|
|
|
# Copyright 2012-2017 The Meson development team
|
|
|
|
|
|
|
|
# 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.
|
|
|
|
|
|
|
|
import re
|
|
|
|
import glob
|
|
|
|
import os.path
|
|
|
|
import subprocess
|
|
|
|
import functools
|
|
|
|
import itertools
|
|
|
|
from pathlib import Path
|
|
|
|
from typing import List
|
|
|
|
|
|
|
|
from .. import mlog
|
|
|
|
from .. import coredata
|
|
|
|
from . import compilers
|
|
|
|
from ..mesonlib import (
|
|
|
|
EnvironmentException, MachineChoice, MesonException, Popen_safe, listify,
|
|
|
|
version_compare, for_windows, for_darwin, for_cygwin, for_haiku,
|
|
|
|
for_openbsd, darwin_get_object_archs
|
|
|
|
)
|
|
|
|
from .c_function_attributes import C_FUNC_ATTRIBUTES
|
|
|
|
|
|
|
|
from .compilers import (
|
|
|
|
get_largefile_args,
|
|
|
|
gnu_winlibs,
|
|
|
|
msvc_winlibs,
|
|
|
|
vs32_instruction_set_args,
|
|
|
|
vs64_instruction_set_args,
|
|
|
|
ArmCompiler,
|
|
|
|
ArmclangCompiler,
|
|
|
|
ClangCompiler,
|
|
|
|
Compiler,
|
|
|
|
CompilerArgs,
|
|
|
|
CompilerType,
|
|
|
|
CrossNoRunException,
|
|
|
|
GnuCompiler,
|
|
|
|
ElbrusCompiler,
|
|
|
|
IntelCompiler,
|
|
|
|
PGICompiler,
|
|
|
|
RunResult,
|
|
|
|
CcrxCompiler,
|
|
|
|
)
|
|
|
|
|
|
|
|
gnu_compiler_internal_libs = ('m', 'c', 'pthread', 'dl', 'rt')
|
|
|
|
|
|
|
|
|
|
|
|
class CCompiler(Compiler):
|
|
|
|
# TODO: Replace this manual cache with functools.lru_cache
|
|
|
|
library_dirs_cache = {}
|
|
|
|
program_dirs_cache = {}
|
|
|
|
find_library_cache = {}
|
|
|
|
find_framework_cache = {}
|
|
|
|
internal_libs = gnu_compiler_internal_libs
|
|
|
|
|
|
|
|
@staticmethod
|
|
|
|
def attribute_check_func(name):
|
|
|
|
try:
|
|
|
|
return C_FUNC_ATTRIBUTES[name]
|
|
|
|
except KeyError:
|
|
|
|
raise MesonException('Unknown function attribute "{}"'.format(name))
|
|
|
|
|
|
|
|
def __init__(self, exelist, version, is_cross, exe_wrapper=None, **kwargs):
|
|
|
|
# If a child ObjC or CPP class has already set it, don't set it ourselves
|
|
|
|
if not hasattr(self, 'language'):
|
|
|
|
self.language = 'c'
|
|
|
|
super().__init__(exelist, version, **kwargs)
|
|
|
|
self.id = 'unknown'
|
|
|
|
self.is_cross = is_cross
|
|
|
|
self.can_compile_suffixes.add('h')
|
|
|
|
# If the exe wrapper was not found, pretend it wasn't set so that the
|
|
|
|
# sanity check is skipped and compiler checks use fallbacks.
|
|
|
|
if not exe_wrapper or not exe_wrapper.found():
|
|
|
|
self.exe_wrapper = None
|
|
|
|
else:
|
|
|
|
self.exe_wrapper = exe_wrapper.get_command()
|
|
|
|
|
|
|
|
# Set to None until we actually need to check this
|
|
|
|
self.has_fatal_warnings_link_arg = None
|
|
|
|
|
|
|
|
def needs_static_linker(self):
|
|
|
|
return True # When compiling static libraries, so yes.
|
|
|
|
|
|
|
|
def get_always_args(self):
|
|
|
|
'''
|
|
|
|
Args that are always-on for all C compilers other than MSVC
|
|
|
|
'''
|
|
|
|
return ['-pipe'] + get_largefile_args(self)
|
|
|
|
|
|
|
|
def get_linker_debug_crt_args(self):
|
|
|
|
"""
|
|
|
|
Arguments needed to select a debug crt for the linker
|
|
|
|
This is only needed for MSVC
|
|
|
|
"""
|
|
|
|
return []
|
|
|
|
|
|
|
|
def get_no_stdinc_args(self):
|
|
|
|
return ['-nostdinc']
|
|
|
|
|
|
|
|
def get_no_stdlib_link_args(self):
|
|
|
|
return ['-nostdlib']
|
|
|
|
|
|
|
|
def get_warn_args(self, level):
|
|
|
|
return self.warn_args[level]
|
|
|
|
|
|
|
|
def get_no_warn_args(self):
|
|
|
|
# Almost every compiler uses this for disabling warnings
|
|
|
|
return ['-w']
|
|
|
|
|
macos: Rewrite install_name for dependent built libraries on install
On macOS, we set the install_name for built libraries to
@rpath/libfoo.dylib, and when linking to the library, we set the RPATH
to its path in the build directory. This allows all built binaries to
be run as-is from the build directory (uninstalled).
However, on install, we have to strip all the RPATHs because they
point to the build directory, and we change the install_name of all
built libraries to the absolute path to the library. This causes the
install name in binaries to be out of date.
We now change that install name to point to the absolute path to each
built library after installation.
Fixes https://github.com/mesonbuild/meson/issues/3038
Fixes https://github.com/mesonbuild/meson/issues/3077
With this, the default workflow on macOS matches what everyone seems
to do, including Autotools and CMake. The next step is providing a way
for build files to override the install_name that is used after
installation for use with, f.ex., private libraries when combined with
the install_rpath: kwarg on targets.
7 years ago
|
|
|
def get_soname_args(self, *args):
|
|
|
|
return []
|
|
|
|
|
|
|
|
def split_shlib_to_parts(self, fname):
|
|
|
|
return None, fname
|
|
|
|
|
|
|
|
# The default behavior is this, override in MSVC
|
|
|
|
@functools.lru_cache(maxsize=None)
|
|
|
|
def build_rpath_args(self, build_dir, from_dir, rpath_paths, build_rpath, install_rpath):
|
|
|
|
if self.compiler_type.is_windows_compiler:
|
|
|
|
return []
|
|
|
|
return self.build_unix_rpath_args(build_dir, from_dir, rpath_paths, build_rpath, install_rpath)
|
|
|
|
|
|
|
|
def get_dependency_gen_args(self, outtarget, outfile):
|
|
|
|
return ['-MD', '-MQ', outtarget, '-MF', outfile]
|
|
|
|
|
|
|
|
def depfile_for_object(self, objfile):
|
|
|
|
return objfile + '.' + self.get_depfile_suffix()
|
|
|
|
|
|
|
|
def get_depfile_suffix(self):
|
|
|
|
return 'd'
|
|
|
|
|
|
|
|
def get_exelist(self):
|
|
|
|
return self.exelist[:]
|
|
|
|
|
|
|
|
def get_linker_exelist(self):
|
|
|
|
return self.exelist[:]
|
|
|
|
|
|
|
|
def get_preprocess_only_args(self):
|
|
|
|
return ['-E', '-P']
|
|
|
|
|
|
|
|
def get_compile_only_args(self):
|
|
|
|
return ['-c']
|
|
|
|
|
|
|
|
def get_no_optimization_args(self):
|
|
|
|
return ['-O0']
|
|
|
|
|
|
|
|
def get_compiler_check_args(self):
|
|
|
|
'''
|
|
|
|
Get arguments useful for compiler checks such as being permissive in
|
|
|
|
the code quality and not doing any optimization.
|
|
|
|
'''
|
|
|
|
return self.get_no_optimization_args()
|
|
|
|
|
|
|
|
def get_output_args(self, target):
|
|
|
|
return ['-o', target]
|
|
|
|
|
|
|
|
def get_linker_output_args(self, outputname):
|
|
|
|
return ['-o', outputname]
|
|
|
|
|
|
|
|
def get_coverage_args(self):
|
|
|
|
return ['--coverage']
|
|
|
|
|
|
|
|
def get_coverage_link_args(self):
|
|
|
|
return ['--coverage']
|
|
|
|
|
|
|
|
def get_werror_args(self):
|
|
|
|
return ['-Werror']
|
|
|
|
|
|
|
|
def get_std_exe_link_args(self):
|
|
|
|
return []
|
|
|
|
|
|
|
|
def get_include_args(self, path, is_system):
|
|
|
|
if path == '':
|
|
|
|
path = '.'
|
|
|
|
if is_system:
|
|
|
|
return ['-isystem', path]
|
|
|
|
return ['-I' + path]
|
|
|
|
|
|
|
|
def get_std_shared_lib_link_args(self):
|
|
|
|
return ['-shared']
|
|
|
|
|
|
|
|
@functools.lru_cache()
|
|
|
|
def _get_search_dirs(self, env):
|
|
|
|
extra_args = ['--print-search-dirs']
|
|
|
|
stdo = None
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
with self._build_wrapper('', env, extra_args=extra_args,
|
|
|
|
dependencies=None, mode='compile',
|
|
|
|
want_output=True) as p:
|
|
|
|
stdo = p.stdo
|
|
|
|
return stdo
|
|
|
|
|
|
|
|
@staticmethod
|
|
|
|
def _split_fetch_real_dirs(pathstr, sep=':'):
|
|
|
|
paths = []
|
|
|
|
for p in pathstr.split(sep):
|
|
|
|
# GCC returns paths like this:
|
|
|
|
# /usr/lib/gcc/x86_64-linux-gnu/8/../../../../x86_64-linux-gnu/lib
|
|
|
|
# It would make sense to normalize them to get rid of the .. parts
|
|
|
|
# Sadly when you are on a merged /usr fs it also kills these:
|
|
|
|
# /lib/x86_64-linux-gnu
|
|
|
|
# since /lib is a symlink to /usr/lib. This would mean
|
|
|
|
# paths under /lib would be considered not a "system path",
|
|
|
|
# which is wrong and breaks things. Store everything, just to be sure.
|
|
|
|
pobj = Path(p)
|
|
|
|
unresolved = pobj.as_posix()
|
|
|
|
if pobj.exists():
|
|
|
|
if unresolved not in paths:
|
|
|
|
paths.append(unresolved)
|
|
|
|
try:
|
|
|
|
resolved = Path(p).resolve().as_posix()
|
|
|
|
if resolved not in paths:
|
|
|
|
paths.append(resolved)
|
|
|
|
except FileNotFoundError:
|
|
|
|
pass
|
|
|
|
return tuple(paths)
|
|
|
|
|
|
|
|
def get_compiler_dirs(self, env, name):
|
|
|
|
'''
|
|
|
|
Get dirs from the compiler, either `libraries:` or `programs:`
|
|
|
|
'''
|
|
|
|
stdo = self._get_search_dirs(env)
|
|
|
|
for line in stdo.split('\n'):
|
|
|
|
if line.startswith(name + ':'):
|
|
|
|
return CCompiler._split_fetch_real_dirs(line.split('=', 1)[1])
|
|
|
|
return ()
|
|
|
|
|
|
|
|
@functools.lru_cache()
|
|
|
|
def get_library_dirs(self, env, elf_class = None):
|
|
|
|
dirs = self.get_compiler_dirs(env, 'libraries')
|
|
|
|
if elf_class is None or elf_class == 0:
|
|
|
|
return dirs
|
|
|
|
|
|
|
|
# if we do have an elf class for 32-bit or 64-bit, we want to check that
|
|
|
|
# the directory in question contains libraries of the appropriate class. Since
|
|
|
|
# system directories aren't mixed, we only need to check one file for each
|
|
|
|
# directory and go by that. If we can't check the file for some reason, assume
|
|
|
|
# the compiler knows what it's doing, and accept the directory anyway.
|
|
|
|
retval = []
|
|
|
|
for d in dirs:
|
|
|
|
files = [f for f in os.listdir(d) if f.endswith('.so') and os.path.isfile(os.path.join(d, f))]
|
|
|
|
# if no files, accept directory and move on
|
|
|
|
if len(files) == 0:
|
|
|
|
retval.append(d)
|
|
|
|
continue
|
|
|
|
file_to_check = os.path.join(d, files[0])
|
|
|
|
with open(file_to_check, 'rb') as fd:
|
|
|
|
header = fd.read(5)
|
|
|
|
# if file is not an ELF file, it's weird, but accept dir
|
|
|
|
# if it is elf, and the class matches, accept dir
|
|
|
|
if header[1:4] != b'ELF' or int(header[4]) == elf_class:
|
|
|
|
retval.append(d)
|
|
|
|
# at this point, it's an ELF file which doesn't match the
|
|
|
|
# appropriate elf_class, so skip this one
|
|
|
|
pass
|
|
|
|
return tuple(retval)
|
|
|
|
|
|
|
|
@functools.lru_cache()
|
|
|
|
def get_program_dirs(self, env):
|
|
|
|
'''
|
|
|
|
Programs used by the compiler. Also where toolchain DLLs such as
|
|
|
|
libstdc++-6.dll are found with MinGW.
