Log2P;
+            spec_case = 1;
+        }
+    }
+
+    /* Arrange for convenient computation of quotients:
+     * shift left if necessary so divisor has 4 leading 0 bits.
+     *
+     * Perhaps we should just compute leading 28 bits of S once
+     * and for all and pass them and a shift to quorem, so it
+     * can do shifts and ors to compute the numerator for q.
+     */
+#define iInc 28
+    i = dshift(S, s2);
+    b2 += i;
+    m2 += i;
+    s2 += i;
+    if (b2 > 0) {
+        b = lshift(b, b2);
+        if (b == NULL)
+            goto failed_malloc;
+    }
+    if (s2 > 0) {
+        S = lshift(S, s2);
+        if (S == NULL)
+            goto failed_malloc;
+    }
+    if (k_check) {
+        if (cmp(b,S) < 0) {
+            k--;
+            b = multadd(b, 10, 0);      /* we botched the k estimate */
+            if (b == NULL)
+                goto failed_malloc;
+            if (leftright) {
+                mhi = multadd(mhi, 10, 0);
+                if (mhi == NULL)
+                    goto failed_malloc;
+            }
+            ilim = ilim1;
+        }
+    }
+    if (ilim <= 0 && (mode == 3 || mode == 5)) {
+        if (ilim < 0) {
+            /* no digits, fcvt style */
+          no_digits:
+            k = -1 - ndigits;
+            goto ret;
+        }
+        else {
+            S = multadd(S, 5, 0);
+            if (S == NULL)
+                goto failed_malloc;
+            if (cmp(b, S) <= 0)
+                goto no_digits;
+        }
+      one_digit:
+        *s++ = '1';
+        k++;
+        goto ret;
+    }
+    if (leftright) {
+        if (m2 > 0) {
+            mhi = lshift(mhi, m2);
+            if (mhi == NULL)
+                goto failed_malloc;
+        }
+
+        /* Compute mlo -- check for special case
+         * that d is a normalized power of 2.
+         */
+
+        mlo = mhi;
+        if (spec_case) {
+            mhi = Balloc(mhi->k);
+            if (mhi == NULL)
+                goto failed_malloc;
+            Bcopy(mhi, mlo);
+            mhi = lshift(mhi, Log2P);
+            if (mhi == NULL)
+                goto failed_malloc;
+        }
+
+        for(i = 1;;i++) {
+            dig = quorem(b,S) + '0';
+            /* Do we yet have the shortest decimal string
+             * that will round to d?
+             */
+            j = cmp(b, mlo);
+            delta = diff(S, mhi);
+            if (delta == NULL)
+                goto failed_malloc;
+            j1 = delta->sign ? 1 : cmp(b, delta);
+            Bfree(delta);
+            if (j1 == 0 && mode != 1 && !(word1(&u) & 1)
+                ) {
+                if (dig == '9')
+                    goto round_9_up;
+                if (j > 0)
+                    dig++;
+                *s++ = dig;
+                goto ret;
+            }
+            if (j < 0 || (j == 0 && mode != 1
+                          && !(word1(&u) & 1)
+                    )) {
+                if (!b->x[0] && b->wds <= 1) {
+                    goto accept_dig;
+                }
+                if (j1 > 0) {
+                    b = lshift(b, 1);
+                    if (b == NULL)
+                        goto failed_malloc;
+                    j1 = cmp(b, S);
+                    if ((j1 > 0 || (j1 == 0 && dig & 1))
+                        && dig++ == '9')
+                        goto round_9_up;
+                }
+              accept_dig:
+                *s++ = dig;
+                goto ret;
+            }
+            if (j1 > 0) {
+                if (dig == '9') { /* possible if i == 1 */
+                  round_9_up:
+                    *s++ = '9';
+                    goto roundoff;
+                }
+                *s++ = dig + 1;
+                goto ret;
+            }
+            *s++ = dig;
+            if (i == ilim)
+                break;
+            b = multadd(b, 10, 0);
+            if (b == NULL)
+                goto failed_malloc;
+            if (mlo == mhi) {
+                mlo = mhi = multadd(mhi, 10, 0);
+                if (mlo == NULL)
+                    goto failed_malloc;
+            }
+            else {
+                mlo = multadd(mlo, 10, 0);
+                if (mlo == NULL)
+                    goto failed_malloc;
+                mhi = multadd(mhi, 10, 0);
+                if (mhi == NULL)
+                    goto failed_malloc;
+            }
+        }
+    }
+    else
+        for(i = 1;; i++) {
+            *s++ = dig = quorem(b,S) + '0';
+            if (!b->x[0] && b->wds <= 1) {
+                goto ret;
+            }
+            if (i >= ilim)
+                break;
+            b = multadd(b, 10, 0);
+            if (b == NULL)
+                goto failed_malloc;
+        }
+
+    /* Round off last digit */
+
+    b = lshift(b, 1);
+    if (b == NULL)
+        goto failed_malloc;
+    j = cmp(b, S);
+    if (j > 0 || (j == 0 && dig & 1)) {
+      roundoff:
+        while(*--s == '9')
+            if (s == s0) {
+                k++;
+                *s++ = '1';
+                goto ret;
+            }
+        ++*s++;
+    }
+    else {
+        while(*--s == '0');
+        s++;
+    }
+  ret:
+    Bfree(S);
+    if (mhi) {
+        if (mlo && mlo != mhi)
+            Bfree(mlo);
+        Bfree(mhi);
+    }
+  ret1:
+    Bfree(b);
+    *s = 0;
+    *decpt = k + 1;
+    if (rve)
+        *rve = s;
+    return s0;
+  failed_malloc:
+    if (S)
+        Bfree(S);
+    if (mlo && mlo != mhi)
+        Bfree(mlo);
+    if (mhi)
+        Bfree(mhi);
+    if (b)
+        Bfree(b);
+    if (s0)
+        _Py_dg_freedtoa(s0);
+    return NULL;
+}
+#ifdef __cplusplus
+}
+#endif
+
+#endif  /* PY_NO_SHORT_FLOAT_REPR */

lib_python/2.7_Python/formatter_string.c

+// This file is originally from CPython 2.7, with modifications for Pyston
+
+/***********************************************************************/
+/* Implements the string (as opposed to unicode) version of the
+   built-in formatters for string, int, float.  That is, the versions
+   of int.__format__, etc., that take and return string objects */
+
+#include "Python.h"
+// Pyston change: had to change this path
+#include "../2.7_Objects/stringlib/stringdefs.h"
+
+#define FORMAT_STRING  _PyBytes_FormatAdvanced
+#define FORMAT_LONG    _PyLong_FormatAdvanced
+#define FORMAT_INT     _PyInt_FormatAdvanced
+#define FORMAT_FLOAT   _PyFloat_FormatAdvanced
+#ifndef WITHOUT_COMPLEX
+#define FORMAT_COMPLEX _PyComplex_FormatAdvanced
+#endif
+
+// Pyston change: had to change this path
+#include "../2.7_Objects/stringlib/formatter.h"
+

