Commit 4546aaec authored by Kirill Smelkov's avatar Kirill Smelkov

golang_str: Switch bstr/ustr to cdef classes

For gpython to switch builtin str/unicode to bstr/ustr we will need
bstr/ustr to have exactly the same C layout as builtin string types.
This is possible to achieve only via `cdef class`. It is also good to
switch to `cdef class` for RAM savings - from https://github.com/cython/cython/pull/5212#issuecomment-1387659026 :

    # what Cython does at runtime for `class MyBytes(bytes)`
    In [3]: MyBytes = type('MyBytes', (bytes,), {'__slots__': ()})

    In [4]: MyBytes
    Out[4]: __main__.MyBytes

    In [5]: a = bytes(b'123')

    In [6]: b = MyBytes(b'123')

    In [7]: a
    Out[7]: b'123'

    In [8]: b
    Out[8]: b'123'

    In [9]: a == b
    Out[9]: True

    In [10]: import sys

    In [11]: sys.getsizeof(a)
    Out[11]: 36

    In [12]: sys.getsizeof(b)
    Out[12]: 52

So with `cdef class` we gain more control and optimize memory usage.

This was not done before because cython forbids to `cdef class X(bytes)` due to
https://github.com/cython/cython/issues/711. We work it around in setup.py with
draft for proper patch pre-posted to upstream in https://github.com/cython/cython/pull/5212 .
parent 21fab975
......@@ -29,6 +29,15 @@ from cpython cimport Py_EQ, Py_NE, Py_LT, Py_GT, Py_LE, Py_GE
from cpython.iterobject cimport PySeqIter_New
from cpython cimport PyThreadState_GetDict, PyDict_SetItem
from cpython cimport PyObject_CheckBuffer
cdef extern from "Python.h":
ctypedef struct PyBytesObject:
pass
cdef extern from "Python.h":
ctypedef struct PyUnicodeObject:
pass
cdef extern from "Python.h":
"""
#if PY_MAJOR_VERSION < 3
......@@ -45,6 +54,7 @@ cdef extern from "Python.h":
cdef extern from "Python.h":
ctypedef int (*initproc)(object, PyObject *, PyObject *) except -1
ctypedef struct _XPyTypeObject "PyTypeObject":
PyObject* tp_new(PyTypeObject*, PyObject*, PyObject*) except NULL
initproc tp_init
PySequenceMethods *tp_as_sequence
......@@ -54,6 +64,8 @@ cdef extern from "Python.h":
object (*sq_slice) (object, Py_ssize_t, Py_ssize_t) # present only on py2
from cython cimport no_gc
from libc.stdint cimport uint8_t
from libc.stdio cimport FILE
......@@ -128,7 +140,12 @@ cdef _pyb(bcls, s): # -> ~bstr | None
return None
assert type(s) is bytes
return bytes.__new__(bcls, s)
# like bytes.__new__(bcls, s) but call bytes.tp_new directly
# else tp_new_wrapper complains because pybstr.tp_new != bytes.tp_new
argv = (s,)
obj = <object>(<_XPyTypeObject*>bytes).tp_new(<PyTypeObject*>bcls, <PyObject*>argv, NULL)
Py_DECREF(obj)
return obj
cdef _pyu(ucls, s): # -> ~ustr | None
if type(s) is ucls:
......@@ -147,7 +164,12 @@ cdef _pyu(ucls, s): # -> ~ustr | None
return None
assert type(s) is unicode
return unicode.__new__(ucls, s)
# like unicode .__new__(bcls, s) but call unicode.tp_new directly
# else tp_new_wrapper complains because pyustr.tp_new != unicode.tp_new
argv = (s,)
obj = <object>(<_XPyTypeObject*>unicode).tp_new(<PyTypeObject*>ucls, <PyObject*>argv, NULL)
Py_DECREF(obj)
return obj
# _ifbuffer_data returns contained data if obj provides buffer interface.
cdef _ifbuffer_data(obj): # -> bytes|None
......@@ -220,8 +242,8 @@ def pyuchr(int i): # -> 1-character ustr
return pyu(unichr(i))
# XXX cannot `cdef class`: github.com/cython/cython/issues/711
class pybstr(bytes):