|
|
|
|
'''
|
|
|
|
return self.get_compiler_dirs(env, 'programs')
|
|
|
|
|
|
|
|
def get_pic_args(self):
|
|
|
|
return ['-fPIC']
|
|
|
|
|
|
|
|
def name_string(self):
|
|
|
|
return ' '.join(self.exelist)
|
|
|
|
|
|
|
|
def get_pch_use_args(self, pch_dir, header):
|
|
|
|
return ['-include', os.path.basename(header)]
|
|
|
|
|
|
|
|
def get_pch_name(self, header_name):
|
|
|
|
return os.path.basename(header_name) + '.' + self.get_pch_suffix()
|
|
|
|
|
|
|
|
def get_linker_search_args(self, dirname):
|
|
|
|
return ['-L' + dirname]
|
|
|
|
|
|
|
|
def get_default_include_dirs(self):
|
|
|
|
return []
|
|
|
|
|
|
|
|
def gen_export_dynamic_link_args(self, env):
|
|
|
|
if for_windows(env.is_cross_build(), env) or for_cygwin(env.is_cross_build(), env):
|
|
|
|
return ['-Wl,--export-all-symbols']
|
|
|
|
elif for_darwin(env.is_cross_build(), env):
|
|
|
|
return []
|
|
|
|
else:
|
|
|
|
return ['-Wl,-export-dynamic']
|
|
|
|
|
|
|
|
def gen_import_library_args(self, implibname):
|
|
|
|
"""
|
|
|
|
The name of the outputted import library
|
|
|
|
|
|
|
|
This implementation is used only on Windows by compilers that use GNU ld
|
|
|
|
"""
|
|
|
|
return ['-Wl,--out-implib=' + implibname]
|
|
|
|
|
|
|
|
def sanity_check_impl(self, work_dir, environment, sname, code):
|
|
|
|
mlog.debug('Sanity testing ' + self.get_display_language() + ' compiler:', ' '.join(self.exelist))
|
|
|
|
mlog.debug('Is cross compiler: %s.' % str(self.is_cross))
|
|
|
|
|
|
|
|
extra_flags = []
|
|
|
|
source_name = os.path.join(work_dir, sname)
|
|
|
|
binname = sname.rsplit('.', 1)[0]
|
|
|
|
if self.is_cross:
|
|
|
|
binname += '_cross'
|
|
|
|
if self.exe_wrapper is None:
|
|
|
|
# Linking cross built apps is painful. You can't really
|
|
|
|
# tell if you should use -nostdlib or not and for example
|
|
|
|
# on OSX the compiler binary is the same but you need
|
|
|
|
# a ton of compiler flags to differentiate between
|
|
|
|
# arm and x86_64. So just compile.
|
|
|
|
extra_flags += self.get_compile_only_args()
|
|
|
|
# Is a valid executable output for all toolchains and platforms
|
|
|
|
binname += '.exe'
|
|
|
|
# Write binary check source
|
|
|
|
binary_name = os.path.join(work_dir, binname)
|
|
|
|
with open(source_name, 'w') as ofile:
|
|
|
|
ofile.write(code)
|
|
|
|
# Compile sanity check
|
|
|
|
cmdlist = self.exelist + extra_flags + [source_name] + self.get_output_args(binary_name)
|
|
|
|
pc, stdo, stde = Popen_safe(cmdlist, cwd=work_dir)
|
|
|
|
mlog.debug('Sanity check compiler command line:', ' '.join(cmdlist))
|
|
|
|
mlog.debug('Sanity check compile stdout:')
|
|
|
|
mlog.debug(stdo)
|
|
|
|
mlog.debug('-----\nSanity check compile stderr:')
|
|
|
|
mlog.debug(stde)
|
|
|
|
mlog.debug('-----')
|
|
|
|
if pc.returncode != 0:
|
|
|
|
raise EnvironmentException('Compiler {0} can not compile programs.'.format(self.name_string()))
|
|
|
|
# Run sanity check
|
|
|
|
if self.is_cross:
|
|
|
|
if self.exe_wrapper is None:
|
|
|
|
# Can't check if the binaries run so we have to assume they do
|
|
|
|
return
|
|
|
|
cmdlist = self.exe_wrapper + [binary_name]
|
|
|
|
else:
|
|
|
|
cmdlist = [binary_name]
|
|
|
|
mlog.debug('Running test binary command: ' + ' '.join(cmdlist))
|
|
|
|
try:
|
|
|
|
pe = subprocess.Popen(cmdlist)
|
|
|
|
except Exception as e:
|
|
|
|
raise EnvironmentException('Could not invoke sanity test executable: %s.' % str(e))
|
|
|
|
pe.wait()
|
|
|
|
if pe.returncode != 0:
|
|
|
|
raise EnvironmentException('Executables created by {0} compiler {1} are not runnable.'.format(self.language, self.name_string()))
|
|
|
|
|
|
|
|
def sanity_check(self, work_dir, environment):
|
|
|
|
code = 'int main(int argc, char **argv) { int class=0; return class; }\n'
|
|
|
|
return self.sanity_check_impl(work_dir, environment, 'sanitycheckc.c', code)
|
|
|
|
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
def check_header(self, hname, prefix, env, *, extra_args=None, dependencies=None):
|
|
|
|
fargs = {'prefix': prefix, 'header': hname}
|
|
|
|
code = '''{prefix}
|
|
|
|
#include <{header}>'''
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
return self.compiles(code.format(**fargs), env, extra_args=extra_args,
|
|
|
|
dependencies=dependencies)
|
|
|
|
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
def has_header(self, hname, prefix, env, *, extra_args=None, dependencies=None):
|
|
|
|
fargs = {'prefix': prefix, 'header': hname}
|
|
|
|
code = '''{prefix}
|
|
|
|
#ifdef __has_include
|
|
|
|
#if !__has_include("{header}")
|
|
|
|
#error "Header '{header}' could not be found"
|
|
|
|
#endif
|
|
|
|
#else
|
|
|
|
#include <{header}>
|
|
|
|
#endif'''
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
return self.compiles(code.format(**fargs), env, extra_args=extra_args,
|
|
|
|
dependencies=dependencies, mode='preprocess')
|
|
|
|
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
def has_header_symbol(self, hname, symbol, prefix, env, *, extra_args=None, dependencies=None):
|
|
|
|
fargs = {'prefix': prefix, 'header': hname, 'symbol': symbol}
|
|
|
|
t = '''{prefix}
|
|
|
|
#include <{header}>
|
|
|
|
int main () {{
|
|
|
|
/* If it's not defined as a macro, try to use as a symbol */
|
|
|
|
#ifndef {symbol}
|
|
|
|
{symbol};
|
|
|
|
#endif
|
|
|
|
}}'''
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
return self.compiles(t.format(**fargs), env, extra_args=extra_args,
|
|
|
|
dependencies=dependencies)
|
|
|
|
|
|
|
|
def _get_compiler_check_args(self, env, extra_args, dependencies, mode='compile'):
|
|
|
|
if extra_args is None:
|
|
|
|
extra_args = []
|
|
|
|
else:
|
|
|
|
extra_args = listify(extra_args)
|
|
|
|
extra_args = listify([e(mode) if callable(e) else e for e in extra_args])
|
|
|
|
|
|
|
|
if dependencies is None:
|
|
|
|
dependencies = []
|
|
|
|
elif not isinstance(dependencies, list):
|
|
|
|
dependencies = [dependencies]
|
|
|
|
# Collect compiler arguments
|
|
|
|
args = CompilerArgs(self)
|
|
|
|
for d in dependencies:
|
|
|
|
# Add compile flags needed by dependencies
|
|
|
|
args += d.get_compile_args()
|
|
|
|
if d.need_threads():
|
|
|
|
args += self.thread_flags(env)
|
|
|
|
elif d.need_openmp():
|
|
|
|
args += self.openmp_flags()
|
|
|
|
if mode == 'link':
|
|
|
|
# Add link flags needed to find dependencies
|
|
|
|
args += d.get_link_args()
|
|
|
|
if d.need_threads():
|
|
|
|
args += self.thread_link_flags(env)
|
|
|
|
# Select a CRT if needed since we're linking
|
|
|
|
if mode == 'link':
|
|
|
|
args += self.get_linker_debug_crt_args()
|
|
|
|
if env.is_cross_build() and not self.is_cross:
|
|
|
|
for_machine = MachineChoice.BUILD
|
|
|
|
else:
|
|
|
|
for_machine = MachineChoice.HOST
|
|
|
|
if mode == 'preprocess':
|
|
|
|
# Add CPPFLAGS from the env.
|
|
|
|
args += env.coredata.get_external_preprocess_args(for_machine, self.language)
|
|
|
|
elif mode == 'compile':
|
|
|
|
# Add CFLAGS/CXXFLAGS/OBJCFLAGS/OBJCXXFLAGS from the env
|
|
|
|
sys_args = env.coredata.get_external_args(for_machine, self.language)
|
|
|
|
# Apparently it is a thing to inject linker flags both
|
|
|
|
# via CFLAGS _and_ LDFLAGS, even though the former are
|
|
|
|
# also used during linking. These flags can break
|
|
|
|
# argument checks. Thanks, Autotools.
|
|
|
|
cleaned_sys_args = self.remove_linkerlike_args(sys_args)
|
|
|
|
args += cleaned_sys_args
|
|
|
|
elif mode == 'link':
|
|
|
|
# Add LDFLAGS from the env
|
|
|
|
args += env.coredata.get_external_link_args(for_machine, self.language)
|
|
|
|
args += self.get_compiler_check_args()
|
|
|
|
# extra_args must override all other arguments, so we add them last
|
|
|
|
args += extra_args
|
|
|
|
return args
|
|
|
|
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
def compiles(self, code, env, *, extra_args=None, dependencies=None, mode='compile'):
|
|
|
|
with self._build_wrapper(code, env, extra_args, dependencies, mode) as p:
|
|
|
|
return p.returncode == 0
|
|
|
|
|
|
|
|
def _build_wrapper(self, code, env, extra_args, dependencies=None, mode='compile', want_output=False):
|
|
|
|
args = self._get_compiler_check_args(env, extra_args, dependencies, mode)
|
|
|
|
return self.compile(code, args, mode, want_output=want_output)
|
|
|
|
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
def links(self, code, env, *, extra_args=None, dependencies=None):
|
|
|
|
return self.compiles(code, env, extra_args=extra_args,
|
|
|
|
dependencies=dependencies, mode='link')
|
|
|
|
|
|
|
|
def run(self, code: str, env, *, extra_args=None, dependencies=None):
|
|
|
|
if self.is_cross and self.exe_wrapper is None:
|
|
|
|
raise CrossNoRunException('Can not run test applications in this cross environment.')
