cpython/Lib/distutils/command/build_ext.py
Bernhard M. Wiedemann 0d30ae1a03 bpo-36302: Sort list of sources (GH-12341)
When building packages (e.g. for openSUSE Linux)
(random) filesystem order of input files
influences ordering of functions in the output .so files.
Thus without the patch, builds (in disposable VMs) would usually differ.

Without this patch, all callers have to be patched individually
https://github.com/dugsong/libdnet/pull/42
https://github.com/sass/libsass-python/pull/212
https://github.com/tahoe-lafs/pycryptopp/pull/41
https://github.com/yt-project/yt/pull/2206
https://github.com/pyproj4/pyproj/pull/142
https://github.com/pytries/datrie/pull/49
https://github.com/Roche/pyreadstat/pull/37
but that is an infinite effort.

See https://reproducible-builds.org/ for why this matters.
2019-08-01 15:18:03 +02:00

746 lines
31 KiB
Python

"""distutils.command.build_ext
Implements the Distutils 'build_ext' command, for building extension
modules (currently limited to C extensions, should accommodate C++
extensions ASAP)."""
import contextlib
import os
import re
import sys
from distutils.core import Command
from distutils.errors import *
from distutils.sysconfig import customize_compiler, get_python_version
from distutils.sysconfig import get_config_h_filename
from distutils.dep_util import newer_group
from distutils.extension import Extension
from distutils.util import get_platform
from distutils import log
from site import USER_BASE
# An extension name is just a dot-separated list of Python NAMEs (ie.
# the same as a fully-qualified module name).
extension_name_re = re.compile \
(r'^[a-zA-Z_][a-zA-Z_0-9]*(\.[a-zA-Z_][a-zA-Z_0-9]*)*$')
def show_compilers ():
from distutils.ccompiler import show_compilers
show_compilers()
class build_ext(Command):
description = "build C/C++ extensions (compile/link to build directory)"
# XXX thoughts on how to deal with complex command-line options like
# these, i.e. how to make it so fancy_getopt can suck them off the
# command line and make it look like setup.py defined the appropriate
# lists of tuples of what-have-you.
# - each command needs a callback to process its command-line options
# - Command.__init__() needs access to its share of the whole
# command line (must ultimately come from
# Distribution.parse_command_line())
# - it then calls the current command class' option-parsing
# callback to deal with weird options like -D, which have to
# parse the option text and churn out some custom data
# structure
# - that data structure (in this case, a list of 2-tuples)
# will then be present in the command object by the time
# we get to finalize_options() (i.e. the constructor
# takes care of both command-line and client options
# in between initialize_options() and finalize_options())
sep_by = " (separated by '%s')" % os.pathsep
user_options = [
('build-lib=', 'b',
"directory for compiled extension modules"),
('build-temp=', 't',
"directory for temporary files (build by-products)"),
('plat-name=', 'p',
"platform name to cross-compile for, if supported "
"(default: %s)" % get_platform()),
('inplace', 'i',
"ignore build-lib and put compiled extensions into the source " +
"directory alongside your pure Python modules"),
('include-dirs=', 'I',
"list of directories to search for header files" + sep_by),
('define=', 'D',
"C preprocessor macros to define"),
('undef=', 'U',
"C preprocessor macros to undefine"),
('libraries=', 'l',
"external C libraries to link with"),
('library-dirs=', 'L',
"directories to search for external C libraries" + sep_by),
('rpath=', 'R',
"directories to search for shared C libraries at runtime"),
('link-objects=', 'O',
"extra explicit link objects to include in the link"),
('debug', 'g',
"compile/link with debugging information"),
('force', 'f',
"forcibly build everything (ignore file timestamps)"),
('compiler=', 'c',
"specify the compiler type"),
('parallel=', 'j',
"number of parallel build jobs"),
('swig-cpp', None,
"make SWIG create C++ files (default is C)"),
('swig-opts=', None,
"list of SWIG command line options"),
('swig=', None,
"path to the SWIG executable"),
('user', None,
"add user include, library and rpath")
]
boolean_options = ['inplace', 'debug', 'force', 'swig-cpp', 'user']
help_options = [
('help-compiler', None,
"list available compilers", show_compilers),
]
def initialize_options(self):
self.extensions = None
self.build_lib = None
self.plat_name = None
self.build_temp = None
self.inplace = 0
self.package = None
self.include_dirs = None
self.define = None
self.undef = None
self.libraries = None
self.library_dirs = None
self.rpath = None
self.link_objects = None
self.debug = None
self.force = None
self.compiler = None
self.swig = None
self.swig_cpp = None
self.swig_opts = None
self.user = None
self.parallel = None
def finalize_options(self):
from distutils import sysconfig
self.set_undefined_options('build',
('build_lib', 'build_lib'),
('build_temp', 'build_temp'),
('compiler', 'compiler'),
('debug', 'debug'),
('force', 'force'),
('parallel', 'parallel'),
('plat_name', 'plat_name'),
)
if self.package is None:
self.package = self.distribution.ext_package
self.extensions = self.distribution.ext_modules
# Make sure Python's include directories (for Python.h, pyconfig.h,
# etc.) are in the include search path.
py_include = sysconfig.get_python_inc()
plat_py_include = sysconfig.get_python_inc(plat_specific=1)
if self.include_dirs is None:
self.include_dirs = self.distribution.include_dirs or []
if isinstance(self.include_dirs, str):
self.include_dirs = self.include_dirs.split(os.pathsep)
# If in a virtualenv, add its include directory
# Issue 16116
if sys.exec_prefix != sys.base_exec_prefix:
self.include_dirs.append(os.path.join(sys.exec_prefix, 'include'))
# Put the Python "system" include dir at the end, so that
# any local include dirs take precedence.
self.include_dirs.extend(py_include.split(os.path.pathsep))
if plat_py_include != py_include:
self.include_dirs.extend(
plat_py_include.split(os.path.pathsep))
self.ensure_string_list('libraries')
self.ensure_string_list('link_objects')
# Life is easier if we're not forever checking for None, so
# simplify these options to empty lists if unset
if self.libraries is None:
self.libraries = []
if self.library_dirs is None:
self.library_dirs = []
elif isinstance(self.library_dirs, str):
self.library_dirs = self.library_dirs.split(os.pathsep)
if self.rpath is None:
self.rpath = []
elif isinstance(self.rpath, str):
self.rpath = self.rpath.split(os.pathsep)
# for extensions under windows use different directories
# for Release and Debug builds.
# also Python's library directory must be appended to library_dirs
if os.name == 'nt':
# the 'libs' directory is for binary installs - we assume that
# must be the *native* platform. But we don't really support
# cross-compiling via a binary install anyway, so we let it go.
self.library_dirs.append(os.path.join(sys.exec_prefix, 'libs'))
if sys.base_exec_prefix != sys.prefix: # Issue 16116
self.library_dirs.append(os.path.join(sys.base_exec_prefix, 'libs'))
if self.debug:
self.build_temp = os.path.join(self.build_temp, "Debug")
else:
self.build_temp = os.path.join(self.build_temp, "Release")
# Append the source distribution include and library directories,
# this allows distutils on windows to work in the source tree
self.include_dirs.append(os.path.dirname(get_config_h_filename()))
_sys_home = getattr(sys, '_home', None)
if _sys_home:
self.library_dirs.append(_sys_home)
# Use the .lib files for the correct architecture
if self.plat_name == 'win32':
suffix = 'win32'
else:
# win-amd64
suffix = self.plat_name[4:]
new_lib = os.path.join(sys.exec_prefix, 'PCbuild')
if suffix:
new_lib = os.path.join(new_lib, suffix)
self.library_dirs.append(new_lib)
# For extensions under Cygwin, Python's library directory must be
# appended to library_dirs
if sys.platform[:6] == 'cygwin':
if sys.executable.startswith(os.path.join(sys.exec_prefix, "bin")):
# building third party extensions
self.library_dirs.append(os.path.join(sys.prefix, "lib",
"python" + get_python_version(),
"config"))
else:
# building python standard extensions
self.library_dirs.append('.')