lib_python/2.7_Python/pystrtod.c

+// This file is originally from CPython 2.7, with modifications for Pyston
+
+/* -*- Mode: C; c-file-style: "python" -*- */
+
+#include <Python.h>
+#include <locale.h>
+
+/* Case-insensitive string match used for nan and inf detection; t should be
+   lower-case.  Returns 1 for a successful match, 0 otherwise. */
+
+static int
+case_insensitive_match(const char *s, const char *t)
+{
+    while(*t && Py_TOLOWER(*s) == *t) {
+        s++;
+        t++;
+    }
+    return *t ? 0 : 1;
+}
+
+/* _Py_parse_inf_or_nan: Attempt to parse a string of the form "nan", "inf" or
+   "infinity", with an optional leading sign of "+" or "-".  On success,
+   return the NaN or Infinity as a double and set *endptr to point just beyond
+   the successfully parsed portion of the string.  On failure, return -1.0 and
+   set *endptr to point to the start of the string. */
+
+double
+_Py_parse_inf_or_nan(const char *p, char **endptr)
+{
+    double retval;
+    const char *s;
+    int negate = 0;
+
+    s = p;
+    if (*s == '-') {
+        negate = 1;
+        s++;
+    }
+    else if (*s == '+') {
+        s++;
+    }
+    if (case_insensitive_match(s, "inf")) {
+        s += 3;
+        if (case_insensitive_match(s, "inity"))
+            s += 5;
+        retval = negate ? -Py_HUGE_VAL : Py_HUGE_VAL;
+    }
+#ifdef Py_NAN
+    else if (case_insensitive_match(s, "nan")) {
+        s += 3;
+        retval = negate ? -Py_NAN : Py_NAN;
+    }
+#endif
+    else {
+        s = p;
+        retval = -1.0;
+    }
+    *endptr = (char *)s;
+    return retval;
+}
+
+/**
+ * PyOS_ascii_strtod:
+ * @nptr:    the string to convert to a numeric value.
+ * @endptr:  if non-%NULL, it returns the character after
+ *           the last character used in the conversion.
+ *
+ * Converts a string to a #gdouble value.
+ * This function behaves like the standard strtod() function
+ * does in the C locale. It does this without actually
+ * changing the current locale, since that would not be
+ * thread-safe.
+ *
+ * This function is typically used when reading configuration
+ * files or other non-user input that should be locale independent.
+ * To handle input from the user you should normally use the
+ * locale-sensitive system strtod() function.
+ *
+ * If the correct value would cause overflow, plus or minus %HUGE_VAL
+ * is returned (according to the sign of the value), and %ERANGE is
+ * stored in %errno. If the correct value would cause underflow,
+ * zero is returned and %ERANGE is stored in %errno.
+ * If memory allocation fails, %ENOMEM is stored in %errno.
+ *
+ * This function resets %errno before calling strtod() so that
+ * you can reliably detect overflow and underflow.
+ *
+ * Return value: the #gdouble value.
+ **/
+
+#ifndef PY_NO_SHORT_FLOAT_REPR
+
+double
+_PyOS_ascii_strtod(const char *nptr, char **endptr)
+{
+    double result;
+    _Py_SET_53BIT_PRECISION_HEADER;
+
+    assert(nptr != NULL);
+    /* Set errno to zero, so that we can distinguish zero results
+       and underflows */
+    errno = 0;
+
+    _Py_SET_53BIT_PRECISION_START;
+    result = _Py_dg_strtod(nptr, endptr);
+    _Py_SET_53BIT_PRECISION_END;
+
+    if (*endptr == nptr)
+        /* string might represent an inf or nan */
+        result = _Py_parse_inf_or_nan(nptr, endptr);
+
+    return result;
+
+}
+
+#else
+
+/*
+   Use system strtod;  since strtod is locale aware, we may
+   have to first fix the decimal separator.
+
+   Note that unlike _Py_dg_strtod, the system strtod may not always give
+   correctly rounded results.
+*/
+
+double
+_PyOS_ascii_strtod(const char *nptr, char **endptr)
+{
+    char *fail_pos;
+    double val = -1.0;
+    struct lconv *locale_data;
+    const char *decimal_point;
+    size_t decimal_point_len;
+    const char *p, *decimal_point_pos;
+    const char *end = NULL; /* Silence gcc */
+    const char *digits_pos = NULL;
+    int negate = 0;
+
+    assert(nptr != NULL);
+
+    fail_pos = NULL;
+
+    locale_data = localeconv();
+    decimal_point = locale_data->decimal_point;
+    decimal_point_len = strlen(decimal_point);
+
+    assert(decimal_point_len != 0);
+
+    decimal_point_pos = NULL;
+
+    /* Parse infinities and nans */
+    val = _Py_parse_inf_or_nan(nptr, endptr);
+    if (*endptr != nptr)
+        return val;
+
+    /* Set errno to zero, so that we can distinguish zero results
+       and underflows */
+    errno = 0;
+
+    /* We process the optional sign manually, then pass the remainder to
+       the system strtod.  This ensures that the result of an underflow
+       has the correct sign. (bug #1725)  */
+    p = nptr;
+    /* Process leading sign, if present */
+    if (*p == '-') {
+        negate = 1;
+        p++;
+    }
+    else if (*p == '+') {
+        p++;
+    }
+
+    /* Some platform strtods accept hex floats; Python shouldn't (at the
+       moment), so we check explicitly for strings starting with '0x'. */
+    if (*p == '0' && (*(p+1) == 'x' || *(p+1) == 'X'))
+        goto invalid_string;
+
+    /* Check that what's left begins with a digit or decimal point */
+    if (!Py_ISDIGIT(*p) && *p != '.')
+        goto invalid_string;
+
+    digits_pos = p;
+    if (decimal_point[0] != '.' ||
+        decimal_point[1] != 0)
+    {
+        /* Look for a '.' in the input; if present, it'll need to be
+           swapped for the current locale's decimal point before we
+           call strtod.  On the other hand, if we find the current
+           locale's decimal point then the input is invalid. */
+        while (Py_ISDIGIT(*p))
+            p++;
+
+        if (*p == '.')
+        {
+            decimal_point_pos = p++;
+
+            /* locate end of number */
+            while (Py_ISDIGIT(*p))
+                p++;
+
+            if (*p == 'e' || *p == 'E')
+                p++;
+            if (*p == '+' || *p == '-')
+                p++;
+            while (Py_ISDIGIT(*p))
+                p++;
+            end = p;
+        }
+        else if (strncmp(p, decimal_point, decimal_point_len) == 0)
+            /* Python bug #1417699 */
+            goto invalid_string;
+        /* For the other cases, we need not convert the decimal
+           point */
+    }
+
+    if (decimal_point_pos) {
+        char *copy, *c;
+        /* Create a copy of the input, with the '.' converted to the
+           locale-specific decimal point */
+        copy = (char *)PyMem_MALLOC(end - digits_pos +
+                                    1 + decimal_point_len);
+        if (copy == NULL) {
+            *endptr = (char *)nptr;
+            errno = ENOMEM;
+            return val;
+        }
+
+        c = copy;
+        memcpy(c, digits_pos, decimal_point_pos - digits_pos);
+        c += decimal_point_pos - digits_pos;
+        memcpy(c, decimal_point, decimal_point_len);
+        c += decimal_point_len;
+        memcpy(c, decimal_point_pos + 1,
+               end - (decimal_point_pos + 1));
+        c += end - (decimal_point_pos + 1);
+        *c = 0;
+
+        val = strtod(copy, &fail_pos);
+
+        if (fail_pos)
+        {
+            if (fail_pos > decimal_point_pos)
+                fail_pos = (char *)digits_pos +
+                    (fail_pos - copy) -
+                    (decimal_point_len - 1);
+            else
+                fail_pos = (char *)digits_pos +
+                    (fail_pos - copy);
+        }
+
+        PyMem_FREE(copy);
+
+    }
+    else {
+        val = strtod(digits_pos, &fail_pos);
+    }
+
+    if (fail_pos == digits_pos)
+        goto invalid_string;
+
+    if (negate && fail_pos != nptr)
+        val = -val;
+    *endptr = fail_pos;
+
+    return val;
+
+  invalid_string:
+    *endptr = (char*)nptr;
+    errno = EINVAL;
+    return -1.0;
+}
+
+#endif
+
+/* PyOS_ascii_strtod is DEPRECATED in Python 2.7 and 3.1 */
+
+double
+PyOS_ascii_strtod(const char *nptr, char **endptr)
+{
+    char *fail_pos;
+    const char *p;
+    double x;
+
+    if (PyErr_WarnEx(PyExc_DeprecationWarning,
+                     "PyOS_ascii_strtod and PyOS_ascii_atof are "
+                     "deprecated.  Use PyOS_string_to_double "
+                     "instead.", 1) < 0)
+        return -1.0;
+
+    /* _PyOS_ascii_strtod already does everything that we want,