|
|
|
|
with self._build_wrapper(code, env, extra_args, dependencies, mode='link', want_output=True) as p:
|
|
|
|
if p.returncode != 0:
|
|
|
|
mlog.debug('Could not compile test file %s: %d\n' % (
|
|
|
|
p.input_name,
|
|
|
|
p.returncode))
|
|
|
|
return RunResult(False)
|
|
|
|
if self.is_cross:
|
|
|
|
cmdlist = self.exe_wrapper + [p.output_name]
|
|
|
|
else:
|
|
|
|
cmdlist = p.output_name
|
|
|
|
try:
|
|
|
|
pe, so, se = Popen_safe(cmdlist)
|
|
|
|
except Exception as e:
|
|
|
|
mlog.debug('Could not run: %s (error: %s)\n' % (cmdlist, e))
|
|
|
|
return RunResult(False)
|
|
|
|
|
|
|
|
mlog.debug('Program stdout:\n')
|
|
|
|
mlog.debug(so)
|
|
|
|
mlog.debug('Program stderr:\n')
|
|
|
|
mlog.debug(se)
|
|
|
|
return RunResult(True, pe.returncode, so, se)
|
|
|
|
|
|
|
|
def _compile_int(self, expression, prefix, env, extra_args, dependencies):
|
|
|
|
fargs = {'prefix': prefix, 'expression': expression}
|
|
|
|
t = '''#include <stdio.h>
|
|
|
|
{prefix}
|
|
|
|
int main() {{ static int a[1-2*!({expression})]; a[0]=0; return 0; }}'''
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
return self.compiles(t.format(**fargs), env, extra_args=extra_args,
|
|
|
|
dependencies=dependencies)
|
|
|
|
|
|
|
|
def cross_compute_int(self, expression, low, high, guess, prefix, env, extra_args, dependencies):
|
|
|
|
# Try user's guess first
|
|
|
|
if isinstance(guess, int):
|
|
|
|
if self._compile_int('%s == %d' % (expression, guess), prefix, env, extra_args, dependencies):
|
|
|
|
return guess
|
|
|
|
|
|
|
|
# If no bounds are given, compute them in the limit of int32
|
|
|
|
maxint = 0x7fffffff
|
|
|
|
minint = -0x80000000
|
|
|
|
if not isinstance(low, int) or not isinstance(high, int):
|
|
|
|
if self._compile_int('%s >= 0' % (expression), prefix, env, extra_args, dependencies):
|
|
|
|
low = cur = 0
|
|
|
|
while self._compile_int('%s > %d' % (expression, cur), prefix, env, extra_args, dependencies):
|
|
|
|
low = cur + 1
|
|
|
|
if low > maxint:
|
|
|
|
raise EnvironmentException('Cross-compile check overflowed')
|
|
|
|
cur = cur * 2 + 1
|
|
|
|
if cur > maxint:
|
|
|
|
cur = maxint
|
|
|
|
high = cur
|
|
|
|
else:
|
|
|
|
low = cur = -1
|
|
|
|
while self._compile_int('%s < %d' % (expression, cur), prefix, env, extra_args, dependencies):
|
|
|
|
high = cur - 1
|
|
|
|
if high < minint:
|
|
|
|
raise EnvironmentException('Cross-compile check overflowed')
|
|
|
|
cur = cur * 2
|
|
|
|
if cur < minint:
|
|
|
|
cur = minint
|
|
|
|
low = cur
|
|
|
|
else:
|
|
|
|
# Sanity check limits given by user
|
|
|
|
if high < low:
|
|
|
|
raise EnvironmentException('high limit smaller than low limit')
|
|
|
|
condition = '%s <= %d && %s >= %d' % (expression, high, expression, low)
|
|
|
|
if not self._compile_int(condition, prefix, env, extra_args, dependencies):
|
|
|
|
raise EnvironmentException('Value out of given range')
|
|
|
|
|
|
|
|
# Binary search
|
|
|
|
while low != high:
|
|
|
|
cur = low + int((high - low) / 2)
|
|
|
|
if self._compile_int('%s <= %d' % (expression, cur), prefix, env, extra_args, dependencies):
|
|
|
|
high = cur
|
|
|
|
else:
|
|
|
|
low = cur + 1
|
|
|
|
|
|
|
|
return low
|
|
|
|
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
def compute_int(self, expression, low, high, guess, prefix, env, *, extra_args=None, dependencies=None):
|
|
|
|
if extra_args is None:
|
|
|
|
extra_args = []
|
|
|
|
if self.is_cross:
|
|
|
|
return self.cross_compute_int(expression, low, high, guess, prefix, env, extra_args, dependencies)
|
|
|
|
fargs = {'prefix': prefix, 'expression': expression}
|
|
|
|
t = '''#include<stdio.h>
|
|
|
|
{prefix}
|
|
|
|
int main(int argc, char **argv) {{
|
|
|
|
printf("%ld\\n", (long)({expression}));
|
|
|
|
return 0;
|
|
|
|
}};'''
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
res = self.run(t.format(**fargs), env, extra_args=extra_args,
|
|
|
|
dependencies=dependencies)
|
|
|
|
if not res.compiled:
|
|
|
|
return -1
|
|
|
|
if res.returncode != 0:
|
|
|
|
raise EnvironmentException('Could not run compute_int test binary.')
|
|
|
|
return int(res.stdout)
|
|
|
|
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
def cross_sizeof(self, typename, prefix, env, *, extra_args=None, dependencies=None):
|
|
|
|
if extra_args is None:
|
|
|
|
extra_args = []
|
|
|
|
fargs = {'prefix': prefix, 'type': typename}
|
|
|
|
t = '''#include <stdio.h>
|
|
|
|
{prefix}
|
|
|
|
int main(int argc, char **argv) {{
|
|
|
|
{type} something;
|
|
|
|
}}'''
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
if not self.compiles(t.format(**fargs), env, extra_args=extra_args,
|
|
|
|
dependencies=dependencies):
|
|
|
|
return -1
|
|
|
|
return self.cross_compute_int('sizeof(%s)' % typename, None, None, None, prefix, env, extra_args, dependencies)
|
|
|
|
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
def sizeof(self, typename, prefix, env, *, extra_args=None, dependencies=None):
|
|
|
|
if extra_args is None:
|
|
|
|
extra_args = []
|
|
|
|
fargs = {'prefix': prefix, 'type': typename}
|
|
|
|
if self.is_cross:
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
return self.cross_sizeof(typename, prefix, env, extra_args=extra_args,
|
|
|
|
dependencies=dependencies)
|
|
|
|
t = '''#include<stdio.h>
|
|
|
|
{prefix}
|
|
|
|
int main(int argc, char **argv) {{
|
|
|
|
printf("%ld\\n", (long)(sizeof({type})));
|
|
|
|
return 0;
|
|
|
|
}};'''
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
res = self.run(t.format(**fargs), env, extra_args=extra_args,
|
|
|
|
dependencies=dependencies)
|
|
|
|
if not res.compiled:
|
|
|
|
return -1
|
|
|
|
if res.returncode != 0:
|
|
|
|
raise EnvironmentException('Could not run sizeof test binary.')
|
|
|
|
return int(res.stdout)
|
|
|
|
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
def cross_alignment(self, typename, prefix, env, *, extra_args=None, dependencies=None):
|
|
|
|
if extra_args is None:
|
|
|
|
extra_args = []
|
|
|
|
fargs = {'prefix': prefix, 'type': typename}
|
|
|
|
t = '''#include <stdio.h>
|
|
|
|
{prefix}
|
|
|
|
int main(int argc, char **argv) {{
|
|
|
|
{type} something;
|
|
|
|
}}'''
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
if not self.compiles(t.format(**fargs), env, extra_args=extra_args,
|
|
|
|
dependencies=dependencies):
|
|
|
|
return -1
|
|
|
|
t = '''#include <stddef.h>
|
|
|
|
{prefix}
|
|
|
|
struct tmp {{
|
|
|
|
char c;
|
|
|
|
{type} target;
|
|
|
|
}};'''
|
|
|
|
return self.cross_compute_int('offsetof(struct tmp, target)', None, None, None, t.format(**fargs), env, extra_args, dependencies)
|
|
|
|
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
def alignment(self, typename, prefix, env, *, extra_args=None, dependencies=None):
|
|
|
|
if extra_args is None:
|
|
|
|
extra_args = []
|
|
|
|
if self.is_cross:
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
return self.cross_alignment(typename, prefix, env, extra_args=extra_args,
|
|
|
|
dependencies=dependencies)
|
|
|
|
fargs = {'prefix': prefix, 'type': typename}
|
|
|
|
t = '''#include <stdio.h>
|
|
|
|
#include <stddef.h>
|
|
|
|
{prefix}
|
|
|
|
struct tmp {{
|
|
|
|
char c;
|
|
|
|
{type} target;
|
|
|
|
}};
|
|
|
|
int main(int argc, char **argv) {{
|
|
|
|
printf("%d", (int)offsetof(struct tmp, target));
|
|
|
|
return 0;
|
|
|
|
}}'''
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
res = self.run(t.format(**fargs), env, extra_args=extra_args,
|
|
|
|
dependencies=dependencies)
|
|
|
|
if not res.compiled:
|
|
|
|
raise EnvironmentException('Could not compile alignment test.')
|
|
|
|
if res.returncode != 0:
|
|
|
|
raise EnvironmentException('Could not run alignment test binary.')
|
|
|
|
align = int(res.stdout)
|
|
|
|
if align == 0:
|
|
|
|
raise EnvironmentException('Could not determine alignment of %s. Sorry. You might want to file a bug.' % typename)
|
|
|
|
return align
|
|
|
|
|
|
|
|
def get_define(self, dname, prefix, env, extra_args, dependencies):
|
|
|
|
delim = '"MESON_GET_DEFINE_DELIMITER"'
|
|
|
|
fargs = {'prefix': prefix, 'define': dname, 'delim': delim}
|
|
|
|
code = '''
|
|
|
|
{prefix}
|
|
|
|
#ifndef {define}
|
|
|
|
# define {define}
|
|
|
|
#endif
|
|
|
|
{delim}\n{define}'''
|
|
|
|
args = self._get_compiler_check_args(env, extra_args, dependencies,
|
|
|
|
mode='preprocess').to_native()
|
|
|
|
with self.compile(code.format(**fargs), args, 'preprocess') as p:
|
|
|
|
if p.returncode != 0:
|
|
|
|
raise EnvironmentException('Could not get define {!r}'.format(dname))
|
|
|
|
# Get the preprocessed value after the delimiter,
|
|
|
|
# minus the extra newline at the end and
|
|
|
|
# merge string literals.
|
|
|
|
return CCompiler.concatenate_string_literals(p.stdo.split(delim + '\n')[-1][:-1])
|
|
|
|
|
|
|
|
def get_return_value(self, fname, rtype, prefix, env, extra_args, dependencies):
|
|
|
|
if rtype == 'string':
|
|
|
|
fmt = '%s'
|
|
|
|
cast = '(char*)'
|
|
|
|
elif rtype == 'int':
|
|
|
|
fmt = '%lli'
|
|
|
|
cast = '(long long int)'
|
|
|
|
else:
|
|
|
|
raise AssertionError('BUG: Unknown return type {!r}'.format(rtype))
|
|
|
|
fargs = {'prefix': prefix, 'f': fname, 'cast': cast, 'fmt': fmt}
|
|
|
|
code = '''{prefix}
|
|
|
|
#include <stdio.h>
|
|
|
|
int main(int argc, char *argv[]) {{
|
|
|
|
printf ("{fmt}", {cast} {f}());
|
|
|
|
}}'''.format(**fargs)
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
res = self.run(code, env, extra_args=extra_args, dependencies=dependencies)
|
|
|
|
if not res.compiled:
|
|
|
|
m = 'Could not get return value of {}()'
|
|
|
|
raise EnvironmentException(m.format(fname))
|
|
|
|
if rtype == 'string':
|
|
|
|
return res.stdout
|
|
|
|
elif rtype == 'int':
|
|
|
|
try:
|
|
|
|
return int(res.stdout.strip())
|
|
|
|
except ValueError:
|
|
|
|
m = 'Return value of {}() is not an int'
|
|
|
|
raise EnvironmentException(m.format(fname))
|
|
|
|
|
|
|
|
@staticmethod
|
|
|
|
def _no_prototype_templ():
|
|
|
|
"""
|
|
|
|
Try to find the function without a prototype from a header by defining
|
|
|
|
our own dummy prototype and trying to link with the C library (and
|
|
|
|
whatever else the compiler links in by default). This is very similar
|
|
|
|
to the check performed by Autoconf for AC_CHECK_FUNCS.
|
|
|
|
"""
|
|
|
|
# Define the symbol to something else since it is defined by the
|
|
|
|
# includes or defines listed by the user or by the compiler. This may
|
|
|
|
# include, for instance _GNU_SOURCE which must be defined before
|
|
|
|
# limits.h, which includes features.h
|
|
|
|
# Then, undef the symbol to get rid of it completely.
|
|
|
|
head = '''
|
|
|
|
#define {func} meson_disable_define_of_{func}
|
|
|
|
{prefix}
|
|
|
|
#include <limits.h>
|
|
|
|
#undef {func}
|
|
|
|
'''
|
|
|
|
# Override any GCC internal prototype and declare our own definition for
|
|
|
|
# the symbol. Use char because that's unlikely to be an actual return
|
|
|
|
# value for a function which ensures that we override the definition.
|
|
|
|
head += '''
|
|
|
|
#ifdef __cplusplus
|
|
|
|
extern "C"
|
|
|
|
#endif
|
|
|
|
char {func} ();
|
|
|
|
'''
|
|
|
|
# The actual function call
|
|
|
|
main = '''
|
|
|
|
int main () {{
|
|
|
|
return {func} ();
|
|
|
|
}}'''
|
|
|
|
return head, main
|
|
|
|
|
|
|
|
@staticmethod
|
|
|
|
def _have_prototype_templ():
|
|
|
|
"""
|
|
|
|
Returns a head-er and main() call that uses the headers listed by the
|
|
|
|
user for the function prototype while checking if a function exists.
|
|
|
|
"""
|
|
|
|
# Add the 'prefix', aka defines, includes, etc that the user provides
|
|
|
|
# This may include, for instance _GNU_SOURCE which must be defined
|
|
|
|
# before limits.h, which includes features.h
|
|
|
|
head = '{prefix}\n#include <limits.h>\n'
|
|
|
|
# We don't know what the function takes or returns, so return it as an int.
|
|
|
|
# Just taking the address or comparing it to void is not enough because
|
|
|
|
# compilers are smart enough to optimize it away. The resulting binary
|
|
|
|
# is not run so we don't care what the return value is.
|
|
|
|
main = '''\nint main() {{
|
|
|
|
void *a = (void*) &{func};
|
|
|
|
long b = (long) a;
|
|
|
|
return (int) b;
|
|
|
|
}}'''
|
|
|
|
return head, main
|
|
|
|
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
def has_function(self, funcname, prefix, env, *, extra_args=None, dependencies=None):
|
|
|
|
"""
|
|
|
|
First, this function looks for the symbol in the default libraries
|
|
|
|
provided by the compiler (stdlib + a few others usually). If that
|
|
|
|
fails, it checks if any of the headers specified in the prefix provide
|
|
|
|
an implementation of the function, and if that fails, it checks if it's
|
|
|
|
implemented as a compiler-builtin.
|
|
|
|
"""
|
|
|
|
if extra_args is None:
|
|
|
|
extra_args = []
|
|
|
|
|
|
|
|
# Short-circuit if the check is already provided by the cross-info file
|
|
|
|
varname = 'has function ' + funcname
|
|
|
|
varname = varname.replace(' ', '_')
|
|
|
|
if self.is_cross:
|
|
|
|
val = env.properties.host.get(varname, None)
|
|
|
|
if val is not None:
|
|
|
|
if isinstance(val, bool):
|
|
|
|
return val
|
|
|
|
raise EnvironmentException('Cross variable {0} is not a boolean.'.format(varname))
|
|
|
|
|
|
|
|
fargs = {'prefix': prefix, 'func': funcname}
|
|
|
|
|
|
|
|
# glibc defines functions that are not available on Linux as stubs that
|
|
|
|
# fail with ENOSYS (such as e.g. lchmod). In this case we want to fail
|
|
|
|
# instead of detecting the stub as a valid symbol.
|
|
|
|
# We already included limits.h earlier to ensure that these are defined
|
|
|
|
# for stub functions.
|
|
|
|
stubs_fail = '''
|
|
|
|
#if defined __stub_{func} || defined __stub___{func}
|
|
|
|
fail fail fail this function is not going to work
|
|
|
|
#endif
|
|
|
|
'''
|
|
|
|
|
|
|
|
# If we have any includes in the prefix supplied by the user, assume
|
|
|
|
# that the user wants us to use the symbol prototype defined in those
|
|
|
|
# includes. If not, then try to do the Autoconf-style check with
|
|
|
|
# a dummy prototype definition of our own.
|
|
|
|
# This is needed when the linker determines symbol availability from an
|
|
|
|
# SDK based on the prototype in the header provided by the SDK.
|
|
|
|
# Ignoring this prototype would result in the symbol always being
|
|
|
|
# marked as available.
|
|
|
|
if '#include' in prefix:
|
|
|
|
head, main = self._have_prototype_templ()
|
|
|
|
else:
|
|
|
|
head, main = self._no_prototype_templ()
|
|
|
|
templ = head + stubs_fail + main
|
|
|
|
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
if self.links(templ.format(**fargs), env, extra_args=extra_args,
|
|
|
|
dependencies=dependencies):
|
|
|
|
return True
|
|
|
|
|
|
|
|
# MSVC does not have compiler __builtin_-s.
|
|
|
|
if self.get_id() == 'msvc':
|
|
|
|
return False
|
|
|
|
|
|
|
|
# Detect function as a built-in
|
|
|
|
#
|
|
|
|
# Some functions like alloca() are defined as compiler built-ins which
|
|
|
|
# are inlined by the compiler and you can't take their address, so we
|
|
|
|
# need to look for them differently. On nice compilers like clang, we
|
|
|
|
# can just directly use the __has_builtin() macro.
|
|
|
|
fargs['no_includes'] = '#include' not in prefix
|
|
|
|
t = '''{prefix}
|
|
|
|
int main() {{
|
|
|
|
#ifdef __has_builtin
|
|
|
|
#if !__has_builtin(__builtin_{func})
|
|
|
|
#error "__builtin_{func} not found"
|
|
|
|
#endif
|
|
|
|
#elif ! defined({func})
|
|
|
|
/* Check for __builtin_{func} only if no includes were added to the
|
|
|
|
* prefix above, which means no definition of {func} can be found.
|
|
|
|
* We would always check for this, but we get false positives on
|
|
|
|
* MSYS2 if we do. Their toolchain is broken, but we can at least
|
|
|
|
* give them a workaround. */
|
|
|
|
#if {no_includes:d}
|
|
|
|
__builtin_{func};
|
|
|
|
#else
|
|
|
|
#error "No definition for __builtin_{func} found in the prefix"
|
|
|
|
#endif
|
|
|
|
#endif
|
|
|
|
}}'''
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
return self.links(t.format(**fargs), env, extra_args=extra_args,
|
|
|
|
dependencies=dependencies)
|
|
|
|
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
def has_members(self, typename, membernames, prefix, env, *, extra_args=None, dependencies=None):
|
|
|
|
if extra_args is None:
|
|
|
|
extra_args = []
|
|
|
|
fargs = {'prefix': prefix, 'type': typename, 'name': 'foo'}
|
|
|
|
# Create code that accesses all members
|
|
|
|
members = ''
|
|
|
|
for member in membernames:
|
|
|
|
members += '{}.{};\n'.format(fargs['name'], member)
|
|
|
|
fargs['members'] = members
|
|
|
|
t = '''{prefix}
|
|
|
|
void bar() {{
|
|
|
|
{type} {name};
|
|
|
|
{members}
|
|
|
|
}};'''
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
return self.compiles(t.format(**fargs), env, extra_args=extra_args,
|
|
|
|
dependencies=dependencies)
|
|
|
|
|
|
|
|
def has_type(self, typename, prefix, env, extra_args, dependencies=None):
|
|
|
|
fargs = {'prefix': prefix, 'type': typename}
|
|
|
|
t = '''{prefix}
|
|
|
|
void bar() {{
|
|
|
|
sizeof({type});
|
|
|
|
}};'''
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
return self.compiles(t.format(**fargs), env, extra_args=extra_args,
|
|
|
|
dependencies=dependencies)
|
|
|
|
|
|
|
|
def symbols_have_underscore_prefix(self, env):
|
|
|
|
'''
|
|
|
|
Check if the compiler prefixes an underscore to global C symbols
|
|
|
|
'''
|
|
|
|
symbol_name = b'meson_uscore_prefix'
|
|
|
|
code = '''#ifdef __cplusplus
|
|
|
|
extern "C" {
|
|
|
|
#endif
|
|
|
|
void ''' + symbol_name.decode() + ''' () {}
|
|
|
|
#ifdef __cplusplus
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
'''
|
|
|
|
args = self.get_compiler_check_args()
|
|
|
|
n = 'symbols_have_underscore_prefix'
|
|
|
|
with self.compile(code, args, 'compile', want_output=True) as p:
|
|
|
|
if p.returncode != 0:
|
|
|
|
m = 'BUG: Unable to compile {!r} check: {}'
|
|
|
|
raise RuntimeError(m.format(n, p.stdo))
|
|
|
|
if not os.path.isfile(p.output_name):
|
|
|
|
m = 'BUG: Can\'t find compiled test code for {!r} check'
|
|
|
|
raise RuntimeError(m.format(n))
|
|
|
|
with open(p.output_name, 'rb') as o:
|
|
|
|
for line in o:
|
|
|
|
# Check if the underscore form of the symbol is somewhere
|
|
|
|
# in the output file.
|
|
|
|
if b'_' + symbol_name in line:
|
|
|
|
return True
|
|
|
|
# Else, check if the non-underscored form is present
|
|
|
|
elif symbol_name in line:
|
|
|
|
return False
|
|
|
|
raise RuntimeError('BUG: {!r} check failed unexpectedly'.format(n))
|
|
|
|
|
|
|
|
def _get_patterns(self, env, prefixes, suffixes, shared=False):
|
|
|
|
patterns = []
|
|
|
|
for p in prefixes:
|
|
|
|
for s in suffixes:
|
|
|
|
patterns.append(p + '{}.' + s)
|
|
|
|
if shared and for_openbsd(self.is_cross, env):
|
|
|
|
# Shared libraries on OpenBSD can be named libfoo.so.X.Y:
|
|
|
|
# https://www.openbsd.org/faq/ports/specialtopics.html#SharedLibs
|
|
|
|
#
|
|
|
|
# This globbing is probably the best matching we can do since regex
|
|
|
|
# is expensive. It's wrong in many edge cases, but it will match
|
|
|
|
# correctly-named libraries and hopefully no one on OpenBSD names
|
|
|
|
# their files libfoo.so.9a.7b.1.0
|
|
|
|
patterns.append('lib{}.so.[0-9]*.[0-9]*')
|
|
|
|
return patterns
|
|
|
|
|
|
|
|
def get_library_naming(self, env, libtype, strict=False):
|
|
|
|
'''
|
|
|
|
Get library prefixes and suffixes for the target platform ordered by
|
|
|
|
priority
|
|
|
|
'''
|
|
|
|
stlibext = ['a']
|
|
|
|
# We've always allowed libname to be both `foo` and `libfoo`,
|
|
|
|
# and now people depend on it
|
|
|
|
if strict and not isinstance(self, VisualStudioCCompiler): # lib prefix is not usually used with msvc
|
|
|
|
prefixes = ['lib']
|
|
|
|
else:
|
|
|
|
prefixes = ['lib', '']
|
|
|
|
# Library suffixes and prefixes
|
|
|
|
if for_darwin(env.is_cross_build(), env):
|
|
|
|
shlibext = ['dylib', 'so']
|
|
|
|
elif for_windows(env.is_cross_build(), env):
|
|
|
|
# FIXME: .lib files can be import or static so we should read the
|
|
|
|
# file, figure out which one it is, and reject the wrong kind.
|
|
|
|
if isinstance(self, VisualStudioCCompiler):
|
|
|
|
shlibext = ['lib']
|
|
|
|
else:
|
|
|
|
shlibext = ['dll.a', 'lib', 'dll']
|
|
|
|
# Yep, static libraries can also be foo.lib
|
|
|
|
stlibext += ['lib']
|
|
|
|
elif for_cygwin(env.is_cross_build(), env):
|
|
|
|
shlibext = ['dll', 'dll.a']
|
|
|
|
prefixes = ['cyg'] + prefixes
|
|
|
|
else:
|
|
|
|
# Linux/BSDs
|
|
|
|
shlibext = ['so']
|
|
|
|
patterns = []
|
|
|
|
# Search priority
|
|
|
|
if libtype in ('default', 'shared-static'):
|
|
|
|
patterns += self._get_patterns(env, prefixes, shlibext, True)
|
|
|
|
patterns += self._get_patterns(env, prefixes, stlibext, False)
|
|
|
|
elif libtype == 'static-shared':
|
|
|
|
patterns += self._get_patterns(env, prefixes, stlibext, False)
|
|
|
|
patterns += self._get_patterns(env, prefixes, shlibext, True)
|
|
|
|
elif libtype == 'shared':
|
|
|
|
patterns += self._get_patterns(env, prefixes, shlibext, True)
|
|
|
|
elif libtype == 'static':
|
|
|
|
patterns += self._get_patterns(env, prefixes, stlibext, False)
|
|
|
|
else:
|
|
|
|
raise AssertionError('BUG: unknown libtype {!r}'.format(libtype))
|
|
|
|
return tuple(patterns)
|
|
|
|
|
|
|
|
@staticmethod
|
|
|
|
def _sort_shlibs_openbsd(libs):
|
|
|
|
filtered = []
|
|
|
|
for lib in libs:
|
|
|
|
# Validate file as a shared library of type libfoo.so.X.Y
|
|
|
|
ret = lib.rsplit('.so.', maxsplit=1)
|
|
|
|
if len(ret) != 2:
|
|
|
|
continue
|
|
|
|
try:
|
|
|
|
float(ret[1])
|
|
|
|
except ValueError:
|
|
|
|
continue
|
|
|
|
filtered.append(lib)
|
|
|
|
float_cmp = lambda x: float(x.rsplit('.so.', maxsplit=1)[1])
|
|
|
|
return sorted(filtered, key=float_cmp, reverse=True)
|
|
|
|
|
|
|
|
@classmethod
|
|
|
|
def _get_trials_from_pattern(cls, pattern, directory, libname):
|
|
|
|
f = Path(directory) / pattern.format(libname)
|
|
|
|
# Globbing for OpenBSD
|
|
|
|
if '*' in pattern:
|
|
|
|
# NOTE: globbing matches directories and broken symlinks
|
|
|
|
# so we have to do an isfile test on it later
|
|
|
|
return cls._sort_shlibs_openbsd(glob.glob(str(f)))
|
|
|
|
return [f.as_posix()]
|
|
|
|
|
|
|
|
@staticmethod
|
|
|
|
def _get_file_from_list(env, files: List[str]) -> str:
|
|
|
|
'''
|
|
|
|
We just check whether the library exists. We can't do a link check
|
|
|
|
because the library might have unresolved symbols that require other
|
|
|
|
libraries. On macOS we check if the library matches our target
|
|
|
|
architecture.
|
|
|
|
'''
|
|
|
|
# If not building on macOS for Darwin, do a simple file check
|
|
|
|
if not env.machines.host.is_darwin() or not env.machines.build.is_darwin():
|
|
|
|
for f in files:
|
|
|
|
if os.path.isfile(f):
|
|
|
|
return f
|
|
|
|
# Run `lipo` and check if the library supports the arch we want
|
|
|
|
for f in files:
|
|
|
|
if not os.path.isfile(f):
|
|
|
|
continue
|
|
|
|
archs = darwin_get_object_archs(f)
|
|
|
|
if archs and env.machines.host.cpu_family in archs:
|
|
|
|
return f
|
|
|
|
else:
|
|
|
|
mlog.debug('Rejected {}, supports {} but need {}'
|
|
|
|
.format(f, archs, env.machines.host.cpu_family))
|
|
|
|
return None
|
|
|
|
|
|
|
|
@functools.lru_cache()
|
|
|
|
def output_is_64bit(self, env):
|
|
|
|
'''
|
|
|
|
returns true if the output produced is 64-bit, false if 32-bit
|
|
|
|
'''
|
|
|
|
return self.sizeof('void *', '', env) == 8
|
|
|
|
|
|
|
|
def find_library_real(self, libname, env, extra_dirs, code, libtype):
|
|
|
|
# First try if we can just add the library as -l.
|
|
|
|
# Gcc + co seem to prefer builtin lib dirs to -L dirs.
|
|
|
|
# Only try to find std libs if no extra dirs specified.
|
|
|
|
if not extra_dirs or libname in self.internal_libs:
|
|
|
|
args = ['-l' + libname]
|
|
|
|
largs = self.linker_to_compiler_args(self.get_allow_undefined_link_args())
|
|
|
|
if self.links(code, env, extra_args=(args + largs)):
|
|
|
|
return args
|
|
|
|
# Don't do a manual search for internal libs
|
|
|
|
if libname in self.internal_libs:
|
|
|
|
return None
|
|
|
|
# Not found or we want to use a specific libtype? Try to find the
|
|
|
|
# library file itself.
|
|
|
|
patterns = self.get_library_naming(env, libtype)
|
|
|
|
# try to detect if we are 64-bit or 32-bit. If we can't
|
|
|
|
# detect, we will just skip path validity checks done in
|
|
|
|
# get_library_dirs() call
|
|
|
|
try:
|
|
|
|
if self.output_is_64bit(env):
|
|
|
|
elf_class = 2
|
|
|
|
else:
|
|
|
|
elf_class = 1
|
|
|
|
except:
|
|
|
|
elf_class = 0
|
|
|
|
# Search in the specified dirs, and then in the system libraries
|
|
|
|
for d in itertools.chain(extra_dirs, self.get_library_dirs(env, elf_class)):
|
|
|
|
for p in patterns:
|
|
|
|
trial = self._get_trials_from_pattern(p, d, libname)
|
|
|
|
if not trial:
|
|
|
|
continue
|
|
|
|
trial = self._get_file_from_list(env, trial)
|
|
|
|
if not trial:
|
|
|
|
continue
|
|
|
|
return [trial]
|
|
|
|
return None
|
|
|
|
|
|
|
|
def find_library_impl(self, libname, env, extra_dirs, code, libtype):
|
|
|
|
# These libraries are either built-in or invalid
|
|
|
|
if libname in self.ignore_libs:
|
|
|
|
return []
|
|
|
|
if isinstance(extra_dirs, str):
|
|
|
|
extra_dirs = [extra_dirs]
|
|
|
|
key = (tuple(self.exelist), libname, tuple(extra_dirs), code, libtype)
|
|
|
|
if key not in self.find_library_cache:
|
|
|
|
value = self.find_library_real(libname, env, extra_dirs, code, libtype)
|
|
|
|
self.find_library_cache[key] = value
|
|
|
|
else:
|
|
|
|
value = self.find_library_cache[key]
|
|
|
|
if value is None:
|
|
|
|
return None
|
|
|
|
return value[:]
|
|
|
|
|
|
|
|
def find_library(self, libname, env, extra_dirs, libtype='default'):
|
|
|
|
code = 'int main(int argc, char **argv) { return 0; }'
|
|
|
|
return self.find_library_impl(libname, env, extra_dirs, code, libtype)
|
|
|
|
|
|
|
|
def find_framework_paths(self, env):
|
|
|
|
'''
|
|
|
|
These are usually /Library/Frameworks and /System/Library/Frameworks,
|
|
|
|
unless you select a particular macOS SDK with the -isysroot flag.
|
|
|
|
You can also add to this by setting -F in CFLAGS.
|
|
|
|
'''
|
|
|
|
if self.id != 'clang':
|
|
|
|
raise MesonException('Cannot find framework path with non-clang compiler')
|
|
|
|
# Construct the compiler command-line
|
|
|
|
commands = self.get_exelist() + ['-v', '-E', '-']
|
|
|
|
commands += self.get_always_args()
|
|
|
|
# Add CFLAGS/CXXFLAGS/OBJCFLAGS/OBJCXXFLAGS from the env
|
|
|
|
if env.is_cross_build() and not self.is_cross:
|
|
|
|
for_machine = MachineChoice.BUILD
|
|
|
|
else:
|
|
|
|
for_machine = MachineChoice.HOST
|
|
|
|
commands += env.coredata.get_external_args(for_machine, self.language)
|
|
|
|
mlog.debug('Finding framework path by running: ', ' '.join(commands), '\n')
|
|
|
|
os_env = os.environ.copy()
|
|
|
|
os_env['LC_ALL'] = 'C'
|
|
|
|
_, _, stde = Popen_safe(commands, env=os_env, stdin=subprocess.PIPE)
|
|
|
|
paths = []
|
|
|
|
for line in stde.split('\n'):
|
|
|
|
if '(framework directory)' not in line:
|
|
|
|
continue
|
|
|
|
# line is of the form:
|
|
|
|
# ` /path/to/framework (framework directory)`
|
|
|
|
paths.append(line[:-21].strip())
|
|
|
|
return paths
|
|
|
|
|
|
|
|
def find_framework_real(self, name, env, extra_dirs, allow_system):
|
|
|
|
code = 'int main(int argc, char **argv) { return 0; }'
|
|
|
|
link_args = []
|
|
|
|
for d in extra_dirs:
|
|
|
|
link_args += ['-F' + d]
|
|
|
|
# We can pass -Z to disable searching in the system frameworks, but
|
|
|
|
# then we must also pass -L/usr/lib to pick up libSystem.dylib
|
|
|
|
extra_args = [] if allow_system else ['-Z', '-L/usr/lib']
|
|
|
|
link_args += ['-framework', name]
|
|
|
|
if self.links(code, env, extra_args=(extra_args + link_args)):
|
|
|
|
return link_args
|
|
|
|
|
|
|
|
def find_framework_impl(self, name, env, extra_dirs, allow_system):
|
|
|
|
if isinstance(extra_dirs, str):
|
|
|
|
extra_dirs = [extra_dirs]
|
|
|
|
key = (tuple(self.exelist), name, tuple(extra_dirs), allow_system)
|
|
|
|
if key in self.find_framework_cache:
|
|
|
|
value = self.find_framework_cache[key]
|
|
|
|
else:
|
|
|
|
value = self.find_framework_real(name, env, extra_dirs, allow_system)
|
|
|
|
self.find_framework_cache[key] = value
|
|
|
|
if value is None:
|
|
|
|
return None
|
|
|
|
return value[:]
|
|
|
|
|
|
|
|
def find_framework(self, name, env, extra_dirs, allow_system=True):
|
|
|
|
'''
|
|
|
|
Finds the framework with the specified name, and returns link args for
|
|
|
|
the same or returns None when the framework is not found.