@no_gc # note setup.py assist this to compile despite
cdef class pybstr(bytes): # https://github.com/cython/cython/issues/711
"""bstr is byte-string.
It is based on bytes and can automatically convert to/from unicode.
......@@ -253,11 +275,10 @@ class pybstr(bytes):
See also: b, ustr/u.
"""
# don't allow to set arbitrary attributes.
# won't be needed after switch to -> `cdef class`
__slots__ = ()
def __new__(cls, object='', encoding=None, errors=None):
# XXX due to "cannot `cdef class` with __new__" (https://github.com/cython/cython/issues/799)
# pybstr.__new__ is hand-made in _pybstr_tp_new which invokes ↓ .____new__() .
@staticmethod
def ____new__(cls, object='', encoding=None, errors=None):
# encoding or errors -> object must expose buffer interface
if not (encoding is None and errors is None):
object = _buffer_decode(object, encoding, errors)
......@@ -360,8 +381,10 @@ class pybstr(bytes):
def __add__(a, b):
# NOTE Cython < 3 does not automatically support __radd__ for cdef class
# https://cython.readthedocs.io/en/latest/src/userguide/migrating_to_cy30.html#arithmetic-special-methods
# but pybstr is currently _not_ cdef'ed class
# see also https://github.com/cython/cython/issues/4750
if type(a) is not pybstr:
assert type(b) is pybstr
return b.__radd__(a)
return pyb(bytes.__add__(a, _pyb_coerce(b)))
def __radd__(b, a):
......@@ -377,6 +400,9 @@ class pybstr(bytes):
# __mul__, __rmul__ (no need to override __imul__)
def __mul__(a, b):
if type(a) is not pybstr:
assert type(b) is pybstr
return b.__rmul__(a)
return pyb(bytes.__mul__(a, b))
def __rmul__(b, a):
return b.__mul__(a)
......@@ -436,8 +462,7 @@ class pybstr(bytes):
# all other string methods
def capitalize(self): return pyb(pyu(self).capitalize())
if _strhas('casefold'): # py3.3 TODO provide py2 implementation
def casefold(self): return pyb(pyu(self).casefold())
def casefold(self): return pyb(pyu(self).casefold())
def center(self, width, fillchar=' '): return pyb(pyu(self).center(width, fillchar))
def count(self, sub, start=None, end=None): return bytes.count(self, _pyb_coerce(sub), start, end)
......@@ -463,12 +488,10 @@ class pybstr(bytes):
# isascii(self) no need to override
def isdecimal(self): return pyu(self).isdecimal()
def isdigit(self): return pyu(self).isdigit()
if _strhas('isidentifier'): # py3 TODO provide fallback implementation
def isidentifier(self): return pyu(self).isidentifier()
def isidentifier(self): return pyu(self).isidentifier()
def islower(self): return pyu(self).islower()
def isnumeric(self): return pyu(self).isnumeric()
if _strhas('isprintable'): # py3 TODO provide fallback implementation
def isprintable(self): return pyu(self).isprintable()
def isprintable(self): return pyu(self).isprintable()
def isspace(self): return pyu(self).isspace()
def istitle(self): return pyu(self).istitle()
......@@ -477,10 +500,8 @@ class pybstr(bytes):
def lower(self): return pyb(pyu(self).lower())
def lstrip(self, chars=None): return pyb(pyu(self).lstrip(chars))
def partition(self, sep): return tuple(pyb(_) for _ in bytes.partition(self, _pyb_coerce(sep)))
if _strhas('removeprefix'): # py3.9 TODO provide fallback implementation
def removeprefix(self, prefix): return pyb(pyu(self).removeprefix(prefix))
if _strhas('removesuffix'): # py3.9 TODO provide fallback implementation
def removesuffix(self, suffix): return pyb(pyu(self).removesuffix(suffix))