+       except that it doesn't parse leading whitespace */
+    p = nptr;
+    while (Py_ISSPACE(*p))
+        p++;
+    x = _PyOS_ascii_strtod(p, &fail_pos);
+    if (fail_pos == p)
+        fail_pos = (char *)nptr;
+    if (endptr)
+        *endptr = (char *)fail_pos;
+    return x;
+}
+
+/* PyOS_ascii_strtod is DEPRECATED in Python 2.7 and 3.1 */
+
+double
+PyOS_ascii_atof(const char *nptr)
+{
+    return PyOS_ascii_strtod(nptr, NULL);
+}
+
+/* PyOS_string_to_double is the recommended replacement for the deprecated
+   PyOS_ascii_strtod and PyOS_ascii_atof functions.  It converts a
+   null-terminated byte string s (interpreted as a string of ASCII characters)
+   to a float.  The string should not have leading or trailing whitespace (in
+   contrast, PyOS_ascii_strtod allows leading whitespace but not trailing
+   whitespace).  The conversion is independent of the current locale.
+
+   If endptr is NULL, try to convert the whole string.  Raise ValueError and
+   return -1.0 if the string is not a valid representation of a floating-point
+   number.
+
+   If endptr is non-NULL, try to convert as much of the string as possible.
+   If no initial segment of the string is the valid representation of a
+   floating-point number then *endptr is set to point to the beginning of the
+   string, -1.0 is returned and again ValueError is raised.
+
+   On overflow (e.g., when trying to convert '1e500' on an IEEE 754 machine),
+   if overflow_exception is NULL then +-Py_HUGE_VAL is returned, and no Python
+   exception is raised.  Otherwise, overflow_exception should point to a
+   a Python exception, this exception will be raised, -1.0 will be returned,
+   and *endptr will point just past the end of the converted value.
+
+   If any other failure occurs (for example lack of memory), -1.0 is returned
+   and the appropriate Python exception will have been set.
+*/
+
+double
+PyOS_string_to_double(const char *s,
+                      char **endptr,
+                      PyObject *overflow_exception)
+{
+    double x, result=-1.0;
+    char *fail_pos;
+
+    errno = 0;
+    PyFPE_START_PROTECT("PyOS_string_to_double", return -1.0)
+    x = _PyOS_ascii_strtod(s, &fail_pos);
+    PyFPE_END_PROTECT(x)
+
+    if (errno == ENOMEM) {
+        PyErr_NoMemory();
+        fail_pos = (char *)s;
+    }
+    else if (!endptr && (fail_pos == s || *fail_pos != '\0'))
+        PyErr_Format(PyExc_ValueError,
+                      "could not convert string to float: "
+                      "%.200s", s);
+    else if (fail_pos == s)
+        PyErr_Format(PyExc_ValueError,
+                      "could not convert string to float: "
+                      "%.200s", s);
+    else if (errno == ERANGE && fabs(x) >= 1.0 && overflow_exception)
+        PyErr_Format(overflow_exception,
+                      "value too large to convert to float: "
+                      "%.200s", s);
+    else
+        result = x;
+
+    if (endptr != NULL)
+        *endptr = fail_pos;
+    return result;
+}
+
+/* Given a string that may have a decimal point in the current
+   locale, change it back to a dot.  Since the string cannot get
+   longer, no need for a maximum buffer size parameter. */
+Py_LOCAL_INLINE(void)
+change_decimal_from_locale_to_dot(char* buffer)
+{
+    struct lconv *locale_data = localeconv();
+    const char *decimal_point = locale_data->decimal_point;
+
+    if (decimal_point[0] != '.' || decimal_point[1] != 0) {
+        size_t decimal_point_len = strlen(decimal_point);
+
+        if (*buffer == '+' || *buffer == '-')
+            buffer++;
+        while (Py_ISDIGIT(*buffer))
+            buffer++;
+        if (strncmp(buffer, decimal_point, decimal_point_len) == 0) {
+            *buffer = '.';
+            buffer++;
+            if (decimal_point_len > 1) {
+                /* buffer needs to get smaller */
+                size_t rest_len = strlen(buffer +
+                                     (decimal_point_len - 1));
+                memmove(buffer,
+                    buffer + (decimal_point_len - 1),
+                    rest_len);
+                buffer[rest_len] = 0;
+            }
+        }
+    }
+}
+
+
+/* From the C99 standard, section 7.19.6:
+The exponent always contains at least two digits, and only as many more digits
+as necessary to represent the exponent.
+*/
+#define MIN_EXPONENT_DIGITS 2
+
+/* Ensure that any exponent, if present, is at least MIN_EXPONENT_DIGITS
+   in length. */
+Py_LOCAL_INLINE(void)
+ensure_minimum_exponent_length(char* buffer, size_t buf_size)
+{
+    char *p = strpbrk(buffer, "eE");
+    if (p && (*(p + 1) == '-' || *(p + 1) == '+')) {
+        char *start = p + 2;
+        int exponent_digit_cnt = 0;
+        int leading_zero_cnt = 0;
+        int in_leading_zeros = 1;
+        int significant_digit_cnt;
+
+        /* Skip over the exponent and the sign. */
+        p += 2;
+
+        /* Find the end of the exponent, keeping track of leading
+           zeros. */
+        while (*p && Py_ISDIGIT(*p)) {
+            if (in_leading_zeros && *p == '0')
+                ++leading_zero_cnt;
+            if (*p != '0')
+                in_leading_zeros = 0;
+            ++p;
+            ++exponent_digit_cnt;
+        }
+
+        significant_digit_cnt = exponent_digit_cnt - leading_zero_cnt;
+        if (exponent_digit_cnt == MIN_EXPONENT_DIGITS) {
+            /* If there are 2 exactly digits, we're done,
+               regardless of what they contain */
+        }
+        else if (exponent_digit_cnt > MIN_EXPONENT_DIGITS) {
+            int extra_zeros_cnt;
+
+            /* There are more than 2 digits in the exponent.  See
+               if we can delete some of the leading zeros */
+            if (significant_digit_cnt < MIN_EXPONENT_DIGITS)
+                significant_digit_cnt = MIN_EXPONENT_DIGITS;
+            extra_zeros_cnt = exponent_digit_cnt -
+                significant_digit_cnt;
+
+            /* Delete extra_zeros_cnt worth of characters from the
+               front of the exponent */
+            assert(extra_zeros_cnt >= 0);
+
+            /* Add one to significant_digit_cnt to copy the
+               trailing 0 byte, thus setting the length */
+            memmove(start,
+                start + extra_zeros_cnt,
+                significant_digit_cnt + 1);
+        }
+        else {
+            /* If there are fewer than 2 digits, add zeros
+               until there are 2, if there's enough room */
+            int zeros = MIN_EXPONENT_DIGITS - exponent_digit_cnt;
+            if (start + zeros + exponent_digit_cnt + 1
+                  < buffer + buf_size) {
+                memmove(start + zeros, start,
+                    exponent_digit_cnt + 1);
+                memset(start, '0', zeros);
+            }
+        }
+    }
+}
+
+/* Remove trailing zeros after the decimal point from a numeric string; also
+   remove the decimal point if all digits following it are zero.  The numeric
+   string must end in '\0', and should not have any leading or trailing
+   whitespace.  Assumes that the decimal point is '.'. */
+Py_LOCAL_INLINE(void)
+remove_trailing_zeros(char *buffer)
+{
+    char *old_fraction_end, *new_fraction_end, *end, *p;
+
+    p = buffer;
+    if (*p == '-' || *p == '+')
+        /* Skip leading sign, if present */
+        ++p;
+    while (Py_ISDIGIT(*p))
+        ++p;
+
+    /* if there's no decimal point there's nothing to do */
+    if (*p++ != '.')
+        return;
+
+    /* scan any digits after the point */
+    while (Py_ISDIGIT(*p))
+        ++p;
+    old_fraction_end = p;
+
+    /* scan up to ending '\0' */
+    while (*p != '\0')
+        p++;
+    /* +1 to make sure that we move the null byte as well */
+    end = p+1;
+
+    /* scan back from fraction_end, looking for removable zeros */
+    p = old_fraction_end;
+    while (*(p-1) == '0')
+        --p;
+    /* and remove point if we've got that far */
+    if (*(p-1) == '.')
+        --p;
+    new_fraction_end = p;
+
+    memmove(new_fraction_end, old_fraction_end, end-old_fraction_end);
+}
+
+/* Ensure that buffer has a decimal point in it.  The decimal point will not
+   be in the current locale, it will always be '.'. Don't add a decimal point