|
|
|
|
'''
|
|
|
|
if self.id != 'clang':
|
|
|
|
raise MesonException('Cannot find frameworks with non-clang compiler')
|
|
|
|
return self.find_framework_impl(name, env, extra_dirs, allow_system)
|
|
|
|
|
|
|
|
def thread_flags(self, env):
|
|
|
|
if for_haiku(self.is_cross, env) or for_darwin(self.is_cross, env):
|
|
|
|
return []
|
|
|
|
return ['-pthread']
|
|
|
|
|
|
|
|
def thread_link_flags(self, env):
|
|
|
|
if for_haiku(self.is_cross, env) or for_darwin(self.is_cross, env):
|
|
|
|
return []
|
|
|
|
return ['-pthread']
|
|
|
|
|
|
|
|
def linker_to_compiler_args(self, args):
|
|
|
|
return args
|
|
|
|
|
|
|
|
def has_arguments(self, args, env, code, mode):
|
|
|
|
return self.compiles(code, env, extra_args=args, mode=mode)
|
|
|
|
|
|
|
|
def has_multi_arguments(self, args, env):
|
|
|
|
for arg in args[:]:
|
|
|
|
# some compilers, e.g. GCC, don't warn for unsupported warning-disable
|
|
|
|
# flags, so when we are testing a flag like "-Wno-forgotten-towel", also
|
|
|
|
# check the equivalent enable flag too "-Wforgotten-towel"
|
|
|
|
if arg.startswith('-Wno-'):
|
|
|
|
args.append('-W' + arg[5:])
|
|
|
|
if arg.startswith('-Wl,'):
|
|
|
|
mlog.warning('{} looks like a linker argument, '
|
|
|
|
'but has_argument and other similar methods only '
|
|
|
|
'support checking compiler arguments. Using them '
|
|
|
|
'to check linker arguments are never supported, '
|
|
|
|
'and results are likely to be wrong regardless of '
|
|
|
|
'the compiler you are using. has_link_argument or '
|
|
|
|
'other similar method can be used instead.'
|
|
|
|
.format(arg))
|
|
|
|
code = 'int i;\n'
|
|
|
|
return self.has_arguments(args, env, code, mode='compile')
|
|
|
|
|
|
|
|
def has_multi_link_arguments(self, args, env):
|
|
|
|
# First time we check for link flags we need to first check if we have
|
|
|
|
# --fatal-warnings, otherwise some linker checks could give some
|
|
|
|
# false positive.
|
|
|
|
fatal_warnings_args = ['-Wl,--fatal-warnings']
|
|
|
|
if self.has_fatal_warnings_link_arg is None:
|
|
|
|
self.has_fatal_warnings_link_arg = False
|
|
|
|
self.has_fatal_warnings_link_arg = self.has_multi_link_arguments(fatal_warnings_args, env)
|
|
|
|
|
|
|
|
if self.has_fatal_warnings_link_arg:
|
|
|
|
args = fatal_warnings_args + args
|
|
|
|
|
|
|
|
args = self.linker_to_compiler_args(args)
|
|
|
|
code = 'int main(int argc, char **argv) { return 0; }'
|
|
|
|
return self.has_arguments(args, env, code, mode='link')
|
|
|
|
|
|
|
|
@staticmethod
|
|
|
|
def concatenate_string_literals(s):
|
|
|
|
pattern = re.compile(r'(?P<pre>.*([^\\]")|^")(?P<str1>([^\\"]|\\.)*)"\s+"(?P<str2>([^\\"]|\\.)*)(?P<post>".*)')
|
|
|
|
ret = s
|
|
|
|
m = pattern.match(ret)
|
|
|
|
while m:
|
|
|
|
ret = ''.join(m.group('pre', 'str1', 'str2', 'post'))
|
|
|
|
m = pattern.match(ret)
|
|
|
|
return ret
|
|
|
|
|
|
|
|
def has_func_attribute(self, name, env):
|
|
|
|
# Just assume that if we're not on windows that dllimport and dllexport
|
|
|
|
# don't work
|
|
|
|
if not (for_windows(env.is_cross_build(), env) or
|
|
|
|
for_cygwin(env.is_cross_build(), env)):
|
|
|
|
if name in ['dllimport', 'dllexport']:
|
|
|
|
return False
|
|
|
|
|
|
|
|
# Clang and GCC both return warnings if the __attribute__ is undefined,
|
|
|
|
# so set -Werror
|
|
|
|
return self.compiles(self.attribute_check_func(name), env, extra_args='-Werror')
|
|
|
|
|
|
|
|
|
|
|
|
class ClangCCompiler(ClangCompiler, CCompiler):
|
|
|
|
def __init__(self, exelist, version, compiler_type, is_cross, exe_wrapper=None, **kwargs):
|
|
|
|
CCompiler.__init__(self, exelist, version, is_cross, exe_wrapper, **kwargs)
|
|
|
|
ClangCompiler.__init__(self, compiler_type)
|
|
|
|
default_warn_args = ['-Wall', '-Winvalid-pch']
|
|
|
|
self.warn_args = {'1': default_warn_args,
|
|
|
|
'2': default_warn_args + ['-Wextra'],
|
|
|
|
'3': default_warn_args + ['-Wextra', '-Wpedantic']}
|
|
|
|
|
|
|
|
def get_options(self):
|
|
|
|
opts = CCompiler.get_options(self)
|
|
|
|
opts.update({'c_std': coredata.UserComboOption('c_std', 'C language standard to use',
|
|
|
|
['none', 'c89', 'c99', 'c11',
|
|
|
|
'gnu89', 'gnu99', 'gnu11'],
|
|
|
|
'none')})
|
|
|
|
return opts
|
|
|
|
|
|
|
|
def get_option_compile_args(self, options):
|
|
|
|
args = []
|
|
|
|
std = options['c_std']
|
|
|
|
if std.value != 'none':
|
|
|
|
args.append('-std=' + std.value)
|
|
|
|
return args
|
|
|
|
|
|
|
|
def get_option_link_args(self, options):
|
|
|
|
return []
|
|
|
|
|
|
|
|
def get_linker_always_args(self):
|
|
|
|
basic = super().get_linker_always_args()
|
|
|
|
if self.compiler_type.is_osx_compiler:
|
|
|
|
return basic + ['-Wl,-headerpad_max_install_names']
|
|
|
|
return basic
|
|
|
|
|
|
|
|
|
|
|
|
class ArmclangCCompiler(ArmclangCompiler, CCompiler):
|
|
|
|
def __init__(self, exelist, version, compiler_type, is_cross, exe_wrapper=None, **kwargs):
|
|
|
|
CCompiler.__init__(self, exelist, version, is_cross, exe_wrapper, **kwargs)
|
|
|
|
ArmclangCompiler.__init__(self, compiler_type)
|
|
|
|
default_warn_args = ['-Wall', '-Winvalid-pch']
|
|
|
|
self.warn_args = {'1': default_warn_args,
|
|
|
|
'2': default_warn_args + ['-Wextra'],
|
|
|
|
'3': default_warn_args + ['-Wextra', '-Wpedantic']}
|
|
|
|
|
|
|
|
def get_options(self):
|
|
|
|
opts = CCompiler.get_options(self)
|
|
|
|
opts.update({'c_std': coredata.UserComboOption('c_std', 'C language standard to use',
|
|
|
|
['none', 'c90', 'c99', 'c11',
|
|
|
|
'gnu90', 'gnu99', 'gnu11'],
|
|
|
|
'none')})
|
|
|
|
return opts
|
|
|
|
|
|
|
|
def get_option_compile_args(self, options):
|
|
|
|
args = []
|
|
|
|
std = options['c_std']
|
|
|
|
if std.value != 'none':
|
|
|
|
args.append('-std=' + std.value)
|
|
|
|
return args
|
|
|
|
|
|
|
|
def get_option_link_args(self, options):
|
|
|
|
return []
|
|
|
|
|
|
|
|
|
|
|
|
class GnuCCompiler(GnuCompiler, CCompiler):
|
|
|
|
def __init__(self, exelist, version, compiler_type, is_cross, exe_wrapper=None, defines=None, **kwargs):
|
|
|
|
CCompiler.__init__(self, exelist, version, is_cross, exe_wrapper, **kwargs)
|
|
|
|
GnuCompiler.__init__(self, compiler_type, defines)
|
|
|
|
default_warn_args = ['-Wall', '-Winvalid-pch']
|
|
|
|
self.warn_args = {'1': default_warn_args,
|
|
|
|
'2': default_warn_args + ['-Wextra'],
|
|
|
|
'3': default_warn_args + ['-Wextra', '-Wpedantic']}
|
|
|
|
|
|
|
|
def get_options(self):
|
|
|
|
opts = CCompiler.get_options(self)
|
|
|
|
opts.update({'c_std': coredata.UserComboOption('c_std', 'C language standard to use',
|
|
|
|
['none', 'c89', 'c99', 'c11',
|
|
|
|
'gnu89', 'gnu99', 'gnu11'],
|
|
|
|
'none')})
|
|
|
|
if self.compiler_type.is_windows_compiler:
|
|
|
|
opts.update({
|
|
|
|
'c_winlibs': coredata.UserArrayOption('c_winlibs', 'Standard Win libraries to link against',
|
|
|
|
gnu_winlibs), })
|
|
|
|
return opts
|
|
|
|
|
|
|
|
def get_option_compile_args(self, options):
|
|
|
|
args = []
|
|
|
|
std = options['c_std']
|
|
|
|
if std.value != 'none':
|
|
|
|
args.append('-std=' + std.value)
|
|
|
|
return args
|
|
|
|
|
|
|
|
def get_option_link_args(self, options):
|
|
|
|
if self.compiler_type.is_windows_compiler:
|
|
|
|
return options['c_winlibs'].value[:]
|
|
|
|
return []
|
|
|
|
|
|
|
|
def get_pch_use_args(self, pch_dir, header):
|
|
|
|
return ['-fpch-preprocess', '-include', os.path.basename(header)]
|
|
|
|
|
|
|
|
|
|
|
|
class PGICCompiler(PGICompiler, CCompiler):
|
|
|
|
def __init__(self, exelist, version, is_cross, exe_wrapper=None, **kwargs):
|
|
|
|
CCompiler.__init__(self, exelist, version, is_cross, exe_wrapper, **kwargs)
|
|
|
|
PGICompiler.__init__(self, CompilerType.PGI_STANDARD)
|
|
|
|
|
|
|
|
|
|
|
|
class ElbrusCCompiler(GnuCCompiler, ElbrusCompiler):
|
|
|
|
def __init__(self, exelist, version, compiler_type, is_cross, exe_wrapper=None, defines=None, **kwargs):
|
|
|
|
GnuCCompiler.__init__(self, exelist, version, compiler_type, is_cross, exe_wrapper, defines, **kwargs)
|
|
|
|
ElbrusCompiler.__init__(self, compiler_type, defines)
|
|
|
|
|
|
|
|
# It does support some various ISO standards and c/gnu 90, 9x, 1x in addition to those which GNU CC supports.
|
|
|
|
def get_options(self):
|
|
|
|
opts = CCompiler.get_options(self)
|
|
|
|
opts.update({'c_std': coredata.UserComboOption('c_std', 'C language standard to use',
|
|
|
|
['none', 'c89', 'c90', 'c9x', 'c99', 'c1x', 'c11',
|
|
|
|
'gnu89', 'gnu90', 'gnu9x', 'gnu99', 'gnu1x', 'gnu11',
|
|
|
|
'iso9899:2011', 'iso9899:1990', 'iso9899:199409', 'iso9899:1999'],
|
|
|
|
'none')})
|
|
|
|
return opts
|
|
|
|
|
|
|
|
# Elbrus C compiler does not have lchmod, but there is only linker warning, not compiler error.
|
|
|
|
# So we should explicitly fail at this case.
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
def has_function(self, funcname, prefix, env, *, extra_args=None, dependencies=None):
|
|
|
|
if funcname == 'lchmod':
|
|
|
|
return False
|
|
|
|
else:
|
compilers: Use keyword only arguments for compiler interfaces
Because we need to inherit them in some cases, and python's
keyword-or-positional arguments make this really painful, especially
with inheritance. They do this in two ways:
1) If you want to intercept the arguments you need to check for both a
keyword and a positional argument, because you could get either. Then
you need to make sure that you only pass one of those down to the
next layer.