# For building extensions with a shared Python library,
# Python's library directory must be appended to library_dirs
# See Issues: #1600860, #4366
if (sysconfig.get_config_var('Py_ENABLE_SHARED')):
if not sysconfig.python_build:
# building third party extensions
self.library_dirs.append(sysconfig.get_config_var('LIBDIR'))
else:
# building python standard extensions
self.library_dirs.append('.')
# The argument parsing will result in self.define being a string, but
# it has to be a list of 2-tuples. All the preprocessor symbols
# specified by the 'define' option will be set to '1'. Multiple
# symbols can be separated with commas.
if self.define:
defines = self.define.split(',')
self.define = [(symbol, '1') for symbol in defines]
# The option for macros to undefine is also a string from the
# option parsing, but has to be a list. Multiple symbols can also
# be separated with commas here.
if self.undef:
self.undef = self.undef.split(',')
if self.swig_opts is None:
self.swig_opts = []
else:
self.swig_opts = self.swig_opts.split(' ')
# Finally add the user include and library directories if requested
if self.user:
user_include = os.path.join(USER_BASE, "include")
user_lib = os.path.join(USER_BASE, "lib")
if os.path.isdir(user_include):
self.include_dirs.append(user_include)
if os.path.isdir(user_lib):
self.library_dirs.append(user_lib)
self.rpath.append(user_lib)
if isinstance(self.parallel, str):
try:
self.parallel = int(self.parallel)
except ValueError:
raise DistutilsOptionError("parallel should be an integer")
def run(self):
from distutils.ccompiler import new_compiler
# 'self.extensions', as supplied by setup.py, is a list of
# Extension instances. See the documentation for Extension (in
# distutils.extension) for details.
#
# For backwards compatibility with Distutils 0.8.2 and earlier, we
# also allow the 'extensions' list to be a list of tuples:
# (ext_name, build_info)
# where build_info is a dictionary containing everything that
# Extension instances do except the name, with a few things being
# differently named. We convert these 2-tuples to Extension
# instances as needed.
if not self.extensions:
return
# If we were asked to build any C/C++ libraries, make sure that the
# directory where we put them is in the library search path for
# linking extensions.
if self.distribution.has_c_libraries():
build_clib = self.get_finalized_command('build_clib')
self.libraries.extend(build_clib.get_library_names() or [])
self.library_dirs.append(build_clib.build_clib)
# Setup the CCompiler object that we'll use to do all the
# compiling and linking
self.compiler = new_compiler(compiler=self.compiler,
verbose=self.verbose,
dry_run=self.dry_run,
force=self.force)
customize_compiler(self.compiler)
# If we are cross-compiling, init the compiler now (if we are not
# cross-compiling, init would not hurt, but people may rely on
# late initialization of compiler even if they shouldn't...)
if os.name == 'nt' and self.plat_name != get_platform():
self.compiler.initialize(self.plat_name)
# And make sure that any compile/link-related options (which might
# come from the command-line or from the setup script) are set in
# that CCompiler object -- that way, they automatically apply to
# all compiling and linking done here.
if self.include_dirs is not None:
self.compiler.set_include_dirs(self.include_dirs)
if self.define is not None:
# 'define' option is a list of (name,value) tuples
for (name, value) in self.define:
self.compiler.define_macro(name, value)
if self.undef is not None:
for macro in self.undef:
self.compiler.undefine_macro(macro)
if self.libraries is not None:
self.compiler.set_libraries(self.libraries)
if self.library_dirs is not None:
self.compiler.set_library_dirs(self.library_dirs)
if self.rpath is not None:
self.compiler.set_runtime_library_dirs(self.rpath)
if self.link_objects is not None:
self.compiler.set_link_objects(self.link_objects)
# Now actually compile and link everything.
self.build_extensions()
def check_extensions_list(self, extensions):
"""Ensure that the list of extensions (presumably provided as a
command option 'extensions') is valid, i.e. it is a list of
Extension objects. We also support the old-style list of 2-tuples,
where the tuples are (ext_name, build_info), which are converted to
Extension instances here.
Raise DistutilsSetupError if the structure is invalid anywhere;
just returns otherwise.