+   if an exponent is present.  Also, convert to exponential notation where
+   adding a '.0' would produce too many significant digits (see issue 5864).
+
+   Returns a pointer to the fixed buffer, or NULL on failure.
+*/
+Py_LOCAL_INLINE(char *)
+ensure_decimal_point(char* buffer, size_t buf_size, int precision)
+{
+    int digit_count, insert_count = 0, convert_to_exp = 0;
+    char *chars_to_insert, *digits_start;
+
+    /* search for the first non-digit character */
+    char *p = buffer;
+    if (*p == '-' || *p == '+')
+        /* Skip leading sign, if present.  I think this could only
+           ever be '-', but it can't hurt to check for both. */
+        ++p;
+    digits_start = p;
+    while (*p && Py_ISDIGIT(*p))
+        ++p;
+    digit_count = Py_SAFE_DOWNCAST(p - digits_start, Py_ssize_t, int);
+
+    if (*p == '.') {
+        if (Py_ISDIGIT(*(p+1))) {
+            /* Nothing to do, we already have a decimal
+               point and a digit after it */
+        }
+        else {
+            /* We have a decimal point, but no following
+               digit.  Insert a zero after the decimal. */
+            /* can't ever get here via PyOS_double_to_string */
+            assert(precision == -1);
+            ++p;
+            chars_to_insert = "0";
+            insert_count = 1;
+        }
+    }
+    else if (!(*p == 'e' || *p == 'E')) {
+        /* Don't add ".0" if we have an exponent. */
+        if (digit_count == precision) {
+            /* issue 5864: don't add a trailing .0 in the case
+               where the '%g'-formatted result already has as many
+               significant digits as were requested.  Switch to
+               exponential notation instead. */
+            convert_to_exp = 1;
+            /* no exponent, no point, and we shouldn't land here
+               for infs and nans, so we must be at the end of the
+               string. */
+            assert(*p == '\0');
+        }
+        else {
+            assert(precision == -1 || digit_count < precision);
+            chars_to_insert = ".0";
+            insert_count = 2;
+        }
+    }
+    if (insert_count) {
+        size_t buf_len = strlen(buffer);
+        if (buf_len + insert_count + 1 >= buf_size) {
+            /* If there is not enough room in the buffer
+               for the additional text, just skip it.  It's
+               not worth generating an error over. */
+        }
+        else {
+            memmove(p + insert_count, p,
+                buffer + strlen(buffer) - p + 1);
+            memcpy(p, chars_to_insert, insert_count);
+        }
+    }
+    if (convert_to_exp) {
+        int written;
+        size_t buf_avail;
+        p = digits_start;
+        /* insert decimal point */
+        assert(digit_count >= 1);
+        memmove(p+2, p+1, digit_count); /* safe, but overwrites nul */
+        p[1] = '.';
+        p += digit_count+1;
+        assert(p <= buf_size+buffer);
+        buf_avail = buf_size+buffer-p;
+        if (buf_avail == 0)
+            return NULL;
+        /* Add exponent.  It's okay to use lower case 'e': we only
+           arrive here as a result of using the empty format code or
+           repr/str builtins and those never want an upper case 'E' */
+        written = PyOS_snprintf(p, buf_avail, "e%+.02d", digit_count-1);
+        if (!(0 <= written &&
+              written < Py_SAFE_DOWNCAST(buf_avail, size_t, int)))
+            /* output truncated, or something else bad happened */
+            return NULL;
+        remove_trailing_zeros(buffer);
+    }
+    return buffer;
+}
+
+/* see FORMATBUFLEN in unicodeobject.c */
+#define FLOAT_FORMATBUFLEN 120
+
+/**
+ * PyOS_ascii_formatd:
+ * @buffer: A buffer to place the resulting string in
+ * @buf_size: The length of the buffer.
+ * @format: The printf()-style format to use for the
+ *          code to use for converting.
+ * @d: The #gdouble to convert
+ *
+ * Converts a #gdouble to a string, using the '.' as
+ * decimal point. To format the number you pass in
+ * a printf()-style format string. Allowed conversion
+ * specifiers are 'e', 'E', 'f', 'F', 'g', 'G', and 'Z'.
+ *
+ * 'Z' is the same as 'g', except it always has a decimal and
+ *     at least one digit after the decimal.
+ *
+ * Return value: The pointer to the buffer with the converted string.
+ * On failure returns NULL but does not set any Python exception.
+ **/
+char *
+_PyOS_ascii_formatd(char       *buffer,
+                   size_t      buf_size,
+                   const char *format,
+                   double      d,
+                   int         precision)
+{
+    char format_char;
+    size_t format_len = strlen(format);
+
+    /* Issue 2264: code 'Z' requires copying the format.  'Z' is 'g', but
+       also with at least one character past the decimal. */
+    char tmp_format[FLOAT_FORMATBUFLEN];
+
+    /* The last character in the format string must be the format char */
+    format_char = format[format_len - 1];
+
+    if (format[0] != '%')
+        return NULL;
+
+    /* I'm not sure why this test is here.  It's ensuring that the format
+       string after the first character doesn't have a single quote, a
+       lowercase l, or a percent. This is the reverse of the commented-out
+       test about 10 lines ago. */
+    if (strpbrk(format + 1, "'l%"))
+        return NULL;
+
+    /* Also curious about this function is that it accepts format strings
+       like "%xg", which are invalid for floats.  In general, the
+       interface to this function is not very good, but changing it is
+       difficult because it's a public API. */
+
+    if (!(format_char == 'e' || format_char == 'E' ||
+          format_char == 'f' || format_char == 'F' ||
+          format_char == 'g' || format_char == 'G' ||
+          format_char == 'Z'))
+        return NULL;
+
+    /* Map 'Z' format_char to 'g', by copying the format string and
+       replacing the final char with a 'g' */
+    if (format_char == 'Z') {
+        if (format_len + 1 >= sizeof(tmp_format)) {
+            /* The format won't fit in our copy.  Error out.  In
+               practice, this will never happen and will be
+               detected by returning NULL */
+            return NULL;
+        }
+        strcpy(tmp_format, format);
+        tmp_format[format_len - 1] = 'g';
+        format = tmp_format;
+    }
+
+
+    /* Have PyOS_snprintf do the hard work */
+    PyOS_snprintf(buffer, buf_size, format, d);
+
+    /* Do various fixups on the return string */
+
+    /* Get the current locale, and find the decimal point string.
+       Convert that string back to a dot. */
+    change_decimal_from_locale_to_dot(buffer);
+
+    /* If an exponent exists, ensure that the exponent is at least
+       MIN_EXPONENT_DIGITS digits, providing the buffer is large enough
+       for the extra zeros.  Also, if there are more than
+       MIN_EXPONENT_DIGITS, remove as many zeros as possible until we get
+       back to MIN_EXPONENT_DIGITS */
+    ensure_minimum_exponent_length(buffer, buf_size);
+
+    /* If format_char is 'Z', make sure we have at least one character
+       after the decimal point (and make sure we have a decimal point);
+       also switch to exponential notation in some edge cases where the
+       extra character would produce more significant digits that we
+       really want. */
+    if (format_char == 'Z')
+        buffer = ensure_decimal_point(buffer, buf_size, precision);
+
+    return buffer;
+}
+
+char *
+PyOS_ascii_formatd(char       *buffer,
+                   size_t      buf_size,
+                   const char *format,
+                   double      d)
+{
+    if (PyErr_WarnEx(PyExc_DeprecationWarning,
+                     "PyOS_ascii_formatd is deprecated, "
+                     "use PyOS_double_to_string instead", 1) < 0)
+        return NULL;
+
+    return _PyOS_ascii_formatd(buffer, buf_size, format, d, -1);
+}
+
+#ifdef PY_NO_SHORT_FLOAT_REPR
+
+/* The fallback code to use if _Py_dg_dtoa is not available. */
+
+PyAPI_FUNC(char *) PyOS_double_to_string(double val,
+                                         char format_code,
+                                         int precision,
+                                         int flags,
+                                         int *type)