def removeprefix(self, prefix): return pyb(pyu(self).removeprefix(prefix))
def removesuffix(self, suffix): return pyb(pyu(self).removesuffix(suffix))
def replace(self, old, new, count=-1): return pyb(bytes.replace(self, _pyb_coerce(old), _pyb_coerce(new), count))
# NOTE rfind/rindex & friends should return byte-position, not unicode-position
......@@ -528,8 +549,35 @@ class pybstr(bytes):
return pyustr.maketrans(x, y, z)
# XXX cannot `cdef class` with __new__: https://github.com/cython/cython/issues/799
class pyustr(unicode):
# hand-made pybstr.__new__ (workaround for https://github.com/cython/cython/issues/799)
cdef PyObject* _pybstr_tp_new(PyTypeObject* _cls, PyObject* _argv, PyObject* _kw) except NULL:
argv = ()
if _argv != NULL:
argv = <object>_argv
kw = {}
if _kw != NULL:
kw = <object>_kw
cdef object x = pybstr.____new__(<object>_cls, *argv, **kw)
Py_INCREF(x)
return <PyObject*>x
(<_XPyTypeObject*>pybstr).tp_new = &_pybstr_tp_new
# bytes uses "optimized" and custom .tp_basicsize and .tp_itemsize:
# https://github.com/python/cpython/blob/v2.7.18-0-g8d21aa21f2c/Objects/stringobject.c#L26-L32
# https://github.com/python/cpython/blob/v2.7.18-0-g8d21aa21f2c/Objects/stringobject.c#L3816-L3820
(<PyTypeObject*>pybstr) .tp_basicsize = (<PyTypeObject*>bytes).tp_basicsize
(<PyTypeObject*>pybstr) .tp_itemsize = (<PyTypeObject*>bytes).tp_itemsize
# make sure pybstr C layout corresponds to bytes C layout exactly
# we patched cython to allow from-bytes cdef class inheritance and we also set
# .tp_basicsize directly above. All this works ok only if C layouts for pybstr
# and bytes are completely the same.
assert sizeof(pybstr) == sizeof(PyBytesObject)
@no_gc
cdef class pyustr(unicode):
"""ustr is unicode-string.
It is based on unicode and can automatically convert to/from bytes.
......@@ -556,11 +604,10 @@ class pyustr(unicode):
See also: u, bstr/b.
"""
# don't allow to set arbitrary attributes.
# won't be needed after switch to -> `cdef class`
__slots__ = ()
def __new__(cls, object='', encoding=None, errors=None):
# XXX due to "cannot `cdef class` with __new__" (https://github.com/cython/cython/issues/799)
# pyustr.__new__ is hand-made in _pyustr_tp_new which invokes ↓ .____new__() .
@staticmethod
def ____new__(cls, object='', encoding=None, errors=None):
# encoding or errors -> object must expose buffer interface
if not (encoding is None and errors is None):
object = _buffer_decode(object, encoding, errors)
......@@ -652,8 +699,10 @@ class pyustr(unicode):
def __add__(a, b):
# NOTE Cython < 3 does not automatically support __radd__ for cdef class
# https://cython.readthedocs.io/en/latest/src/userguide/migrating_to_cy30.html#arithmetic-special-methods
# but pyustr is currently _not_ cdef'ed class
# see also https://github.com/cython/cython/issues/4750
if type(a) is not pyustr:
assert type(b) is pyustr
return b.__radd__(a)
return pyu(unicode.__add__(a, _pyu_coerce(b)))
def __radd__(b, a):
......@@ -671,6 +720,9 @@ class pyustr(unicode):
# __mul__, __rmul__ (no need to override __imul__)
def __mul__(a, b):
if type(a) is not pyustr:
assert type(b) is pyustr
return b.__rmul__(a)
return pyu(unicode.__mul__(a, b))
def __rmul__(b, a):
return b.__mul__(a)
......@@ -723,8 +775,7 @@ class pyustr(unicode):
# all other string methods
def capitalize(self): return pyu(unicode.capitalize(self))