"""
if not isinstance(extensions, list):
raise DistutilsSetupError(
"'ext_modules' option must be a list of Extension instances")
for i, ext in enumerate(extensions):
if isinstance(ext, Extension):
continue # OK! (assume type-checking done
# by Extension constructor)
if not isinstance(ext, tuple) or len(ext) != 2:
raise DistutilsSetupError(
"each element of 'ext_modules' option must be an "
"Extension instance or 2-tuple")
ext_name, build_info = ext
log.warn("old-style (ext_name, build_info) tuple found in "
"ext_modules for extension '%s' "
"-- please convert to Extension instance", ext_name)
if not (isinstance(ext_name, str) and
extension_name_re.match(ext_name)):
raise DistutilsSetupError(
"first element of each tuple in 'ext_modules' "
"must be the extension name (a string)")
if not isinstance(build_info, dict):
raise DistutilsSetupError(
"second element of each tuple in 'ext_modules' "
"must be a dictionary (build info)")
# OK, the (ext_name, build_info) dict is type-safe: convert it
# to an Extension instance.
ext = Extension(ext_name, build_info['sources'])
# Easy stuff: one-to-one mapping from dict elements to
# instance attributes.
for key in ('include_dirs', 'library_dirs', 'libraries',
'extra_objects', 'extra_compile_args',
'extra_link_args'):
val = build_info.get(key)
if val is not None:
setattr(ext, key, val)
# Medium-easy stuff: same syntax/semantics, different names.
ext.runtime_library_dirs = build_info.get('rpath')
if 'def_file' in build_info:
log.warn("'def_file' element of build info dict "
"no longer supported")
# Non-trivial stuff: 'macros' split into 'define_macros'
# and 'undef_macros'.
macros = build_info.get('macros')
if macros:
ext.define_macros = []
ext.undef_macros = []
for macro in macros:
if not (isinstance(macro, tuple) and len(macro) in (1, 2)):
raise DistutilsSetupError(
"'macros' element of build info dict "
"must be 1- or 2-tuple")
if len(macro) == 1:
ext.undef_macros.append(macro[0])
elif len(macro) == 2:
ext.define_macros.append(macro)
extensions[i] = ext
def get_source_files(self):
self.check_extensions_list(self.extensions)
filenames = []
# Wouldn't it be neat if we knew the names of header files too...
for ext in self.extensions:
filenames.extend(ext.sources)
return filenames
def get_outputs(self):
# Sanity check the 'extensions' list -- can't assume this is being
# done in the same run as a 'build_extensions()' call (in fact, we
# can probably assume that it *isn't*!).
self.check_extensions_list(self.extensions)
# And build the list of output (built) filenames. Note that this
# ignores the 'inplace' flag, and assumes everything goes in the
# "build" tree.
outputs = []
for ext in self.extensions:
outputs.append(self.get_ext_fullpath(ext.name))
return outputs
def build_extensions(self):
# First, sanity-check the 'extensions' list
self.check_extensions_list(self.extensions)
if self.parallel:
self._build_extensions_parallel()
else:
self._build_extensions_serial()
def _build_extensions_parallel(self):
workers = self.parallel
if self.parallel is True:
workers = os.cpu_count() # may return None
try:
from concurrent.futures import ThreadPoolExecutor
except ImportError:
workers = None
if workers is None:
self._build_extensions_serial()
return
with ThreadPoolExecutor(max_workers=workers) as executor:
futures = [executor.submit(self.build_extension, ext)
for ext in self.extensions]
for ext, fut in zip(self.extensions, futures):
with self._filter_build_errors(ext):
fut.result()
def _build_extensions_serial(self):
for ext in self.extensions:
with self._filter_build_errors(ext):
self.build_extension(ext)
@contextlib.contextmanager
def _filter_build_errors(self, ext):
try:
yield
except (CCompilerError, DistutilsError, CompileError) as e:
if not ext.optional:
raise
self.warn('building extension "%s" failed: %s' %
(ext.name, e))
def build_extension(self, ext):
sources = ext.sources
if sources is None or not isinstance(sources, (list, tuple)):
raise DistutilsSetupError(
"in 'ext_modules' option (extension '%s'), "
"'sources' must be present and must be "
"a list of source filenames" % ext.name)
# sort to make the resulting .so file build reproducible
sources = sorted(sources)
ext_path = self.get_ext_fullpath(ext.name)
depends = sources + ext.depends
if not (self.force or newer_group(depends, ext_path, 'newer')):
log.debug("skipping '%s' extension (up-to-date)", ext.name)
return
else:
log.info("building '%s' extension", ext.name)
# First, scan the sources for SWIG definition files (.i), run
# SWIG on 'em to create .c files, and modify the sources list
# accordingly.
sources = self.swig_sources(sources, ext)