2) After you do that, if the layer below you decides to do the same
thing, but uses the other form (you used keyword by the lower level
uses positional or vice versa), then you'll get a TypeError since two
layers down got the argument as both a positional and a keyword.
All of this is bad. Fortunately python 3.x provides a mechanism to solve
this, keyword only arguments. These arguments cannot be based
positionally, the interpreter will give us an error in that case.
I have made a best effort to do this correctly, and I've verified it
with GCC, Clang, ICC, and MSVC, but there are other compilers like Arm
and Elbrus that I don't have access to.
6 years ago
|
|
|
return super().has_function(funcname, prefix, env,
|
|
|
|
extra_args=extra_args,
|
|
|
|
dependencies=dependencies)
|
|
|
|
|
|
|
|
|
|
|
|
class IntelCCompiler(IntelCompiler, CCompiler):
|
|
|
|
def __init__(self, exelist, version, compiler_type, is_cross, exe_wrapper=None, **kwargs):
|
|
|
|
CCompiler.__init__(self, exelist, version, is_cross, exe_wrapper, **kwargs)
|
|
|
|
IntelCompiler.__init__(self, compiler_type)
|
|
|
|
self.lang_header = 'c-header'
|
|
|
|
default_warn_args = ['-Wall', '-w3', '-diag-disable:remark']
|
|
|
|
self.warn_args = {'1': default_warn_args,
|
|
|
|
'2': default_warn_args + ['-Wextra'],
|
|
|
|
'3': default_warn_args + ['-Wextra']}
|
|
|
|
|
|
|
|
def get_options(self):
|
|
|
|
opts = CCompiler.get_options(self)
|
|
|
|
c_stds = ['c89', 'c99']
|
|
|
|
g_stds = ['gnu89', 'gnu99']
|
|
|
|
if version_compare(self.version, '>=16.0.0'):
|
|
|
|
c_stds += ['c11']
|
|
|
|
opts.update({'c_std': coredata.UserComboOption('c_std', 'C language standard to use',
|
|
|
|
['none'] + c_stds + g_stds,
|
|
|
|
'none')})
|
|
|
|
return opts
|
|
|
|
|
|
|
|
def get_option_compile_args(self, options):
|
|
|
|
args = []
|
|
|
|
std = options['c_std']
|
|
|
|
if std.value != 'none':
|
|
|
|
args.append('-std=' + std.value)
|
|
|
|
return args
|
|
|
|
|
|
|
|
|
|
|
|
class VisualStudioCCompiler(CCompiler):
|
|
|
|
std_warn_args = ['/W3']
|
|
|
|
std_opt_args = ['/O2']
|
|
|
|
ignore_libs = gnu_compiler_internal_libs
|
|
|
|
internal_libs = ()
|
|
|
|
|
|
|
|
crt_args = {'none': [],
|
|
|
|
'md': ['/MD'],
|
|
|
|
'mdd': ['/MDd'],
|
|
|
|
'mt': ['/MT'],
|
|
|
|
'mtd': ['/MTd'],
|
|
|
|
}
|
|
|
|
|
|
|
|
def __init__(self, exelist, version, is_cross, exe_wrap, target):
|
|
|
|
CCompiler.__init__(self, exelist, version, is_cross, exe_wrap)
|
|
|
|
self.id = 'msvc'
|
|
|
|
# /showIncludes is needed for build dependency tracking in Ninja
|
|
|
|
# See: https://ninja-build.org/manual.html#_deps
|
|
|
|
self.always_args = ['/nologo', '/showIncludes']
|
|
|
|
self.warn_args = {'1': ['/W2'],
|
|
|
|
'2': ['/W3'],
|
|
|
|
'3': ['/W4']}
|
|
|
|
self.base_options = ['b_pch', 'b_ndebug', 'b_vscrt'] # FIXME add lto, pgo and the like
|
|
|
|
self.target = target
|
|
|
|
self.is_64 = ('x64' in target) or ('x86_64' in target)
|
|
|
|
|
|
|
|
# Override CCompiler.get_always_args
|
|
|
|
def get_always_args(self):
|
|
|
|
return self.always_args
|
|
|
|
|
|
|
|
def get_linker_debug_crt_args(self):
|
|
|
|
"""
|
|
|
|
Arguments needed to select a debug crt for the linker
|
|
|
|
|
|
|
|
Sometimes we need to manually select the CRT (C runtime) to use with
|
|
|
|
MSVC. One example is when trying to link with static libraries since
|
|
|
|
MSVC won't auto-select a CRT for us in that case and will error out
|
|
|
|
asking us to select one.
|
|
|
|
"""
|
|
|
|
return ['/MDd']
|
|
|
|
|
|
|
|
def get_buildtype_args(self, buildtype):
|
|
|
|
args = compilers.msvc_buildtype_args[buildtype]
|
|
|
|
if self.id == 'msvc' and version_compare(self.version, '<18.0'):
|
|
|
|
args = [arg for arg in args if arg != '/Gw']
|
|
|
|
return args
|
|
|
|
|
|
|
|
def get_buildtype_linker_args(self, buildtype):
|
|
|
|
return compilers.msvc_buildtype_linker_args[buildtype]
|
|
|
|
|
|
|
|
def get_pch_suffix(self):
|
|
|
|
return 'pch'
|
|
|
|
|
|
|
|
def get_pch_name(self, header):
|
|
|
|
chopped = os.path.basename(header).split('.')[:-1]
|
|
|
|
chopped.append(self.get_pch_suffix())
|
|
|
|
pchname = '.'.join(chopped)
|
|
|
|
return pchname
|
|
|
|
|
|
|
|
def get_pch_use_args(self, pch_dir, header):
|
|
|
|
base = os.path.basename(header)
|
|
|
|
if self.id == 'clang-cl':
|
|
|
|
base = header
|
|
|
|
pchname = self.get_pch_name(header)
|
|
|
|
return ['/FI' + base, '/Yu' + base, '/Fp' + os.path.join(pch_dir, pchname)]
|
|
|
|
|
|
|
|
def get_preprocess_only_args(self):
|
|
|
|
return ['/EP']
|
|
|
|
|
|
|
|
def get_compile_only_args(self):
|
|
|
|
return ['/c']
|
|
|
|
|
|
|
|
def get_no_optimization_args(self):
|
|
|
|
return ['/Od']
|
|
|
|
|
|
|
|
def get_output_args(self, target):
|
|
|
|
if target.endswith('.exe'):
|
|
|
|
return ['/Fe' + target]
|
|
|
|
return ['/Fo' + target]
|
|
|
|
|
|
|
|
def get_optimization_args(self, optimization_level):
|
|
|
|
return compilers.msvc_optimization_args[optimization_level]
|
|
|
|
|
|
|
|
def get_debug_args(self, is_debug):
|
|
|
|
return compilers.msvc_debug_args[is_debug]
|
|
|
|
|
|
|
|
def get_dependency_gen_args(self, outtarget, outfile):
|
|
|
|
return []
|
|
|
|
|
|
|
|
def get_linker_exelist(self):
|
|
|
|
# FIXME, should have same path as compiler.
|
|
|
|
# FIXME, should be controllable via cross-file.
|
|
|
|
if self.id == 'clang-cl':
|
|
|
|
return ['lld-link']
|
|
|
|
else:
|
|
|
|
return ['link']
|
|
|
|
|
|
|
|
def get_linker_always_args(self):
|
|
|
|
return ['/nologo']
|
|
|
|
|
|
|
|
def get_linker_output_args(self, outputname):
|
|
|
|
return ['/OUT:' + outputname]
|
|
|
|
|
|
|
|
def get_linker_search_args(self, dirname):
|
|
|
|
return ['/LIBPATH:' + dirname]
|
|
|
|
|
|
|
|
def linker_to_compiler_args(self, args):
|
|
|
|
return ['/link'] + args
|
|
|
|
|
|
|
|
def get_gui_app_args(self, value):
|
|
|
|
# the default is for the linker to guess the subsystem based on presence
|
|
|
|
# of main or WinMain symbols, so always be explicit
|
|
|
|
if value:
|
|
|
|
return ['/SUBSYSTEM:WINDOWS']
|
|
|
|
else:
|
|
|
|
return ['/SUBSYSTEM:CONSOLE']
|
|
|
|
|
|
|
|
def get_pic_args(self):
|
|
|
|
return [] # PIC is handled by the loader on Windows
|
|
|
|
|
|
|
|
def gen_export_dynamic_link_args(self, env):
|
|
|
|
return [] # Not applicable with MSVC
|
|
|
|
|
|
|
|
def get_std_shared_lib_link_args(self):
|
|
|
|
return ['/DLL']
|
|
|
|
|
|
|
|
def gen_vs_module_defs_args(self, defsfile):
|
|
|
|
if not isinstance(defsfile, str):
|
|
|
|
raise RuntimeError('Module definitions file should be str')
|
|
|
|
# With MSVC, DLLs only export symbols that are explicitly exported,
|
|
|
|
# so if a module defs file is specified, we use that to export symbols
|
|
|
|
return ['/DEF:' + defsfile]
|
|
|
|
|
|
|
|
def gen_pch_args(self, header, source, pchname):
|
|
|
|
objname = os.path.splitext(pchname)[0] + '.obj'
|
|
|
|
return objname, ['/Yc' + header, '/Fp' + pchname, '/Fo' + objname]
|
|
|
|
|
|
|
|
def gen_import_library_args(self, implibname):
|
|
|
|
"The name of the outputted import library"
|
|
|
|
return ['/IMPLIB:' + implibname]
|
|
|
|
|
|
|
|
def build_rpath_args(self, build_dir, from_dir, rpath_paths, build_rpath, install_rpath):
|
|
|
|
return []
|
|
|
|
|
|
|
|
def openmp_flags(self):
|
|
|
|
return ['/openmp']
|
|
|
|
|
|
|
|
# FIXME, no idea what these should be.
|
|
|
|
def thread_flags(self, env):
|
|
|
|
return []
|
|
|
|
|
|
|
|
def thread_link_flags(self, env):
|
|
|
|
return []
|
|
|
|
|
|
|
|
def get_options(self):
|
|
|
|
opts = CCompiler.get_options(self)
|
|
|
|
opts.update({'c_winlibs': coredata.UserArrayOption('c_winlibs',
|
|
|
|
'Windows libs to link against.',
|
|
|
|
msvc_winlibs)})
|
|
|
|
return opts
|
|
|
|
|
|
|
|
def get_option_link_args(self, options):
|
|
|
|
return options['c_winlibs'].value[:]
|
|
|
|
|
|
|
|
@classmethod
|
|
|
|
def unix_args_to_native(cls, args):
|
|
|
|
result = []
|
|
|
|
for i in args:
|
|
|
|
# -mms-bitfields is specific to MinGW-GCC
|
|
|
|
# -pthread is only valid for GCC
|
|
|
|
if i in ('-mms-bitfields', '-pthread'):
|
|
|
|
continue
|
|
|
|
if i.startswith('-L'):
|
|
|
|
i = '/LIBPATH:' + i[2:]
|
|
|
|
# Translate GNU-style -lfoo library name to the import library
|
|
|
|
elif i.startswith('-l'):
|
|
|
|
name = i[2:]
|
|
|
|
if name in cls.ignore_libs:
|
|
|
|
# With MSVC, these are provided by the C runtime which is
|
|
|
|
# linked in by default
|
|
|
|
continue
|
|
|
|
else:
|
|
|
|
i = name + '.lib'
|
|
|
|
# -pthread in link flags is only used on Linux
|
|
|
|
elif i == '-pthread':
|
|
|
|
continue
|
|
|
|
result.append(i)
|
|
|
|
return result
|
|
|
|
|
|
|
|
def get_werror_args(self):
|
|
|
|
return ['/WX']
|
|
|
|
|
|
|
|
def get_include_args(self, path, is_system):
|
|
|
|
if path == '':
|
|
|
|
path = '.'