+{
+    char format[32];
+    Py_ssize_t bufsize;
+    char *buf;
+    int t, exp;
+    int upper = 0;
+
+    /* Validate format_code, and map upper and lower case */
+    switch (format_code) {
+    case 'e':          /* exponent */
+    case 'f':          /* fixed */
+    case 'g':          /* general */
+        break;
+    case 'E':
+        upper = 1;
+        format_code = 'e';
+        break;
+    case 'F':
+        upper = 1;
+        format_code = 'f';
+        break;
+    case 'G':
+        upper = 1;
+        format_code = 'g';
+        break;
+    case 'r':          /* repr format */
+        /* Supplied precision is unused, must be 0. */
+        if (precision != 0) {
+            PyErr_BadInternalCall();
+            return NULL;
+        }
+        /* The repr() precision (17 significant decimal digits) is the
+           minimal number that is guaranteed to have enough precision
+           so that if the number is read back in the exact same binary
+           value is recreated.  This is true for IEEE floating point
+           by design, and also happens to work for all other modern
+           hardware. */
+        precision = 17;
+        format_code = 'g';
+        break;
+    default:
+        PyErr_BadInternalCall();
+        return NULL;
+    }
+
+    /* Here's a quick-and-dirty calculation to figure out how big a buffer
+       we need.  In general, for a finite float we need:
+
+         1 byte for each digit of the decimal significand, and
+
+         1 for a possible sign
+         1 for a possible decimal point
+         2 for a possible [eE][+-]
+         1 for each digit of the exponent;  if we allow 19 digits
+           total then we're safe up to exponents of 2**63.
+         1 for the trailing nul byte
+
+       This gives a total of 24 + the number of digits in the significand,
+       and the number of digits in the significand is:
+
+         for 'g' format: at most precision, except possibly
+           when precision == 0, when it's 1.
+         for 'e' format: precision+1
+         for 'f' format: precision digits after the point, at least 1
+           before.  To figure out how many digits appear before the point
+           we have to examine the size of the number.  If fabs(val) < 1.0
+           then there will be only one digit before the point.  If
+           fabs(val) >= 1.0, then there are at most
+
+         1+floor(log10(ceiling(fabs(val))))
+
+           digits before the point (where the 'ceiling' allows for the
+           possibility that the rounding rounds the integer part of val
+           up).  A safe upper bound for the above quantity is
+           1+floor(exp/3), where exp is the unique integer such that 0.5
+           <= fabs(val)/2**exp < 1.0.  This exp can be obtained from
+           frexp.
+
+       So we allow room for precision+1 digits for all formats, plus an
+       extra floor(exp/3) digits for 'f' format.
+
+    */
+
+    if (Py_IS_NAN(val) || Py_IS_INFINITY(val))
+        /* 3 for 'inf'/'nan', 1 for sign, 1 for '\0' */
+        bufsize = 5;
+    else {
+        bufsize = 25 + precision;
+        if (format_code == 'f' && fabs(val) >= 1.0) {
+            frexp(val, &exp);
+            bufsize += exp/3;
+        }
+    }
+
+    buf = PyMem_Malloc(bufsize);
+    if (buf == NULL) {
+        PyErr_NoMemory();
+        return NULL;
+    }
+
+    /* Handle nan and inf. */
+    if (Py_IS_NAN(val)) {
+        strcpy(buf, "nan");
+        t = Py_DTST_NAN;
+    } else if (Py_IS_INFINITY(val)) {
+        if (copysign(1., val) == 1.)
+            strcpy(buf, "inf");
+        else
+            strcpy(buf, "-inf");
+        t = Py_DTST_INFINITE;
+    } else {
+        t = Py_DTST_FINITE;
+        if (flags & Py_DTSF_ADD_DOT_0)
+            format_code = 'Z';
+
+        PyOS_snprintf(format, sizeof(format), "%%%s.%i%c",
+                      (flags & Py_DTSF_ALT ? "#" : ""), precision,
+                      format_code);
+        _PyOS_ascii_formatd(buf, bufsize, format, val, precision);
+    }
+
+    /* Add sign when requested.  It's convenient (esp. when formatting
+     complex numbers) to include a sign even for inf and nan. */
+    if (flags & Py_DTSF_SIGN && buf[0] != '-') {
+        size_t len = strlen(buf);
+        /* the bufsize calculations above should ensure that we've got
+           space to add a sign */
+        assert((size_t)bufsize >= len+2);
+        memmove(buf+1, buf, len+1);
+        buf[0] = '+';
+    }
+    if (upper) {
+        /* Convert to upper case. */
+        char *p1;
+        for (p1 = buf; *p1; p1++)
+            *p1 = Py_TOUPPER(*p1);
+    }
+
+    if (type)
+        *type = t;
+    return buf;
+}
+
+#else
+
+/* _Py_dg_dtoa is available. */
+
+/* I'm using a lookup table here so that I don't have to invent a non-locale
+   specific way to convert to uppercase */
+#define OFS_INF 0
+#define OFS_NAN 1
+#define OFS_E 2
+
+/* The lengths of these are known to the code below, so don't change them */
+static char *lc_float_strings[] = {
+    "inf",
+    "nan",
+    "e",
+};
+static char *uc_float_strings[] = {
+    "INF",
+    "NAN",
+    "E",
+};
+
+
+/* Convert a double d to a string, and return a PyMem_Malloc'd block of
+   memory contain the resulting string.
+
+   Arguments:
+     d is the double to be converted
+     format_code is one of 'e', 'f', 'g', 'r'.  'e', 'f' and 'g'
+       correspond to '%e', '%f' and '%g';  'r' corresponds to repr.
+     mode is one of '0', '2' or '3', and is completely determined by
+       format_code: 'e' and 'g' use mode 2; 'f' mode 3, 'r' mode 0.
+     precision is the desired precision
+     always_add_sign is nonzero if a '+' sign should be included for positive
+       numbers
+     add_dot_0_if_integer is nonzero if integers in non-exponential form
+       should have ".0" added.  Only applies to format codes 'r' and 'g'.
+     use_alt_formatting is nonzero if alternative formatting should be
+       used.  Only applies to format codes 'e', 'f' and 'g'.  For code 'g',
+       at most one of use_alt_formatting and add_dot_0_if_integer should
+       be nonzero.
+     type, if non-NULL, will be set to one of these constants to identify
+       the type of the 'd' argument:
+     Py_DTST_FINITE
+     Py_DTST_INFINITE
+     Py_DTST_NAN
+
+   Returns a PyMem_Malloc'd block of memory containing the resulting string,
+    or NULL on error. If NULL is returned, the Python error has been set.
+ */
+
+static char *
+format_float_short(double d, char format_code,
+                   int mode, Py_ssize_t precision,
+                   int always_add_sign, int add_dot_0_if_integer,
+                   int use_alt_formatting, char **float_strings, int *type)
+{
+    char *buf = NULL;
+    char *p = NULL;
+    Py_ssize_t bufsize = 0;
+    char *digits, *digits_end;
+    int decpt_as_int, sign, exp_len, exp = 0, use_exp = 0;
+    Py_ssize_t decpt, digits_len, vdigits_start, vdigits_end;
+    _Py_SET_53BIT_PRECISION_HEADER;
+
+    /* _Py_dg_dtoa returns a digit string (no decimal point or exponent).
+       Must be matched by a call to _Py_dg_freedtoa. */
+    _Py_SET_53BIT_PRECISION_START;
+    digits = _Py_dg_dtoa(d, mode, precision, &decpt_as_int, &sign,
+                         &digits_end);
+    _Py_SET_53BIT_PRECISION_END;
+
+    decpt = (Py_ssize_t)decpt_as_int;
+    if (digits == NULL) {
+        /* The only failure mode is no memory. */
+        PyErr_NoMemory();
+        goto exit;
+    }
+    assert(digits_end != NULL && digits_end >= digits);
+    digits_len = digits_end - digits;
+
+    if (digits_len && !Py_ISDIGIT(digits[0])) {
+        /* Infinities and nans here; adapt Gay's output,
+           so convert Infinity to inf and NaN to nan, and
+           ignore sign of nan. Then return. */
+
+        /* ignore the actual sign of a nan */
+        if (digits[0] == 'n' || digits[0] == 'N')
+            sign = 0;
+
+        /* We only need 5 bytes to hold the result "+inf\0" . */
+        bufsize = 5; /* Used later in an assert. */
+        buf = (char *)PyMem_Malloc(bufsize);
+        if (buf == NULL) {
+            PyErr_NoMemory();
+            goto exit;
+        }
+        p = buf;
+
+        if (sign == 1) {
+            *p++ = '-';
