cpython/Tools/build/freeze_modules.py

685 lines
21 KiB
Python

"""Freeze modules and regen related files (e.g. Python/frozen.c).
See the notes at the top of Python/frozen.c for more info.
"""
from collections import namedtuple
import hashlib
import os
import ntpath
import posixpath
import argparse
from update_file import updating_file_with_tmpfile
ROOT_DIR = os.path.dirname(os.path.dirname(os.path.dirname(__file__)))
ROOT_DIR = os.path.abspath(ROOT_DIR)
FROZEN_ONLY = os.path.join(ROOT_DIR, 'Tools', 'freeze', 'flag.py')
STDLIB_DIR = os.path.join(ROOT_DIR, 'Lib')
# If FROZEN_MODULES_DIR or DEEPFROZEN_MODULES_DIR is changed then the
# .gitattributes and .gitignore files needs to be updated.
FROZEN_MODULES_DIR = os.path.join(ROOT_DIR, 'Python', 'frozen_modules')
FROZEN_FILE = os.path.join(ROOT_DIR, 'Python', 'frozen.c')
MAKEFILE = os.path.join(ROOT_DIR, 'Makefile.pre.in')
PCBUILD_PROJECT = os.path.join(ROOT_DIR, 'PCbuild', '_freeze_module.vcxproj')
PCBUILD_FILTERS = os.path.join(ROOT_DIR, 'PCbuild', '_freeze_module.vcxproj.filters')
PCBUILD_PYTHONCORE = os.path.join(ROOT_DIR, 'PCbuild', 'pythoncore.vcxproj')
OS_PATH = 'ntpath' if os.name == 'nt' else 'posixpath'
# These are modules that get frozen.
# If you're debugging new bytecode instructions,
# you can delete all sections except 'import system'.
# This also speeds up building somewhat.
TESTS_SECTION = 'Test module'
FROZEN = [
# See parse_frozen_spec() for the format.
# In cases where the frozenid is duplicated, the first one is re-used.
('import system', [
# These frozen modules are necessary for bootstrapping
# the import system.
'importlib._bootstrap : _frozen_importlib',
'importlib._bootstrap_external : _frozen_importlib_external',
# This module is important because some Python builds rely
# on a builtin zip file instead of a filesystem.
'zipimport',
]),
# (You can delete entries from here down to the end of the list.)
('stdlib - startup, without site (python -S)', [
'abc',
'codecs',
# For now we do not freeze the encodings, due # to the noise all
# those extra modules add to the text printed during the build.
# (See https://github.com/python/cpython/pull/28398#pullrequestreview-756856469.)
#'<encodings.*>',
'io',
]),
('stdlib - startup, with site', [
'_collections_abc',
'_sitebuiltins',
'genericpath',
'ntpath',
'posixpath',
# We must explicitly mark os.path as a frozen module
# even though it will never be imported.
f'{OS_PATH} : os.path',
'os',
'site',
'stat',
]),
('runpy - run module with -m', [
"importlib.util",
"importlib.machinery",
"runpy",
]),
(TESTS_SECTION, [
'__hello__',
'__hello__ : __hello_alias__',
'__hello__ : <__phello_alias__>',
'__hello__ : __phello_alias__.spam',
'<__phello__.**.*>',
f'frozen_only : __hello_only__ = {FROZEN_ONLY}',
]),
# (End of stuff you could delete.)
]
BOOTSTRAP = {
'importlib._bootstrap',
'importlib._bootstrap_external',
'zipimport',
}
#######################################
# platform-specific helpers
if os.path is posixpath:
relpath_for_posix_display = os.path.relpath
def relpath_for_windows_display(path, base):
return ntpath.relpath(
ntpath.join(*path.split(os.path.sep)),
ntpath.join(*base.split(os.path.sep)),
)
else:
relpath_for_windows_display = ntpath.relpath
def relpath_for_posix_display(path, base):
return posixpath.relpath(
posixpath.join(*path.split(os.path.sep)),
posixpath.join(*base.split(os.path.sep)),
)
#######################################
# specs
def parse_frozen_specs():
seen = {}
for section, specs in FROZEN:
parsed = _parse_specs(specs, section, seen)
for item in parsed:
frozenid, pyfile, modname, ispkg, section = item
try:
source = seen[frozenid]
except KeyError:
source = FrozenSource.from_id(frozenid, pyfile)
seen[frozenid] = source
else:
assert not pyfile or pyfile == source.pyfile, item
yield FrozenModule(modname, ispkg, section, source)
def _parse_specs(specs, section, seen):
for spec in specs:
info, subs = _parse_spec(spec, seen, section)
yield info
for info in subs or ():
yield info
def _parse_spec(spec, knownids=None, section=None):
"""Yield an info tuple for each module corresponding to the given spec.
The info consists of: (frozenid, pyfile, modname, ispkg, section).
Supported formats:
frozenid
frozenid : modname
frozenid : modname = pyfile
"frozenid" and "modname" must be valid module names (dot-separated
identifiers). If "modname" is not provided then "frozenid" is used.
If "pyfile" is not provided then the filename of the module
corresponding to "frozenid" is used.
Angle brackets around a frozenid (e.g. '<encodings>") indicate
it is a package. This also means it must be an actual module
(i.e. "pyfile" cannot have been provided). Such values can have
patterns to expand submodules:
<encodings.*> - also freeze all direct submodules
<encodings.**.*> - also freeze the full submodule tree
As with "frozenid", angle brackets around "modname" indicate
it is a package. However, in this case "pyfile" should not
have been provided and patterns in "modname" are not supported.
Also, if "modname" has brackets then "frozenid" should not,
and "pyfile" should have been provided..
"""
frozenid, _, remainder = spec.partition(':')
modname, _, pyfile = remainder.partition('=')
frozenid = frozenid.strip()
modname = modname.strip()
pyfile = pyfile.strip()
submodules = None
if modname.startswith('<') and modname.endswith('>'):
assert check_modname(frozenid), spec
modname = modname[1:-1]
assert check_modname(modname), spec
if frozenid in knownids:
pass
elif pyfile:
assert not os.path.isdir(pyfile), spec
else:
pyfile = _resolve_module(frozenid, ispkg=False)
ispkg = True
elif pyfile:
assert check_modname(frozenid), spec
assert not knownids or frozenid not in knownids, spec
assert check_modname(modname), spec
assert not os.path.isdir(pyfile), spec
ispkg = False
elif knownids and frozenid in knownids:
assert check_modname(frozenid), spec
assert check_modname(modname), spec
ispkg = False
else:
assert not modname or check_modname(modname), spec
resolved = iter(resolve_modules(frozenid))
frozenid, pyfile, ispkg = next(resolved)
if not modname:
modname = frozenid
if ispkg:
pkgid = frozenid
pkgname = modname
pkgfiles = {pyfile: pkgid}
def iter_subs():
for frozenid, pyfile, ispkg in resolved:
if pkgname:
modname = frozenid.replace(pkgid, pkgname, 1)
else:
modname = frozenid
if pyfile:
if pyfile in pkgfiles:
frozenid = pkgfiles[pyfile]
pyfile = None
elif ispkg:
pkgfiles[pyfile] = frozenid
yield frozenid, pyfile, modname, ispkg, section
submodules = iter_subs()
info = (frozenid, pyfile or None, modname, ispkg, section)
return info, submodules
#######################################
# frozen source files
class FrozenSource(namedtuple('FrozenSource', 'id pyfile frozenfile')):
@classmethod
def from_id(cls, frozenid, pyfile=None):
if not pyfile:
pyfile = os.path.join(STDLIB_DIR, *frozenid.split('.')) + '.py'
#assert os.path.exists(pyfile), (frozenid, pyfile)
frozenfile = resolve_frozen_file(frozenid, FROZEN_MODULES_DIR)
return cls(frozenid, pyfile, frozenfile)
@property
def frozenid(self):
return self.id
@property
def modname(self):
if self.pyfile.startswith(STDLIB_DIR):
return self.id
return None
@property
def symbol(self):
# This matches what we do in Programs/_freeze_module.c:
name = self.frozenid.replace('.', '_')
return '_Py_M__' + name
@property
def ispkg(self):
if not self.pyfile:
return False
elif self.frozenid.endswith('.__init__'):
return False
else:
return os.path.basename(self.pyfile) == '__init__.py'
@property
def isbootstrap(self):
return self.id in BOOTSTRAP
def resolve_frozen_file(frozenid, destdir):
"""Return the filename corresponding to the given frozen ID.
