Issue #1580: New free format floating point representation based on...

Issue #1580: New free format floating point representation based on "Floating-Point Printer Sample Code", by Robert G. Burger. For example repr(11./5) now returns '2.2' instead of '2.2000000000000002'.

Thanks to noam for the patch! I had to modify doubledigits.c slightly to support X64 and IA64 machines on Windows. I also added the new file to the three project files.

Include/floatobject.h

@@ -76,6 +76,10 @@ PyAPI_FUNC(int) _PyFloat_Pack8(double x, unsigned char *p, int le);
  */
 PyAPI_FUNC(int) _PyFloat_Repr(double x, char *p, size_t len);
 
+/* Used to get the important decimal digits of a double */
+PyAPI_FUNC(int) _PyFloat_Digits(char *buf, double v, int *signum);
+PyAPI_FUNC(void) _PyFloat_DigitsInit(void);
+
 /* The unpack routines read 4 or 8 bytes, starting at p.  le is a bool
  * argument, true if the string is in little-endian format (exponent
  * last, at p+3 or p+7), false if big-endian (exponent first, at p).

Lib/test/test_float.py

 
 import unittest, struct
+import os
 from test import test_support
 
 class FormatFunctionsTestCase(unittest.TestCase):
@@ -146,12 +147,26 @@ class FormatTestCase(unittest.TestCase):
         self.assertRaises(ValueError, format, 3.0, "s")
 
 
+class ReprTestCase(unittest.TestCase):
+    def test_repr(self):
+        floats_file = open(os.path.join(os.path.split(__file__)[0],
+                           'floating_points.txt'))
+        for line in floats_file:
+            line = line.strip()
+            if not line or line.startswith('#'):
+                continue
+            v = eval(line)
+            self.assertEqual(v, eval(repr(v)))
+        floats_file.close()
+
+
 def test_main():
     test_support.run_unittest(
         FormatFunctionsTestCase,
         UnknownFormatTestCase,
         IEEEFormatTestCase,
-        FormatTestCase)
+        FormatTestCase,
+        ReprTestCase)
 
 if __name__ == '__main__':
     test_main()

Makefile.pre.in

@@ -301,6 +301,7 @@ OBJECT_OBJS=	\
 		Objects/genobject.o \
 		Objects/fileobject.o \
 		Objects/floatobject.o \
+		Objects/doubledigits.o \
 		Objects/frameobject.o \
 		Objects/funcobject.o \
 		Objects/iterobject.o \

Misc/NEWS

@@ -12,6 +12,10 @@ What's New in Python 3.0a3?
 Core and Builtins
 -----------------
 
+- Issue #1580: New free format floating point representation based on 
+  "Floating-Point Printer Sample Code", by Robert G. Burger. For example
+  repr(11./5) now returns '2.2' instead of '2.2000000000000002'.
+
 - Issue #1573: Improper use of the keyword-only syntax makes the parser crash
 
 - Issue #1564: The set implementation should special-case PyUnicode instead

Objects/floatobject.c

@@ -281,6 +281,107 @@ format_float(char *buf, size_t buflen, PyFloatObject *v, int precision)
 	format_double(buf, buflen, PyFloat_AS_DOUBLE(v), precision);
 }
 
+/* The following function is based on Tcl_PrintDouble,
+ * from tclUtil.c.
+ */
+
+#define is_infinite(d)	( (d) > DBL_MAX || (d) < -DBL_MAX )
+#define is_nan(d)		((d) != (d))
+
+static void
+format_double_repr(char *dst, double value)
+{
+    char *p, c;
+    int exp;
+    int signum;
+    char buffer[30];
+
+	/*
+	 * Handle NaN.
+	 */
+
+	if (is_nan(value)) {
+	    strcpy(dst, "nan");
+	    return;
+	}
+
+	/*
+	 * Handle infinities.
+	 */
+
+	if (is_infinite(value)) {
+	    if (value < 0) {
+		strcpy(dst, "-inf");
+	    } else {
+		strcpy(dst, "inf");
+	    }
+	    return;
+	}
+
+	/*
+	 * Ordinary (normal and denormal) values.
+	 */
+
+	exp = _PyFloat_Digits(buffer, value, &signum)+1;
+	if (signum) {
+	    *dst++ = '-';
+	}
+	p = buffer;
+	if (exp < -3 || exp > 17) {
+	    /*
+	     * E format for numbers < 1e-3 or >= 1e17.
+	     */
+
+	    *dst++ = *p++;
+	    c = *p;
+	    if (c != '\0') {
+		*dst++ = '.';
+		while (c != '\0') {
+		    *dst++ = c;
+		    c = *++p;
+		}
+	    }
+	    sprintf(dst, "e%+d", exp-1);
+	} else {
+	    /*
+	     * F format for others.
+	     */
+
+	    if (exp <= 0) {
+		*dst++ = '0';
+	    }
+	    c = *p;
+	    while (exp-- > 0) {
+		if (c != '\0') {
+		    *dst++ = c;
+		    c = *++p;
+		} else {
+		    *dst++ = '0';
+		}
+	    }
+	    *dst++ = '.';
+	    if (c == '\0') {
+		*dst++ = '0';
+	    } else {
+		while (++exp < 0) {
+		    *dst++ = '0';
+		}
+		while (c != '\0') {
+		    *dst++ = c;
+		    c = *++p;
+		}
+	    }
+	    *dst++ = '\0';
+	}
+}
+
+static void
+format_float_repr(char *buf, PyFloatObject *v)
+{
+	assert(PyFloat_Check(v));
+	format_double_repr(buf, PyFloat_AS_DOUBLE(v));
+}
+
 /* Macro and helper that convert PyObject obj to a C double and store
    the value in dbl.  If conversion to double raises an exception, obj is
    set to NULL, and the function invoking this macro returns NULL.  If
@@ -333,8 +434,8 @@ convert_to_double(PyObject **v, double *dbl)
 static PyObject *
 float_repr(PyFloatObject *v)
 {
-	char buf[100];
-	format_float(buf, sizeof(buf), v, PREC_REPR);
+	char buf[30];
+	format_float_repr(buf, v);
 	return PyUnicode_FromString(buf);
 }
 
@@ -1226,6 +1327,9 @@ _PyFloat_Init(void)
 
 	double_format = detected_double_format;
 	float_format = detected_float_format;
+	
+	/* Initialize floating point repr */
+	_PyFloat_DigitsInit();
 }
 
 void

PCbuild/pythoncore.vcproj

@@ -488,6 +488,9 @@
 			RelativePath="..\Objects\dictobject.c">
 		</File>
 		<File
+			RelativePath="..\Objects\doubledigits.c">
+		</File>
+        <File
 			RelativePath="..\PC\dl_nt.c">
 		</File>
 		<File

PCbuild8/pythoncore/pythoncore.vcproj

@@ -827,6 +827,10 @@
 				RelativePath="..\..\Objects\dictobject.c"
 				>
 			</File>
+			<File
+				RelativePath="..\..\Objects\doubledigits.c"
+				>
+			</File>
 			<File
 				RelativePath="..\..\Objects\memoryobject.c"
 				>

PCbuild9/pythoncore.vcproj

@@ -1362,6 +1362,10 @@
 				RelativePath="..\Objects\dictobject.c"
 				>
 			</File>
+			<File
+				RelativePath="..\Objects\doubledigits.c"
+				>
+			</File>
 			<File
 				RelativePath="..\Objects\enumobject.c"
 				>