+        }
+        else if (always_add_sign) {
+            *p++ = '+';
+        }
+        if (digits[0] == 'i' || digits[0] == 'I') {
+            strncpy(p, float_strings[OFS_INF], 3);
+            p += 3;
+
+            if (type)
+                *type = Py_DTST_INFINITE;
+        }
+        else if (digits[0] == 'n' || digits[0] == 'N') {
+            strncpy(p, float_strings[OFS_NAN], 3);
+            p += 3;
+
+            if (type)
+                *type = Py_DTST_NAN;
+        }
+        else {
+            /* shouldn't get here: Gay's code should always return
+               something starting with a digit, an 'I',  or 'N' */
+            strncpy(p, "ERR", 3);
+            p += 3;
+            assert(0);
+        }
+        goto exit;
+    }
+
+    /* The result must be finite (not inf or nan). */
+    if (type)
+        *type = Py_DTST_FINITE;
+
+
+    /* We got digits back, format them.  We may need to pad 'digits'
+       either on the left or right (or both) with extra zeros, so in
+       general the resulting string has the form
+
+         [<sign>]<zeros><digits><zeros>[<exponent>]
+
+       where either of the <zeros> pieces could be empty, and there's a
+       decimal point that could appear either in <digits> or in the
+       leading or trailing <zeros>.
+
+       Imagine an infinite 'virtual' string vdigits, consisting of the
+       string 'digits' (starting at index 0) padded on both the left and
+       right with infinite strings of zeros.  We want to output a slice
+
+         vdigits[vdigits_start : vdigits_end]
+
+       of this virtual string.  Thus if vdigits_start < 0 then we'll end
+       up producing some leading zeros; if vdigits_end > digits_len there
+       will be trailing zeros in the output.  The next section of code
+       determines whether to use an exponent or not, figures out the
+       position 'decpt' of the decimal point, and computes 'vdigits_start'
+       and 'vdigits_end'. */
+    vdigits_end = digits_len;
+    switch (format_code) {
+    case 'e':
+        use_exp = 1;
+        vdigits_end = precision;
+        break;
+    case 'f':
+        vdigits_end = decpt + precision;
+        break;
+    case 'g':
+        if (decpt <= -4 || decpt >
+            (add_dot_0_if_integer ? precision-1 : precision))
+            use_exp = 1;
+        if (use_alt_formatting)
+            vdigits_end = precision;
+        break;
+    case 'r':
+        /* convert to exponential format at 1e16.  We used to convert
+           at 1e17, but that gives odd-looking results for some values
+           when a 16-digit 'shortest' repr is padded with bogus zeros.
+           For example, repr(2e16+8) would give 20000000000000010.0;
+           the true value is 20000000000000008.0. */
+        if (decpt <= -4 || decpt > 16)
+            use_exp = 1;
+        break;
+    default:
+        PyErr_BadInternalCall();
+        goto exit;
+    }
+
+    /* if using an exponent, reset decimal point position to 1 and adjust
+       exponent accordingly.*/
+    if (use_exp) {
+        exp = decpt - 1;
+        decpt = 1;
+    }
+    /* ensure vdigits_start < decpt <= vdigits_end, or vdigits_start <
+       decpt < vdigits_end if add_dot_0_if_integer and no exponent */
+    vdigits_start = decpt <= 0 ? decpt-1 : 0;
+    if (!use_exp && add_dot_0_if_integer)
+        vdigits_end = vdigits_end > decpt ? vdigits_end : decpt + 1;
+    else
+        vdigits_end = vdigits_end > decpt ? vdigits_end : decpt;
+
+    /* double check inequalities */
+    assert(vdigits_start <= 0 &&
+           0 <= digits_len &&
+           digits_len <= vdigits_end);
+    /* decimal point should be in (vdigits_start, vdigits_end] */
+    assert(vdigits_start < decpt && decpt <= vdigits_end);
+
+    /* Compute an upper bound how much memory we need. This might be a few
+       chars too long, but no big deal. */
+    bufsize =
+        /* sign, decimal point and trailing 0 byte */
+        3 +
+
+        /* total digit count (including zero padding on both sides) */
+        (vdigits_end - vdigits_start) +
+
+        /* exponent "e+100", max 3 numerical digits */
+        (use_exp ? 5 : 0);
+
+    /* Now allocate the memory and initialize p to point to the start of
+       it. */
+    buf = (char *)PyMem_Malloc(bufsize);
+    if (buf == NULL) {
+        PyErr_NoMemory();
+        goto exit;
+    }
+    p = buf;
+
+    /* Add a negative sign if negative, and a plus sign if non-negative
+       and always_add_sign is true. */
+    if (sign == 1)
+        *p++ = '-';
+    else if (always_add_sign)
+        *p++ = '+';
+
+    /* note that exactly one of the three 'if' conditions is true,
+       so we include exactly one decimal point */
+    /* Zero padding on left of digit string */
+    if (decpt <= 0) {
+        memset(p, '0', decpt-vdigits_start);
+        p += decpt - vdigits_start;
+        *p++ = '.';
+        memset(p, '0', 0-decpt);
+        p += 0-decpt;
+    }
+    else {
+        memset(p, '0', 0-vdigits_start);
+        p += 0 - vdigits_start;
+    }
+
+    /* Digits, with included decimal point */
+    if (0 < decpt && decpt <= digits_len) {
+        strncpy(p, digits, decpt-0);
+        p += decpt-0;
+        *p++ = '.';
+        strncpy(p, digits+decpt, digits_len-decpt);
+        p += digits_len-decpt;
+    }
+    else {
+        strncpy(p, digits, digits_len);
+        p += digits_len;
+    }
+
+    /* And zeros on the right */
+    if (digits_len < decpt) {
+        memset(p, '0', decpt-digits_len);
+        p += decpt-digits_len;
+        *p++ = '.';
+        memset(p, '0', vdigits_end-decpt);
+        p += vdigits_end-decpt;
+    }
+    else {
+        memset(p, '0', vdigits_end-digits_len);
+        p += vdigits_end-digits_len;
+    }
+
+    /* Delete a trailing decimal pt unless using alternative formatting. */
+    if (p[-1] == '.' && !use_alt_formatting)
+        p--;
+
+    /* Now that we've done zero padding, add an exponent if needed. */
+    if (use_exp) {
+        *p++ = float_strings[OFS_E][0];
+        exp_len = sprintf(p, "%+.02d", exp);
+        p += exp_len;
+    }
+  exit:
+    if (buf) {
+        *p = '\0';
+        /* It's too late if this fails, as we've already stepped on
+           memory that isn't ours. But it's an okay debugging test. */
+        assert(p-buf < bufsize);
+    }
+    if (digits)
+        _Py_dg_freedtoa(digits);
+
+    return buf;
+}
+
+
+PyAPI_FUNC(char *) PyOS_double_to_string(double val,
+                                         char format_code,
+                                         int precision,
+                                         int flags,
+                                         int *type)
+{
+    char **float_strings = lc_float_strings;
+    int mode;
+
+    /* Validate format_code, and map upper and lower case. Compute the
+       mode and make any adjustments as needed. */
+    switch (format_code) {
+    /* exponent */
+    case 'E':
+        float_strings = uc_float_strings;
+        format_code = 'e';
+        /* Fall through. */
+    case 'e':
+        mode = 2;
+        precision++;
+        break;
+
+    /* fixed */
+    case 'F':
+        float_strings = uc_float_strings;
+        format_code = 'f';
+        /* Fall through. */
+    case 'f':
+        mode = 3;
+        break;
+
+    /* general */
+    case 'G':
+        float_strings = uc_float_strings;
+        format_code = 'g';
+        /* Fall through. */
+    case 'g':
+        mode = 2;
+        /* precision 0 makes no sense for 'g' format; interpret as 1 */
+        if (precision == 0)
+            precision = 1;
+        break;
+
+    /* repr format */
+    case 'r':
+        mode = 0;
+        /* Supplied precision is unused, must be 0. */
+        if (precision != 0) {
+            PyErr_BadInternalCall();
+            return NULL;
+        }
+        break;
+
+    default:
+        PyErr_BadInternalCall();
+        return NULL;
+    }
+
+    return format_float_short(val, format_code, mode, precision,
+                              flags & Py_DTSF_SIGN,
+                              flags & Py_DTSF_ADD_DOT_0,
+                              flags & Py_DTSF_ALT,
+                              float_strings, type);
+}
+#endif /* ifdef PY_NO_SHORT_FLOAT_REPR */