For stdlib modules the ID will always be the full name
of the source module.
"""
if not isinstance(frozenid, str):
try:
frozenid = frozenid.frozenid
except AttributeError:
raise ValueError(f'unsupported frozenid {frozenid!r}')
# We use a consistent naming convention for all frozen modules.
frozenfile = f'{frozenid}.h'
if not destdir:
return frozenfile
return os.path.join(destdir, frozenfile)
#######################################
# frozen modules
class FrozenModule(namedtuple('FrozenModule', 'name ispkg section source')):
def __getattr__(self, name):
return getattr(self.source, name)
@property
def modname(self):
return self.name
@property
def orig(self):
return self.source.modname
@property
def isalias(self):
orig = self.source.modname
if not orig:
return True
return self.name != orig
def summarize(self):
source = self.source.modname
if source:
source = f'<{source}>'
else:
source = relpath_for_posix_display(self.pyfile, ROOT_DIR)
return {
'module': self.name,
'ispkg': self.ispkg,
'source': source,
'frozen': os.path.basename(self.frozenfile),
'checksum': _get_checksum(self.frozenfile),
}
def _iter_sources(modules):
seen = set()
for mod in modules:
if mod.source not in seen:
yield mod.source
seen.add(mod.source)
#######################################
# generic helpers
def _get_checksum(filename):
with open(filename, "rb") as infile:
contents = infile.read()
m = hashlib.sha256()
m.update(contents)
return m.hexdigest()
def resolve_modules(modname, pyfile=None):
if modname.startswith('<') and modname.endswith('>'):
if pyfile:
assert os.path.isdir(pyfile) or os.path.basename(pyfile) == '__init__.py', pyfile
ispkg = True
modname = modname[1:-1]
rawname = modname
# For now, we only expect match patterns at the end of the name.
_modname, sep, match = modname.rpartition('.')
if sep:
if _modname.endswith('.**'):
modname = _modname[:-3]
match = f'**.{match}'
elif match and not match.isidentifier():
modname = _modname
# Otherwise it's a plain name so we leave it alone.
else:
match = None
else:
ispkg = False
rawname = modname
match = None
if not check_modname(modname):
raise ValueError(f'not a valid module name ({rawname})')
if not pyfile:
pyfile = _resolve_module(modname, ispkg=ispkg)
elif os.path.isdir(pyfile):
pyfile = _resolve_module(modname, pyfile, ispkg)
yield modname, pyfile, ispkg
if match:
pkgdir = os.path.dirname(pyfile)
yield from iter_submodules(modname, pkgdir, match)
def check_modname(modname):
return all(n.isidentifier() for n in modname.split('.'))
def iter_submodules(pkgname, pkgdir=None, match='*'):
if not pkgdir:
pkgdir = os.path.join(STDLIB_DIR, *pkgname.split('.'))
if not match:
match = '**.*'
match_modname = _resolve_modname_matcher(match, pkgdir)
def _iter_submodules(pkgname, pkgdir):
for entry in sorted(os.scandir(pkgdir), key=lambda e: e.name):
matched, recursive = match_modname(entry.name)
if not matched:
continue
modname = f'{pkgname}.{entry.name}'
if modname.endswith('.py'):
yield modname[:-3], entry.path, False
elif entry.is_dir():
pyfile = os.path.join(entry.path, '__init__.py')