|
|
|
|
# msvc does not have a concept of system header dirs.
|
|
|
|
return ['-I' + path]
|
|
|
|
|
|
|
|
def compute_parameters_with_absolute_paths(self, parameter_list, build_dir):
|
|
|
|
for idx, i in enumerate(parameter_list):
|
|
|
|
if i[:2] == '-I' or i[:2] == '/I':
|
|
|
|
parameter_list[idx] = i[:2] + os.path.normpath(os.path.join(build_dir, i[2:]))
|
|
|
|
elif i[:9] == '/LIBPATH:':
|
|
|
|
parameter_list[idx] = i[:9] + os.path.normpath(os.path.join(build_dir, i[9:]))
|
|
|
|
|
|
|
|
return parameter_list
|
|
|
|
|
|
|
|
# Visual Studio is special. It ignores some arguments it does not
|
|
|
|
# understand and you can't tell it to error out on those.
|
|
|
|
# http://stackoverflow.com/questions/15259720/how-can-i-make-the-microsoft-c-compiler-treat-unknown-flags-as-errors-rather-t
|
|
|
|
def has_arguments(self, args, env, code, mode):
|
|
|
|
warning_text = '4044' if mode == 'link' else '9002'
|
|
|
|
if self.id == 'clang-cl' and mode != 'link':
|
|
|
|
args = args + ['-Werror=unknown-argument']
|
|
|
|
with self._build_wrapper(code, env, extra_args=args, mode=mode) as p:
|
|
|
|
if p.returncode != 0:
|
|
|
|
return False
|
|
|
|
return not(warning_text in p.stde or warning_text in p.stdo)
|
|
|
|
|
|
|
|
def get_compile_debugfile_args(self, rel_obj, pch=False):
|
|
|
|
pdbarr = rel_obj.split('.')[:-1]
|
|
|
|
pdbarr += ['pdb']
|
|
|
|
args = ['/Fd' + '.'.join(pdbarr)]
|
|
|
|
# When generating a PDB file with PCH, all compile commands write
|
|
|
|
# to the same PDB file. Hence, we need to serialize the PDB
|
|
|
|
# writes using /FS since we do parallel builds. This slows down the
|
|
|
|
# build obviously, which is why we only do this when PCH is on.
|
|
|
|
# This was added in Visual Studio 2013 (MSVC 18.0). Before that it was
|
|
|
|
# always on: https://msdn.microsoft.com/en-us/library/dn502518.aspx
|
|
|
|
if pch and self.id == 'msvc' and version_compare(self.version, '>=18.0'):
|
|
|
|
args = ['/FS'] + args
|
|
|
|
return args
|
|
|
|
|
|
|
|
def get_link_debugfile_args(self, targetfile):
|
|
|
|
pdbarr = targetfile.split('.')[:-1]
|
|
|
|
pdbarr += ['pdb']
|
|
|
|
return ['/DEBUG', '/PDB:' + '.'.join(pdbarr)]
|
|
|
|
|
|
|
|
def get_link_whole_for(self, args):
|
|
|
|
# Only since VS2015
|
|
|
|
args = listify(args)
|
|
|
|
return ['/WHOLEARCHIVE:' + x for x in args]
|
|
|
|
|
|
|
|
def get_instruction_set_args(self, instruction_set):
|
|
|
|
if self.is_64:
|
|
|
|
return vs64_instruction_set_args.get(instruction_set, None)
|
|
|
|
if self.id == 'msvc' and self.version.split('.')[0] == '16' and instruction_set == 'avx':
|
|
|
|
# VS documentation says that this exists and should work, but
|
|
|
|
# it does not. The headers do not contain AVX intrinsics
|
|
|
|
# and the can not be called.
|
|
|
|
return None
|
|
|
|
return vs32_instruction_set_args.get(instruction_set, None)
|
|
|
|
|
|
|
|
def get_toolset_version(self):
|
|
|
|
if self.id == 'clang-cl':
|
|
|
|
# I have no idea
|
|
|
|
return '14.1'
|
|
|
|
|
|
|
|
# See boost/config/compiler/visualc.cpp for up to date mapping
|
|
|
|
try:
|
|
|
|
version = int(''.join(self.version.split('.')[0:2]))
|
|
|
|
except ValueError:
|
|
|
|
return None
|
|
|
|
if version < 1310:
|
|
|
|
return '7.0'
|
|
|
|
elif version < 1400:
|
|
|
|
return '7.1' # (Visual Studio 2003)
|
|
|
|
elif version < 1500:
|
|
|
|
return '8.0' # (Visual Studio 2005)
|
|
|
|
elif version < 1600:
|
|
|
|
return '9.0' # (Visual Studio 2008)
|
|
|
|
elif version < 1700:
|
|
|
|
return '10.0' # (Visual Studio 2010)
|
|
|
|
elif version < 1800:
|
|
|
|
return '11.0' # (Visual Studio 2012)
|
|
|
|
elif version < 1900:
|
|
|
|
return '12.0' # (Visual Studio 2013)
|
|
|
|
elif version < 1910:
|
|
|
|
return '14.0' # (Visual Studio 2015)
|
|
|
|
elif version < 1920:
|
|
|
|
return '14.1' # (Visual Studio 2017)
|
|
|
|
return None
|
|
|
|
|
|
|
|
def get_default_include_dirs(self):
|
|
|
|
if 'INCLUDE' not in os.environ:
|
|
|
|
return []
|
|
|
|
return os.environ['INCLUDE'].split(os.pathsep)
|
|
|
|
|
|
|
|
def get_crt_compile_args(self, crt_val, buildtype):
|
|
|
|
if crt_val in self.crt_args:
|
|
|
|
return self.crt_args[crt_val]
|
|
|
|
assert(crt_val == 'from_buildtype')
|
|
|
|
# Match what build type flags used to do.
|
|
|
|
if buildtype == 'plain':
|
|
|
|
return []
|
|
|
|
elif buildtype == 'debug':
|
|
|
|
return self.crt_args['mdd']
|
|
|
|
elif buildtype == 'debugoptimized':
|
|
|
|
return self.crt_args['md']
|
|
|
|
elif buildtype == 'release':
|
|
|
|
return self.crt_args['md']
|
|
|
|
elif buildtype == 'minsize':
|
|
|
|
return self.crt_args['md']
|
|
|
|
else:
|
|
|
|
assert(buildtype == 'custom')
|
|
|
|
raise EnvironmentException('Requested C runtime based on buildtype, but buildtype is "custom".')
|
|
|
|
|
|
|
|
def has_func_attribute(self, name, env):
|
|
|
|
# MSVC doesn't have __attribute__ like Clang and GCC do, so just return
|
|
|
|
# false without compiling anything
|
|
|
|
return name in ['dllimport', 'dllexport']
|
|
|
|
|
|
|
|
def get_argument_syntax(self):
|
|
|
|
return 'msvc'
|
|
|
|
|
|
|
|
def get_allow_undefined_link_args(self):
|
|
|
|
# link.exe
|
|
|
|
return ['/FORCE:UNRESOLVED']
|
|
|
|
|
|
|
|
|
|
|
|
class ClangClCCompiler(VisualStudioCCompiler):
|
|
|
|
def __init__(self, exelist, version, is_cross, exe_wrap, target):
|
|
|
|
super().__init__(exelist, version, is_cross, exe_wrap, target)
|
|
|
|
self.id = 'clang-cl'
|
|
|
|
|
|
|
|
|
|
|
|
class ArmCCompiler(ArmCompiler, CCompiler):
|
|
|
|
def __init__(self, exelist, version, compiler_type, is_cross, exe_wrapper=None, **kwargs):
|
|
|
|
CCompiler.__init__(self, exelist, version, is_cross, exe_wrapper, **kwargs)
|
|
|
|
ArmCompiler.__init__(self, compiler_type)
|
|
|
|
|
|
|
|
def get_options(self):
|
|
|
|
opts = CCompiler.get_options(self)
|
|
|
|
opts.update({'c_std': coredata.UserComboOption('c_std', 'C language standard to use',
|
|
|
|
['none', 'c90', 'c99'],
|
|
|
|
'none')})
|
|
|
|
return opts
|
|
|
|
|
|
|
|
def get_option_compile_args(self, options):
|
|
|
|
args = []
|
|
|
|
std = options['c_std']
|
|
|
|
if std.value != 'none':
|
|
|
|
args.append('--' + std.value)
|
|
|
|
return args
|
|
|
|
|
|
|
|
class CcrxCCompiler(CcrxCompiler, CCompiler):
|
|
|
|
def __init__(self, exelist, version, compiler_type, is_cross, exe_wrapper=None, **kwargs):
|
|
|
|
CCompiler.__init__(self, exelist, version, is_cross, exe_wrapper, **kwargs)
|
|
|
|
CcrxCompiler.__init__(self, compiler_type)
|
|
|
|
|
|
|
|
# Override CCompiler.get_always_args
|
|
|
|
def get_always_args(self):
|
|
|
|
return ['-nologo']
|
|
|
|
|
|
|
|
def get_options(self):
|
|
|
|
opts = CCompiler.get_options(self)
|
|
|
|
opts.update({'c_std': coredata.UserComboOption('c_std', 'C language standard to use',
|
|
|
|
['none', 'c89', 'c99'],
|
|
|
|
'none')})
|
|
|
|
return opts
|
|
|
|
|
|
|
|
def get_option_compile_args(self, options):
|
|
|
|
args = []
|
|
|
|
std = options['c_std']
|
|
|
|
if std.value == 'c89':
|
|
|
|
args.append('-lang=c')
|
|
|
|
elif std.value == 'c99':
|
|
|
|
args.append('-lang=c99')
|
|
|
|
return args
|
|
|
|
|
|
|
|
def get_compile_only_args(self):
|
|
|
|
return []
|
|
|
|
|
|
|
|
def get_no_optimization_args(self):
|
|
|
|
return ['-optimize=0']
|
|
|
|
|
|
|
|
def get_output_args(self, target):
|
|
|
|
return ['-output=obj=%s' % target]
|
|
|
|
|
|
|
|
def get_linker_output_args(self, outputname):
|
|
|
|
return ['-output=%s' % outputname]
|
|
|
|
|
|
|
|
def get_werror_args(self):
|
|
|
|
return ['-change_message=error']
|
|
|
|
|
|
|
|
def get_include_args(self, path, is_system):
|
|
|
|
if path == '':
|
|
|
|
path = '.'
|
|
|
|
return ['-include=' + path]
|