lib_python/CMakeLists.txt

@@ -18,10 +18,10 @@ add_custom_target(copy_stdlib ALL DEPENDS ${STDLIB_TARGETS})
 file(GLOB_RECURSE STDMODULE_SRCS 2.7_Modules errnomodule.c shamodule.c sha256module.c sha512module.c _math.c mathmodule.c md5.c md5module.c _randommodule.c _sre.c operator.c binascii.c pwdmodule.c posixmodule.c)
 
 # compile specified files in lib_python/2.7_Objects
-file(GLOB_RECURSE STDOBJECT_SRCS 2.7_Objects structseq.c capsule.c)
+file(GLOB_RECURSE STDOBJECT_SRCS 2.7_Objects structseq.c capsule.c stringobject.c)
 
 # compile specified files in lib_python/2.7_Python
-file(GLOB_RECURSE STDPYTHON_SRCS 2.7_Python getargs.c pyctype.c)
+file(GLOB_RECURSE STDPYTHON_SRCS 2.7_Python getargs.c pyctype.c formatter_string.c pystrtod.c dtoa.c)
 
 set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -Wno-missing-field-initializers -Wno-tautological-compare -Wno-type-limits")
 add_library(FROM_CPYTHON OBJECT ${STDMODULE_SRCS} ${STDOBJECT_SRCS} ${STDPYTHON_SRCS})

src/capi/abstract.cpp

+// Copyright (c) 2014 Dropbox, Inc.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include <stdarg.h>
+#include <string.h>
+
+#include "Python.h"
+
+#include "capi/types.h"
+#include "core/threading.h"
+#include "core/types.h"
+#include "runtime/import.h"
+#include "runtime/objmodel.h"
+#include "runtime/types.h"
+
+namespace pyston {
+
+extern "C" PyObject* PyObject_CallFunctionObjArgs(PyObject* callable, ...) {
+    Py_FatalError("unimplemented");
+}
+}

src/capi/object.cpp

+// Copyright (c) 2014 Dropbox, Inc.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include <stdarg.h>
+#include <string.h>
+
+#include "Python.h"
+
+#include "capi/types.h"
+#include "core/threading.h"
+#include "core/types.h"
+#include "runtime/import.h"
+#include "runtime/objmodel.h"
+#include "runtime/types.h"
+
+namespace pyston {
+
+extern "C" PyObject* PyObject_Unicode(PyObject* v) {
+    Py_FatalError("unimplemented");
+}
+
+extern "C" PyObject* _PyObject_Str(PyObject* v) {
+    if (v == NULL)
+        return boxStrConstant("<NULL>");
+
+    if (v->cls == str_cls)
+        return v;
+
+    try {
+        return str(v);
+    } catch (Box* b) {
+        PyErr_SetObject(b->cls, b);
+        return NULL;
+    }
+}
+
+extern "C" PyObject* PyObject_Str(PyObject* v) {
+    PyObject* res = _PyObject_Str(v);
+    if (res == NULL)
+        return NULL;
+#ifdef Py_USING_UNICODE
+    if (PyUnicode_Check(res)) {
+        PyObject* str;
+        str = PyUnicode_AsEncodedString(res, NULL, NULL);
+        Py_DECREF(res);
+        if (str)
+            res = str;
+        else
+            return NULL;
+    }
+#endif
+    assert(PyString_Check(res));
+    return res;
+}
+}

src/capi/typeobject.cpp

@@ -289,6 +289,11 @@ static PyObject* lookup_maybe(PyObject* self, const char* attrstr, PyObject** at
     return obj;
 }
 
+extern "C" PyObject* _PyObject_LookupSpecial(PyObject* self, const char* attrstr, PyObject** attrobj) {
+    assert(!PyInstance_Check(self));
+    return lookup_maybe(self, attrstr, attrobj);
+}
+
 static PyObject* lookup_method(PyObject* self, const char* attrstr, PyObject** attrobj) noexcept {
     PyObject* res = lookup_maybe(self, attrstr, attrobj);
     if (res == NULL && !PyErr_Occurred())

src/codegen/compvars.cpp

@@ -1366,9 +1366,8 @@ public:
             return NULL;
         BoxedFunction* rtattr_func = static_cast<BoxedFunction*>(rtattr);
 
-        RELEASE_ASSERT(!argspec.has_starargs, "");
-        RELEASE_ASSERT(!argspec.has_kwargs, "");
-        RELEASE_ASSERT(argspec.num_keywords == 0, "");
+        if (argspec.num_keywords || argspec.has_starargs || argspec.has_kwargs)
+            return NULL;
 
         CLFunction* cl = rtattr_func->f;
         assert(cl);

src/runtime/builtin_modules/builtins.cpp

@@ -492,7 +492,7 @@ BoxedClass* BaseException, *Exception, *StandardError, *AssertionError, *Attribu
     *NameError, *KeyError, *IndexError, *IOError, *OSError, *ZeroDivisionError, *ValueError, *UnboundLocalError,
     *RuntimeError, *ImportError, *StopIteration, *Warning, *SyntaxError, *OverflowError, *DeprecationWarning,
     *MemoryError, *LookupError, *EnvironmentError, *ArithmeticError, *BufferError, *KeyboardInterrupt, *SystemExit,
-    *SystemError, *NotImplementedError;
+    *SystemError, *NotImplementedError, *PendingDeprecationWarning;
 }
 
 Box* exceptionNew1(BoxedClass* cls) {
@@ -815,7 +815,6 @@ void setupBuiltins() {
     SyntaxError = makeBuiltinException(StandardError, "SyntaxError");
     OverflowError = makeBuiltinException(ArithmeticError, "OverflowError");
     /*ImportWarning =*/makeBuiltinException(Warning, "ImportWarning");
-    /*PendingDeprecationWarning =*/makeBuiltinException(Warning, "PendingDeprecationWarning");
     DeprecationWarning = makeBuiltinException(Warning, "DeprecationWarning");
     /*BytesWarning =*/makeBuiltinException(Warning, "BytesWarning");
     MemoryError = makeBuiltinException(StandardError, "MemoryError");
@@ -824,6 +823,7 @@ void setupBuiltins() {
     SystemExit = makeBuiltinException(BaseException, "SystemExit");
     SystemError = makeBuiltinException(StandardError, "SystemError");
     NotImplementedError = makeBuiltinException(RuntimeError, "NotImplementedError");
+    PendingDeprecationWarning = makeBuiltinException(Warning, "PendingDeprecationWarning");
 
     EnvironmentError->giveAttr(
         "__init__",

src/runtime/capi.cpp

@@ -262,6 +262,152 @@ extern "C" PyObject* PyObject_Repr(PyObject* obj) {
     }
 }
 
+extern "C" PyObject* PyObject_Format(PyObject* obj, PyObject* format_spec) {
+    PyObject* empty = NULL;
+    PyObject* result = NULL;
+#ifdef Py_USING_UNICODE
+    int spec_is_unicode;
+    int result_is_unicode;
+#endif
+
+    /* If no format_spec is provided, use an empty string */
+    if (format_spec == NULL) {
+        empty = PyString_FromStringAndSize(NULL, 0);
+        format_spec = empty;
+    }
+
+/* Check the format_spec type, and make sure it's str or unicode */
+#ifdef Py_USING_UNICODE
+    if (PyUnicode_Check(format_spec))
+        spec_is_unicode = 1;
+    else if (PyString_Check(format_spec))
+        spec_is_unicode = 0;
+    else {
+#else
+    if (!PyString_Check(format_spec)) {
+#endif
+        PyErr_Format(PyExc_TypeError, "format expects arg 2 to be string "
+                                      "or unicode, not %.100s",
+                     Py_TYPE(format_spec)->tp_name);
+        goto done;
+    }
+
+    /* Check for a __format__ method and call it. */
+    if (PyInstance_Check(obj)) {
+        /* We're an instance of a classic class */
+        PyObject* bound_method = PyObject_GetAttrString(obj, "__format__");
+        if (bound_method != NULL) {
+            result = PyObject_CallFunctionObjArgs(bound_method, format_spec, NULL);
+            Py_DECREF(bound_method);
+        } else {
+            PyObject* self_as_str = NULL;
+            PyObject* format_method = NULL;
+            Py_ssize_t format_len;
+
+            PyErr_Clear();
+/* Per the PEP, convert to str (or unicode,
+   depending on the type of the format
+   specifier).  For new-style classes, this
+   logic is done by object.__format__(). */
+#ifdef Py_USING_UNICODE
+            if (spec_is_unicode) {
+                format_len = PyUnicode_GET_SIZE(format_spec);
+                self_as_str = PyObject_Unicode(obj);
+            } else
+#endif
+            {
+                format_len = PyString_GET_SIZE(format_spec);
+                self_as_str = PyObject_Str(obj);
+            }
+            if (self_as_str == NULL)
+                goto done1;
+
+            if (format_len > 0) {
+                /* See the almost identical code in
+                   typeobject.c for new-style
+                   classes. */
+                if (PyErr_WarnEx(PyExc_PendingDeprecationWarning, "object.__format__ with a non-empty "
+                                                                  "format string is deprecated",
+                                 1) < 0) {
+                    goto done1;
+                }
+                /* Eventually this will become an
+                   error:
+                PyErr_Format(PyExc_TypeError,
+                   "non-empty format string passed to "
+                   "object.__format__");
+                goto done1;
+                */
+            }
+
+            /* Then call str.__format__ on that result */
+            format_method = PyObject_GetAttrString(self_as_str, "__format__");
+            if (format_method == NULL) {
+                goto done1;
+            }
+            result = PyObject_CallFunctionObjArgs(format_method, format_spec, NULL);
+        done1:
+            Py_XDECREF(self_as_str);
+            Py_XDECREF(format_method);
+            if (result == NULL)
+                goto done;
+        }
+    } else {
+        /* Not an instance of a classic class, use the code
+           from py3k */
+        static PyObject* format_cache = NULL;
+
+        /* Find the (unbound!) __format__ method (a borrowed
+           reference) */
+        PyObject* method = _PyObject_LookupSpecial(obj, "__format__", &format_cache);
+        if (method == NULL) {
+            if (!PyErr_Occurred())
+                PyErr_Format(PyExc_TypeError, "Type %.100s doesn't define __format__", Py_TYPE(obj)->tp_name);
+            goto done;
+        }
+        /* And call it. */
+        result = PyObject_CallFunctionObjArgs(method, format_spec, NULL);
+        Py_DECREF(method);
+    }
+
+    if (result == NULL)
+        goto done;
+
+/* Check the result type, and make sure it's str or unicode */
+#ifdef Py_USING_UNICODE
+    if (PyUnicode_Check(result))
+        result_is_unicode = 1;
+    else if (PyString_Check(result))
+        result_is_unicode = 0;
+    else {
+#else
+    if (!PyString_Check(result)) {
+#endif
+        PyErr_Format(PyExc_TypeError, "%.100s.__format__ must return string or "
+                                      "unicode, not %.100s",
+                     Py_TYPE(obj)->tp_name, Py_TYPE(result)->tp_name);
+        Py_DECREF(result);
+        result = NULL;
+        goto done;
+    }
+
+/* Convert to unicode, if needed.  Required if spec is unicode
+   and result is str */
+#ifdef Py_USING_UNICODE
+    if (spec_is_unicode && !result_is_unicode) {
+        PyObject* tmp = PyObject_Unicode(result);
+        /* This logic works whether or not tmp is NULL */
+        Py_DECREF(result);
+        result = tmp;
+    }
+#endif
+
+done:
+    Py_XDECREF(empty);
+    return result;
+}
+
+
 extern "C" PyObject* PyObject_GetAttr(PyObject* o, PyObject* attr_name) {
     if (!isSubclass(attr_name->cls, str_cls)) {
         PyErr_Format(PyExc_TypeError, "attribute name must be string, not '%.200s'", Py_TYPE(attr_name)->tp_name);