# Next, compile the source code to object files.
# XXX not honouring 'define_macros' or 'undef_macros' -- the
# CCompiler API needs to change to accommodate this, and I
# want to do one thing at a time!
# Two possible sources for extra compiler arguments:
# - 'extra_compile_args' in Extension object
# - CFLAGS environment variable (not particularly
# elegant, but people seem to expect it and I
# guess it's useful)
# The environment variable should take precedence, and
# any sensible compiler will give precedence to later
# command line args. Hence we combine them in order:
extra_args = ext.extra_compile_args or []
macros = ext.define_macros[:]
for undef in ext.undef_macros:
macros.append((undef,))
objects = self.compiler.compile(sources,
output_dir=self.build_temp,
macros=macros,
include_dirs=ext.include_dirs,
debug=self.debug,
extra_postargs=extra_args,
depends=ext.depends)
# XXX outdated variable, kept here in case third-part code
# needs it.
self._built_objects = objects[:]
# Now link the object files together into a "shared object" --
# of course, first we have to figure out all the other things
# that go into the mix.
if ext.extra_objects:
objects.extend(ext.extra_objects)
extra_args = ext.extra_link_args or []
# Detect target language, if not provided
language = ext.language or self.compiler.detect_language(sources)
self.compiler.link_shared_object(
objects, ext_path,
libraries=self.get_libraries(ext),
library_dirs=ext.library_dirs,
runtime_library_dirs=ext.runtime_library_dirs,
extra_postargs=extra_args,
export_symbols=self.get_export_symbols(ext),
debug=self.debug,
build_temp=self.build_temp,
target_lang=language)
def swig_sources(self, sources, extension):
"""Walk the list of source files in 'sources', looking for SWIG
interface (.i) files. Run SWIG on all that are found, and
return a modified 'sources' list with SWIG source files replaced
by the generated C (or C++) files.
"""
new_sources = []
swig_sources = []
swig_targets = {}
# XXX this drops generated C/C++ files into the source tree, which
# is fine for developers who want to distribute the generated
# source -- but there should be an option to put SWIG output in
# the temp dir.
if self.swig_cpp:
log.warn("--swig-cpp is deprecated - use --swig-opts=-c++")
if self.swig_cpp or ('-c++' in self.swig_opts) or \
('-c++' in extension.swig_opts):
target_ext = '.cpp'
else:
target_ext = '.c'
for source in sources:
(base, ext) = os.path.splitext(source)
if ext == ".i": # SWIG interface file
new_sources.append(base + '_wrap' + target_ext)
swig_sources.append(source)
swig_targets[source] = new_sources[-1]
else:
new_sources.append(source)
if not swig_sources:
return new_sources
swig = self.swig or self.find_swig()
swig_cmd = [swig, "-python"]
swig_cmd.extend(self.swig_opts)
if self.swig_cpp:
swig_cmd.append("-c++")
# Do not override commandline arguments
if not self.swig_opts:
for o in extension.swig_opts:
swig_cmd.append(o)
for source in swig_sources:
target = swig_targets[source]
log.info("swigging %s to %s", source, target)
self.spawn(swig_cmd + ["-o", target, source])
return new_sources
def find_swig(self):
"""Return the name of the SWIG executable. On Unix, this is
just "swig" -- it should be in the PATH. Tries a bit harder on
Windows.
"""
if os.name == "posix":
return "swig"
elif os.name == "nt":
# Look for SWIG in its standard installation directory on
# Windows (or so I presume!). If we find it there, great;
# if not, act like Unix and assume it's in the PATH.
for vers in ("1.3", "1.2", "1.1"):
fn = os.path.join("c:\\swig%s" % vers, "swig.exe")
if os.path.isfile(fn):
return fn
else:
return "swig.exe"
else:
raise DistutilsPlatformError(
"I don't know how to find (much less run) SWIG "
"on platform '%s'" % os.name)
# -- Name generators -----------------------------------------------
# (extension names, filenames, whatever)
def get_ext_fullpath(self, ext_name):
"""Returns the path of the filename for a given extension.
The file is located in `build_lib` or directly in the package
(inplace option).
"""
fullname = self.get_ext_fullname(ext_name)
modpath = fullname.split('.')
filename = self.get_ext_filename(modpath[-1])
if not self.inplace:
# no further work needed
# returning :
# build_dir/package/path/filename
filename = os.path.join(*modpath[:-1]+[filename])
return os.path.join(self.build_lib, filename)
# the inplace option requires to find the package directory
# using the build_py command for that
package = '.'.join(modpath[0:-1])
build_py = self.get_finalized_command('build_py')
package_dir = os.path.abspath(build_py.get_package_dir(package))
# returning
# package_dir/filename
return os.path.join(package_dir, filename)
def get_ext_fullname(self, ext_name):
"""Returns the fullname of a given extension name.
Adds the `package.` prefix"""
if self.package is None:
return ext_name
else:
return self.package + '.' + ext_name
def get_ext_filename(self, ext_name):
r"""Convert the name of an extension (eg. "foo.bar") into the name
of the file from which it will be loaded (eg. "foo/bar.so", or
"foo\bar.pyd").
"""
from distutils.sysconfig import get_config_var
ext_path = ext_name.split('.')
ext_suffix = get_config_var('EXT_SUFFIX')
return os.path.join(*ext_path) + ext_suffix
def get_export_symbols(self, ext):
"""Return the list of symbols that a shared extension has to
export. This either uses 'ext.export_symbols' or, if it's not
provided, "PyInit_" + module_name. Only relevant on Windows, where
the .pyd file (DLL) must export the module "PyInit_" function.
"""
initfunc_name = "PyInit_" + ext.name.split('.')[-1]
if initfunc_name not in ext.export_symbols:
ext.export_symbols.append(initfunc_name)
return ext.export_symbols
def get_libraries(self, ext):
"""Return the list of libraries to link against when building a
shared extension. On most platforms, this is just 'ext.libraries';
on Windows, we add the Python library (eg. python20.dll).
"""
# The python library is always needed on Windows. For MSVC, this
# is redundant, since the library is mentioned in a pragma in
# pyconfig.h that MSVC groks. The other Windows compilers all seem
# to need it mentioned explicitly, though, so that's what we do.
# Append '_d' to the python import library on debug builds.
if sys.platform == "win32":
from distutils._msvccompiler import MSVCCompiler
if not isinstance(self.compiler, MSVCCompiler):
template = "python%d%d"
if self.debug:
template = template + '_d'
pythonlib = (template %
(sys.hexversion >> 24, (sys.hexversion >> 16) & 0xff))
# don't extend ext.libraries, it may be shared with other
# extensions, it is a reference to the original list
return ext.libraries + [pythonlib]
else:
# On Android only the main executable and LD_PRELOADs are considered
# to be RTLD_GLOBAL, all the dependencies of the main executable
# remain RTLD_LOCAL and so the shared libraries must be linked with
# libpython when python is built with a shared python library (issue
# bpo-21536).
# On Cygwin (and if required, other POSIX-like platforms based on
# Windows like MinGW) it is simply necessary that all symbols in
# shared libraries are resolved at link time.
from distutils.sysconfig import get_config_var
link_libpython = False
if get_config_var('Py_ENABLE_SHARED'):
# A native build on an Android device or on Cygwin
if hasattr(sys, 'getandroidapilevel'):
link_libpython = True
elif sys.platform == 'cygwin':
link_libpython = True
elif '_PYTHON_HOST_PLATFORM' in os.environ:
# We are cross-compiling for one of the relevant platforms
if get_config_var('ANDROID_API_LEVEL') != 0:
link_libpython = True
elif get_config_var('MACHDEP') == 'cygwin':
link_libpython = True
if link_libpython:
ldversion = get_config_var('LDVERSION')
return ext.libraries + ['python' + ldversion]
return ext.libraries