# We ignore namespace packages.
if os.path.exists(pyfile):
yield modname, pyfile, True
if recursive:
yield from _iter_submodules(modname, entry.path)
return _iter_submodules(pkgname, pkgdir)
def _resolve_modname_matcher(match, rootdir=None):
if isinstance(match, str):
if match.startswith('**.'):
recursive = True
pat = match[3:]
assert match
else:
recursive = False
pat = match
if pat == '*':
def match_modname(modname):
return True, recursive
else:
raise NotImplementedError(match)
elif callable(match):
match_modname = match(rootdir)
else:
raise ValueError(f'unsupported matcher {match!r}')
return match_modname
def _resolve_module(modname, pathentry=STDLIB_DIR, ispkg=False):
assert pathentry, pathentry
pathentry = os.path.normpath(pathentry)
assert os.path.isabs(pathentry)
if ispkg:
return os.path.join(pathentry, *modname.split('.'), '__init__.py')
return os.path.join(pathentry, *modname.split('.')) + '.py'
#######################################
# regenerating dependent files
def find_marker(lines, marker, file):
for pos, line in enumerate(lines):
if marker in line:
return pos
raise Exception(f"Can't find {marker!r} in file {file}")
def replace_block(lines, start_marker, end_marker, replacements, file):
start_pos = find_marker(lines, start_marker, file)
end_pos = find_marker(lines, end_marker, file)
if end_pos <= start_pos:
raise Exception(f"End marker {end_marker!r} "
f"occurs before start marker {start_marker!r} "
f"in file {file}")
replacements = [line.rstrip() + '\n' for line in replacements]
return lines[:start_pos + 1] + replacements + lines[end_pos:]
class UniqueList(list):
def __init__(self):
self._seen = set()
def append(self, item):
if item in self._seen:
return
super().append(item)
self._seen.add(item)
def regen_frozen(modules):
headerlines = []
parentdir = os.path.dirname(FROZEN_FILE)
for src in _iter_sources(modules):
# Adding a comment to separate sections here doesn't add much,
# so we don't.
header = relpath_for_posix_display(src.frozenfile, parentdir)
headerlines.append(f'#include "{header}"')
externlines = UniqueList()
bootstraplines = []
stdliblines = []
testlines = []
aliaslines = []
indent = ' '
lastsection = None
for mod in modules:
if mod.isbootstrap:
lines = bootstraplines
elif mod.section == TESTS_SECTION:
lines = testlines
else:
lines = stdliblines
if mod.section != lastsection:
if lastsection is not None:
lines.append('')
lines.append(f'/* {mod.section} */')
lastsection = mod.section
pkg = 'true' if mod.ispkg else 'false'
size = f"(int)sizeof({mod.symbol})"
line = f'{{"{mod.name}", {mod.symbol}, {size}, {pkg}}},'
lines.append(line)
if mod.isalias:
if not mod.orig:
entry = '{"%s", NULL},' % (mod.name,)
elif mod.source.ispkg:
entry = '{"%s", "<%s"},' % (mod.name, mod.orig)
else:
entry = '{"%s", "%s"},' % (mod.name, mod.orig)
aliaslines.append(indent + entry)
for lines in (bootstraplines, stdliblines, testlines):
# TODO: Is this necessary any more?
if lines and not lines[0]:
del lines[0]
for i, line in enumerate(lines):
if line:
lines[i] = indent + line
print(f'# Updating {os.path.relpath(FROZEN_FILE)}')
with updating_file_with_tmpfile(FROZEN_FILE) as (infile, outfile):
lines = infile.readlines()