if _strhas('casefold'): # py3.3 TODO provide fallback implementation
def casefold(self): return pyu(unicode.casefold(self))
def casefold(self): return pyu(unicode.casefold(self))
def center(self, width, fillchar=' '): return pyu(unicode.center(self, width, _pyu_coerce(fillchar)))
def count(self, sub, start=None, end=None):
# cython optimizes unicode.count to directly call PyUnicode_Count -
......@@ -768,10 +819,8 @@ class pyustr(unicode):
def lower(self): return pyu(unicode.lower(self))
def lstrip(self, chars=None): return pyu(unicode.lstrip(self, _xpyu_coerce(chars)))
def partition(self, sep): return tuple(pyu(_) for _ in unicode.partition(self, _pyu_coerce(sep)))
if _strhas('removeprefix'): # py3.9 TODO provide fallback implementation
def removeprefix(self, prefix): return pyu(unicode.removeprefix(self, _pyu_coerce(prefix)))
if _strhas('removesuffix'): # py3.9 TODO provide fallback implementation
def removesuffix(self, suffix): return pyu(unicode.removesuffix(self, _pyu_coerce(suffix)))
def removeprefix(self, prefix): return pyu(unicode.removeprefix(self, _pyu_coerce(prefix)))
def removesuffix(self, suffix): return pyu(unicode.removesuffix(self, _pyu_coerce(suffix)))
def replace(self, old, new, count=-1): return pyu(unicode.replace(self, _pyu_coerce(old), _pyu_coerce(new), count))
def rfind(self, sub, start=None, end=None):
if start is None: start = 0
......@@ -864,6 +913,24 @@ class pyustr(unicode):
return t
# hand-made pyustr.__new__ (workaround for https://github.com/cython/cython/issues/799)
cdef PyObject* _pyustr_tp_new(PyTypeObject* _cls, PyObject* _argv, PyObject* _kw) except NULL:
argv = ()
if _argv != NULL:
argv = <object>_argv
kw = {}
if _kw != NULL:
kw = <object>_kw
cdef object x = pyustr.____new__(<object>_cls, *argv, **kw)
Py_INCREF(x)
return <PyObject*>x
(<_XPyTypeObject*>pyustr).tp_new = &_pyustr_tp_new
# similarly to bytes - want same C layout for pyustr vs unicode
assert sizeof(pyustr) == sizeof(PyUnicodeObject)
# _pyustrIter wraps unicode iterator to return pyustr for each yielded character.
cdef class _pyustrIter:
cdef object uiter
......@@ -941,6 +1008,31 @@ if PY2:
(<_XPyTypeObject*>pyustr) .tp_as_sequence.sq_slice = NULL
# ---- adjust bstr/ustr classes after what cython generated ----
# remove unsupported bstr/ustr methods. do it outside of `cdef class` to
# workaround https://github.com/cython/cython/issues/4556 (`if ...` during
# `cdef class` is silently handled wrongly)
cdef _bstrustr_remove_unsupported_slots():
vslot = (
'casefold', # py3.3 TODO provide py2 implementation
'isidentifier', # py3 TODO provide fallback implementation
'isprintable', # py3 TODO provide fallback implementation
'removeprefix', # py3.9 TODO provide fallback implementation
'removesuffix', # py3.9 TODO provide fallback implementation
)
for slot in vslot:
if not hasattr(unicode, slot):
_patch_slot(<PyTypeObject*>pybstr, slot, DEL)
try:
_patch_slot(<PyTypeObject*>pyustr, slot, DEL)
except KeyError: # e.g. we do not define ustr.isprintable ourselves
pass
_bstrustr_remove_unsupported_slots()
# ---- quoting ----
# _bpysmartquote_u3b2 quotes bytes/bytearray s the same way python would do for string.
#
# nonascii_escape indicates whether \xNN with NN >= 0x80 is present in the output.
......@@ -1321,12 +1413,15 @@ cdef _InBStringify _inbstringify_get():