src/runtime/classobj.cpp

@@ -23,7 +23,9 @@
 
 namespace pyston {
 
+extern "C" {
 BoxedClass* classobj_cls, *instance_cls;
+}
 
 bool classobjIssubclass(BoxedClassobj* child, BoxedClassobj* parent) {
     if (child == parent)

src/runtime/classobj.h

@@ -24,7 +24,9 @@ void setupClassobj();
 class BoxedClass;
 class BoxedClassobj;
 class BoxedInstance;
+extern "C" {
 extern BoxedClass* classobj_cls, *instance_cls;
+}
 
 bool instanceIsinstance(BoxedInstance* obj, BoxedClassobj* cls);
 bool classobjIssubclass(BoxedClassobj* child, BoxedClassobj* parent);

src/runtime/complex.cpp

@@ -34,6 +34,22 @@ extern "C" Py_complex PyComplex_AsCComplex(PyObject* op) {
     Py_FatalError("unimplemented");
 }
 
+extern "C" double PyComplex_RealAsDouble(PyObject* op) {
+    if (PyComplex_Check(op)) {
+        return static_cast<BoxedComplex*>(op)->real;
+    } else {
+        return PyFloat_AsDouble(op);
+    }
+}
+
+extern "C" double PyComplex_ImagAsDouble(PyObject* op) {
+    if (PyComplex_Check(op)) {
+        return static_cast<BoxedComplex*>(op)->imag;
+    } else {
+        return 0.0;
+    }
+}
+
 // addition
 
 extern "C" Box* complexAddComplex(BoxedComplex* lhs, BoxedComplex* rhs) {

src/runtime/int.cpp

@@ -58,6 +58,10 @@ extern "C" PyObject* PyInt_FromLong(long n) {
     return boxInt(n);
 }
 
+extern "C" PyAPI_FUNC(PyObject*) _PyInt_Format(PyIntObject* v, int base, int newstyle) {
+    Py_FatalError("unimplemented");
+}
+
 BoxedInt* interned_ints[NUM_INTERNED_INTS];
 
 // If we don't have fast overflow-checking builtins, provide some slow variants:

src/runtime/long.cpp

@@ -113,6 +113,10 @@ extern "C" double PyLong_AsDouble(PyObject* vv) {
     return mpz_get_d(l->n);
 }
 
+extern "C" PyAPI_FUNC(PyObject*) _PyLong_Format(PyObject* aa, int base, int addL, int newstyle) {
+    Py_FatalError("unimplemented");
+}
+
 extern "C" PyObject* PyLong_FromDouble(double v) {
     Py_FatalError("unimplemented");
 }
@@ -129,6 +133,23 @@ extern "C" PyObject* PyLong_FromUnsignedLong(unsigned long ival) {
     return rtn;
 }
 
+#define IS_LITTLE_ENDIAN (int)*(unsigned char*)&one
+#define PY_ABS_LLONG_MIN (0 - (unsigned PY_LONG_LONG)PY_LLONG_MIN)
+
+extern "C" PyObject* PyLong_FromSsize_t(Py_ssize_t ival) {
+    Py_ssize_t bytes = ival;
+    int one = 1;
+    return _PyLong_FromByteArray((unsigned char*)&bytes, SIZEOF_SIZE_T, IS_LITTLE_ENDIAN, 1);
+}
+
+extern "C" PyObject* PyLong_FromSize_t(size_t ival) {
+    size_t bytes = ival;
+    int one = 1;
+    return _PyLong_FromByteArray((unsigned char*)&bytes, SIZEOF_SIZE_T, IS_LITTLE_ENDIAN, 0);
+}
+
+#undef IS_LITTLE_ENDIAN
+
 extern "C" double _PyLong_Frexp(PyLongObject* a, Py_ssize_t* e) {
     Py_FatalError("unimplemented");
 }

src/runtime/str.cpp

@@ -24,6 +24,7 @@
 #include "core/types.h"
 #include "core/util.h"
 #include "gc/collector.h"
+#include "runtime/capi.h"
 #include "runtime/dict.h"
 #include "runtime/objmodel.h"
 #include "runtime/types.h"
@@ -650,6 +651,18 @@ Box* strPartition(BoxedString* self, BoxedString* sep) {
                                                self->s.size() - found_idx - sep->s.size()) });
 }
 
+extern "C" PyObject* do_string_format(PyObject* self, PyObject* args, PyObject* kwargs);
+
+Box* strFormat(BoxedString* self, BoxedTuple* args, BoxedDict* kwargs) {
+    assert(args->cls == tuple_cls);
+    assert(kwargs->cls == dict_cls);
+
+    Box* rtn = do_string_format(self, args, kwargs);
+    checkAndThrowCAPIException();
+    assert(rtn);
+    return rtn;
+}
+
 Box* strSplit(BoxedString* self, BoxedString* sep, BoxedInt* _max_split) {
     assert(self->cls == str_cls);
     if (_max_split->cls != int_cls)
@@ -1039,7 +1052,13 @@ extern "C" Py_ssize_t PyString_Size(PyObject* s) {
 }
 
 extern "C" int _PyString_Resize(PyObject** pv, Py_ssize_t newsize) {
-    Py_FatalError("unimplemented");
+    // This is only allowed to be called when there is only one user of the string (ie a refcount of 1 in CPython)
+
+    assert(pv);
+    assert((*pv)->cls == str_cls);
+    BoxedString* s = static_cast<BoxedString*>(*pv);
+    s->s.resize(newsize, '\0');
+    return 0;
 }
 
 static Py_ssize_t string_buffer_getreadbuf(PyObject* self, Py_ssize_t index, const void** ptr) {
@@ -1120,6 +1139,8 @@ void setupStr() {
 
     str_cls->giveAttr("partition", new BoxedFunction(boxRTFunction((void*)strPartition, UNKNOWN, 2)));
 
+    str_cls->giveAttr("format", new BoxedFunction(boxRTFunction((void*)strFormat, UNKNOWN, 1, 0, true, true)));
+
     str_cls->giveAttr("__add__", new BoxedFunction(boxRTFunction((void*)strAdd, UNKNOWN, 2)));
     str_cls->giveAttr("__mod__", new BoxedFunction(boxRTFunction((void*)strMod, STR, 2)));
     str_cls->giveAttr("__mul__", new BoxedFunction(boxRTFunction((void*)strMul, UNKNOWN, 2)));

test/tests/str_functions.py

@@ -78,3 +78,5 @@ print "hello world".partition("hello")
 print "hello world".partition("o")
 
 print "hello world"[False:True:True]
+
+print "{hello}".format(hello="world")