# TODO: Use more obvious markers, e.g.
# $START GENERATED FOOBAR$ / $END GENERATED FOOBAR$
lines = replace_block(
lines,
"/* Includes for frozen modules: */",
"/* End includes */",
headerlines,
FROZEN_FILE,
)
lines = replace_block(
lines,
"static const struct _frozen bootstrap_modules[] =",
"/* bootstrap sentinel */",
bootstraplines,
FROZEN_FILE,
)
lines = replace_block(
lines,
"static const struct _frozen stdlib_modules[] =",
"/* stdlib sentinel */",
stdliblines,
FROZEN_FILE,
)
lines = replace_block(
lines,
"static const struct _frozen test_modules[] =",
"/* test sentinel */",
testlines,
FROZEN_FILE,
)
lines = replace_block(
lines,
"const struct _module_alias aliases[] =",
"/* aliases sentinel */",
aliaslines,
FROZEN_FILE,
)
outfile.writelines(lines)
def regen_makefile(modules):
pyfiles = []
frozenfiles = []
rules = ['']
for src in _iter_sources(modules):
frozen_header = relpath_for_posix_display(src.frozenfile, ROOT_DIR)
frozenfiles.append(f'\t\t{frozen_header} \\')
pyfile = relpath_for_posix_display(src.pyfile, ROOT_DIR)
pyfiles.append(f'\t\t{pyfile} \\')
if src.isbootstrap:
freezecmd = '$(FREEZE_MODULE_BOOTSTRAP)'
freezedep = '$(FREEZE_MODULE_BOOTSTRAP_DEPS)'
else:
freezecmd = '$(FREEZE_MODULE)'
freezedep = '$(FREEZE_MODULE_DEPS)'
freeze = (f'{freezecmd} {src.frozenid} '
f'$(srcdir)/{pyfile} {frozen_header}')
rules.extend([
f'{frozen_header}: {pyfile} {freezedep}',
f'\t{freeze}',
'',
])
pyfiles[-1] = pyfiles[-1].rstrip(" \\")
frozenfiles[-1] = frozenfiles[-1].rstrip(" \\")
print(f'# Updating {os.path.relpath(MAKEFILE)}')
with updating_file_with_tmpfile(MAKEFILE) as (infile, outfile):
lines = infile.readlines()
lines = replace_block(
lines,
"FROZEN_FILES_IN =",
"# End FROZEN_FILES_IN",
pyfiles,
MAKEFILE,
)
lines = replace_block(
lines,
"FROZEN_FILES_OUT =",
"# End FROZEN_FILES_OUT",
frozenfiles,
MAKEFILE,
)
lines = replace_block(
lines,
"# BEGIN: freezing modules",
"# END: freezing modules",
rules,
MAKEFILE,
)
outfile.writelines(lines)
def regen_pcbuild(modules):
projlines = []
filterlines = []
corelines = []
for src in _iter_sources(modules):
pyfile = relpath_for_windows_display(src.pyfile, ROOT_DIR)
header = relpath_for_windows_display(src.frozenfile, ROOT_DIR)
intfile = ntpath.splitext(ntpath.basename(header))[0] + '.g.h'
projlines.append(f' <None Include="..\\{pyfile}">')
projlines.append(f' <ModName>{src.frozenid}</ModName>')
projlines.append(f' <IntFile>$(IntDir){intfile}</IntFile>')
projlines.append(f' <OutFile>$(GeneratedFrozenModulesDir){header}</OutFile>')
projlines.append(f' </None>')
filterlines.append(f' <None Include="..\\{pyfile}">')
filterlines.append(' <Filter>Python Files</Filter>')
filterlines.append(' </None>')
print(f'# Updating {os.path.relpath(PCBUILD_PROJECT)}')
with updating_file_with_tmpfile(PCBUILD_PROJECT) as (infile, outfile):
lines = infile.readlines()
lines = replace_block(
lines,
'<!-- BEGIN frozen modules -->',
'<!-- END frozen modules -->',
projlines,
PCBUILD_PROJECT,
)
outfile.writelines(lines)
print(f'# Updating {os.path.relpath(PCBUILD_FILTERS)}')
with updating_file_with_tmpfile(PCBUILD_FILTERS) as (infile, outfile):
lines = infile.readlines()
lines = replace_block(
lines,
'<!-- BEGIN frozen modules -->',
'<!-- END frozen modules -->',
filterlines,
PCBUILD_FILTERS,
)
outfile.writelines(lines)
#######################################
# the script
def main():
# Expand the raw specs, preserving order.
modules = list(parse_frozen_specs())
# Regen build-related files.
regen_makefile(modules)
regen_pcbuild(modules)
regen_frozen(modules)
if __name__ == '__main__':
main()