#
# if func_or_descr is descriptor (has __get__), it is installed as is.
# otherwise it is wrapped with "unbound method" descriptor.
#
# if func_or_descr is DEL the slot is removed from typ's __dict__.
cdef DEL = object()
cdef _patch_slot(PyTypeObject* typ, str name, object func_or_descr):
typdict = <dict>(typ.tp_dict)
#print("\npatching %s.%s with %r" % (typ.tp_name, name, func_or_descr))
#print("old: %r" % typdict.get(name))
if hasattr(func_or_descr, '__get__'):
if hasattr(func_or_descr, '__get__') or func_or_descr is DEL:
descr = func_or_descr
else:
func = func_or_descr
......@@ -1335,7 +1430,10 @@ cdef _patch_slot(PyTypeObject* typ, str name, object func_or_descr):
else:
descr = _UnboundMethod(func)
typdict[name] = descr
if descr is DEL:
del typdict[name]
else:
typdict[name] = descr
#print("new: %r" % typdict.get(name))
PyType_Modified(typ)
......@@ -1686,10 +1784,6 @@ class _BFormatter(pystring.Formatter):
# ---- misc ----
# _strhas returns whether unicode string type has specified method.
cdef bint _strhas(str meth) except *:
return hasattr(unicode, meth)
cdef object _xpyu_coerce(obj):
return _pyu_coerce(obj) if obj is not None else None
......
......@@ -31,7 +31,7 @@ import sys
import six
from six import text_type as unicode, unichr
from six.moves import range as xrange
import re, pickle, copy, types
import gc, re, pickle, copy, types
import array, collections
......@@ -284,6 +284,25 @@ def test_strings_basic():
bs.hello = 1
# verify that bstr/ustr are created with correct refcount.
def test_strings_refcount():
# first verify our logic on std type
obj = xbytes(u'abc'); assert type(obj) is bytes
gc.collect(); assert sys.getrefcount(obj) == 1+1 # +1 due to obj passed to getrefcount call
# bstr
obj = b('abc'); assert type(obj) is bstr
gc.collect(); assert sys.getrefcount(obj) == 1+1
obj = bstr('abc'); assert type(obj) is bstr
gc.collect(); assert sys.getrefcount(obj) == 1+1
# ustr
obj = u('abc'); assert type(obj) is ustr
gc.collect(); assert sys.getrefcount(obj) == 1+1
obj = ustr('abc'); assert type(obj) is ustr
gc.collect(); assert sys.getrefcount(obj) == 1+1
# verify memoryview(bstr|ustr).
def test_strings_memoryview():
bs = b('мир')
......
......@@ -18,6 +18,25 @@
# See COPYING file for full licensing terms.
# See https://www.nexedi.com/licensing for rationale and options.
# patch cython to allow `cdef class X(bytes)` while building pygolang to
# workaround https://github.com/cython/cython/issues/711
# see `cdef class pybstr` in golang/_golang_str.pyx for details.
# (should become unneeded with cython 3 once https://github.com/cython/cython/pull/5212 is finished)
import inspect
from Cython.Compiler.PyrexTypes import BuiltinObjectType
def pygo_cy_builtin_type_name_set(self, v):
self._pygo_name = v
def pygo_cy_builtin_type_name_get(self):
name = self._pygo_name
if name == 'bytes':
caller = inspect.currentframe().f_back.f_code.co_name
if caller == 'analyse_declarations':
# need anything different from 'bytes' to deactivate check in
# https://github.com/cython/cython/blob/c21b39d4/Cython/Compiler/Nodes.py#L4759-L4762
name = 'xxx'
return name
BuiltinObjectType.name = property(pygo_cy_builtin_type_name_get, pygo_cy_builtin_type_name_set)
from setuptools import find_packages
from setuptools.command.install_scripts import install_scripts as _install_scripts
from setuptools.command.develop import develop as _develop
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment