meson/mesonbuild/backend/ninjabackend.py

# 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.

from collections import OrderedDict
from enum import Enum, unique
from functools import lru_cache
from pathlib import PurePath, Path
from textwrap import dedent
import itertools
import json
import os
import pickle
import re
import shlex
import subprocess
import typing as T

from . import backends
from .. import modules
from .. import environment, mesonlib
from .. import build
from .. import mlog
from .. import compilers
from ..arglist import CompilerArgs
from ..compilers import (
    Compiler, CCompiler,
    FortranCompiler,
    mixins,
    PGICCompiler,
    VisualStudioLikeCompiler,
)
from ..linkers import ArLinker, RSPFileSyntax
from ..mesonlib import (
    File, LibType, MachineChoice, MesonException, OrderedSet, PerMachine,
    ProgressBar, quote_arg
)
from ..mesonlib import get_compiler_for_source, has_path_sep, OptionKey
from .backends import CleanTrees
from ..build import GeneratedList, InvalidArguments, ExtractedObjects
from ..interpreter import Interpreter

if T.TYPE_CHECKING:
    from .._typing import ImmutableListProtocol
    from ..linkers import DynamicLinker, StaticLinker
    from ..compilers.cs import CsCompiler
    from ..interpreter.interpreter import SourceOutputs


FORTRAN_INCLUDE_PAT = r"^\s*#?include\s*['\"](\w+\.\w+)['\"]"
FORTRAN_MODULE_PAT = r"^\s*\bmodule\b\s+(\w+)\s*(?:!+.*)*$"
FORTRAN_SUBMOD_PAT = r"^\s*\bsubmodule\b\s*\((\w+:?\w+)\)\s*(\w+)"
FORTRAN_USE_PAT = r"^\s*use,?\s*(?:non_intrinsic)?\s*(?:::)?\s*(\w+)"

def cmd_quote(s):
    # see: https://docs.microsoft.com/en-us/windows/desktop/api/shellapi/nf-shellapi-commandlinetoargvw#remarks

    # backslash escape any existing double quotes
    # any existing backslashes preceding a quote are doubled
    s = re.sub(r'(\\*)"', lambda m: '\\' * (len(m.group(1)) * 2 + 1) + '"', s)
    # any terminal backslashes likewise need doubling
    s = re.sub(r'(\\*)$', lambda m: '\\' * (len(m.group(1)) * 2), s)
    # and double quote
    s = f'"{s}"'

    return s

def gcc_rsp_quote(s):
    # see: the function buildargv() in libiberty
    #
    # this differs from sh-quoting in that a backslash *always* escapes the
    # following character, even inside single quotes.

    s = s.replace('\\', '\\\\')

    return shlex.quote(s)

# How ninja executes command lines differs between Unix and Windows
# (see https://ninja-build.org/manual.html#ref_rule_command)
if mesonlib.is_windows():
    quote_func = cmd_quote
    execute_wrapper = ['cmd', '/c']  # unused
    rmfile_prefix = ['del', '/f', '/s', '/q', '{}', '&&']
else:
    quote_func = quote_arg
    execute_wrapper = []
    rmfile_prefix = ['rm', '-f', '{}', '&&']


def get_rsp_threshold():
    '''Return a conservative estimate of the commandline size in bytes
    above which a response file should be used.  May be overridden for
    debugging by setting environment variable MESON_RSP_THRESHOLD.'''

    if mesonlib.is_windows():
        # Usually 32k, but some projects might use cmd.exe,
        # and that has a limit of 8k.
        limit = 8192
    else:
        # On Linux, ninja always passes the commandline as a single
        # big string to /bin/sh, and the kernel limits the size of a
        # single argument; see MAX_ARG_STRLEN
        limit = 131072
    # Be conservative
    limit = limit / 2
    return int(os.environ.get('MESON_RSP_THRESHOLD', limit))

# a conservative estimate of the command-line length limit
rsp_threshold = get_rsp_threshold()

# ninja variables whose value should remain unquoted. The value of these ninja
# variables (or variables we use them in) is interpreted directly by ninja
# (e.g. the value of the depfile variable is a pathname that ninja will read
# from, etc.), so it must not be shell quoted.
raw_names = {'DEPFILE_UNQUOTED', 'DESC', 'pool', 'description', 'targetdep', 'dyndep'}

NINJA_QUOTE_BUILD_PAT = re.compile(r"[$ :\n]")
NINJA_QUOTE_VAR_PAT = re.compile(r"[$ \n]")

def ninja_quote(text: str, is_build_line=False) -> str:
    if is_build_line:
        quote_re = NINJA_QUOTE_BUILD_PAT
    else:
        quote_re = NINJA_QUOTE_VAR_PAT
    # Fast path for when no quoting is necessary
    if not quote_re.search(text):
        return text
    if '\n' in text:
        errmsg = f'''Ninja does not support newlines in rules. The content was:

{text}

Please report this error with a test case to the Meson bug tracker.'''
        raise MesonException(errmsg)
    return quote_re.sub(r'$\g<0>', text)

class TargetDependencyScannerInfo:
    def __init__(self, private_dir: str, source2object: T.Dict[str, str]):
        self.private_dir = private_dir
        self.source2object = source2object

@unique
class Quoting(Enum):
    both = 0
    notShell = 1
    notNinja = 2
    none = 3

class NinjaCommandArg:
    def __init__(self, s, quoting = Quoting.both):
        self.s = s
        self.quoting = quoting

    def __str__(self):
        return self.s

    @staticmethod
    def list(l, q):
        return [NinjaCommandArg(i, q) for i in l]

class NinjaComment:
    def __init__(self, comment):
        self.comment = comment

    def write(self, outfile):
        for l in self.comment.split('\n'):
            outfile.write('# ')
            outfile.write(l)
            outfile.write('\n')
        outfile.write('\n')

class NinjaRule:
    def __init__(self, rule, command, args, description,
                 rspable = False, deps = None, depfile = None, extra = None,
                 rspfile_quote_style: RSPFileSyntax = RSPFileSyntax.GCC):

        def strToCommandArg(c):
            if isinstance(c, NinjaCommandArg):
                return c

            # deal with common cases here, so we don't have to explicitly
            # annotate the required quoting everywhere
            if c == '&&':
                # shell constructs shouldn't be shell quoted
                return NinjaCommandArg(c, Quoting.notShell)
            if c.startswith('$'):
                var = re.search(r'\$\{?(\w*)\}?', c).group(1)
                if var not in raw_names:
                    # ninja variables shouldn't be ninja quoted, and their value
                    # is already shell quoted
                    return NinjaCommandArg(c, Quoting.none)
                else:
                    # shell quote the use of ninja variables whose value must
                    # not be shell quoted (as it also used by ninja)
                    return NinjaCommandArg(c, Quoting.notNinja)

            return NinjaCommandArg(c)

        self.name = rule
        self.command = list(map(strToCommandArg, command))  # includes args which never go into a rspfile
        self.args = list(map(strToCommandArg, args))  # args which will go into a rspfile, if used
        self.description = description
        self.deps = deps  # depstyle 'gcc' or 'msvc'
        self.depfile = depfile
        self.extra = extra
        self.rspable = rspable  # if a rspfile can be used
        self.refcount = 0
        self.rsprefcount = 0
        self.rspfile_quote_style = rspfile_quote_style

        if self.depfile == '$DEPFILE':
            self.depfile += '_UNQUOTED'

    @staticmethod
    def _quoter(x, qf = quote_func):
        if isinstance(x, NinjaCommandArg):
            if x.quoting == Quoting.none:
                return x.s
            elif x.quoting == Quoting.notNinja:
                return qf(x.s)
            elif x.quoting == Quoting.notShell:
                return ninja_quote(x.s)
            # fallthrough
        return ninja_quote(qf(str(x)))

    def write(self, outfile):
        if self.rspfile_quote_style is RSPFileSyntax.MSVC:
            rspfile_quote_func = cmd_quote
        else:
            rspfile_quote_func = gcc_rsp_quote

        def rule_iter():
            if self.refcount:
                yield ''
            if self.rsprefcount:
                yield '_RSP'

        for rsp in rule_iter():
            outfile.write(f'rule {self.name}{rsp}\n')
            if rsp == '_RSP':
                outfile.write(' command = {} @$out.rsp\n'.format(' '.join([self._quoter(x) for x in self.command])))
                outfile.write(' rspfile = $out.rsp\n')
                outfile.write(' rspfile_content = {}\n'.format(' '.join([self._quoter(x, rspfile_quote_func) for x in self.args])))
            else:
                outfile.write(' command = {}\n'.format(' '.join([self._quoter(x) for x in self.command + self.args])))
            if self.deps:
                outfile.write(f' deps = {self.deps}\n')
            if self.depfile:
                outfile.write(f' depfile = {self.depfile}\n')
            outfile.write(f' description = {self.description}\n')
            if self.extra:
                for l in self.extra.split('\n'):
                    outfile.write(' ')
                    outfile.write(l)
                    outfile.write('\n')
            outfile.write('\n')

    def length_estimate(self, infiles, outfiles, elems):
        # determine variables
        # this order of actions only approximates ninja's scoping rules, as
        # documented at: https://ninja-build.org/manual.html#ref_scope
        ninja_vars = {}
        for e in elems:
            (name, value) = e
            ninja_vars[name] = value
        ninja_vars['deps'] = self.deps
        ninja_vars['depfile'] = self.depfile
        ninja_vars['in'] = infiles
        ninja_vars['out'] = outfiles

        # expand variables in command
        command = ' '.join([self._quoter(x) for x in self.command + self.args])
        estimate = len(command)
        for m in re.finditer(r'(\${\w+}|\$\w+)?[^$]*', command):
            if m.start(1) != -1:
                estimate -= m.end(1) - m.start(1) + 1
                chunk = m.group(1)
                if chunk[1] == '{':
                    chunk = chunk[2:-1]
                else:
                    chunk = chunk[1:]
                chunk = ninja_vars.get(chunk, []) # undefined ninja variables are empty
                estimate += len(' '.join(chunk))

        # determine command length
        return estimate

class NinjaBuildElement:
    def __init__(self, all_outputs, outfilenames, rulename, infilenames, implicit_outs=None):
        self.implicit_outfilenames = implicit_outs or []
        if isinstance(outfilenames, str):
            self.outfilenames = [outfilenames]
        else:
            self.outfilenames = outfilenames
        assert isinstance(rulename, str)
        self.rulename = rulename
        if isinstance(infilenames, str):
            self.infilenames = [infilenames]
        else:
            self.infilenames = infilenames
        self.deps = OrderedSet()
        self.orderdeps = OrderedSet()
        self.elems = []
        self.all_outputs = all_outputs

    def add_dep(self, dep):
        if isinstance(dep, list):
            self.deps.update(dep)
        else:
            self.deps.add(dep)

    def add_orderdep(self, dep):
        if isinstance(dep, list):
            self.orderdeps.update(dep)
        else:
            self.orderdeps.add(dep)

    def add_item(self, name, elems):
        # Always convert from GCC-style argument naming to the naming used by the
        # current compiler. Also filter system include paths, deduplicate, etc.
        if isinstance(elems, CompilerArgs):
            elems = elems.to_native()
        if isinstance(elems, str):
            elems = [elems]
        self.elems.append((name, elems))

        if name == 'DEPFILE':
            self.elems.append((name + '_UNQUOTED', elems))

    def _should_use_rspfile(self):
        # 'phony' is a rule built-in to ninja
        if self.rulename == 'phony':
            return False

        if not self.rule.rspable:
            return False

        infilenames = ' '.join([ninja_quote(i, True) for i in self.infilenames])
        outfilenames = ' '.join([ninja_quote(i, True) for i in self.outfilenames])

        return self.rule.length_estimate(infilenames,
                                         outfilenames,
                                         self.elems) >= rsp_threshold

    def count_rule_references(self):
        if self.rulename != 'phony':
            if self._should_use_rspfile():
                self.rule.rsprefcount += 1
            else:
                self.rule.refcount += 1

    def write(self, outfile):
        self.check_outputs()
        ins = ' '.join([ninja_quote(i, True) for i in self.infilenames])
        outs = ' '.join([ninja_quote(i, True) for i in self.outfilenames])
        implicit_outs = ' '.join([ninja_quote(i, True) for i in self.implicit_outfilenames])
        if implicit_outs:
            implicit_outs = ' | ' + implicit_outs
        use_rspfile = self._should_use_rspfile()
        if use_rspfile:
            rulename = self.rulename + '_RSP'
            mlog.debug(f'Command line for building {self.outfilenames} is long, using a response file')
        else:
            rulename = self.rulename
        line = f'build {outs}{implicit_outs}: {rulename} {ins}'
        if len(self.deps) > 0:
            line += ' | ' + ' '.join([ninja_quote(x, True) for x in sorted(self.deps)])
        if len(self.orderdeps) > 0:
            line += ' || ' + ' '.join([ninja_quote(x, True) for x in sorted(self.orderdeps)])
        line += '\n'
        # This is the only way I could find to make this work on all
        # platforms including Windows command shell. Slash is a dir separator
        # on Windows, too, so all characters are unambiguous and, more importantly,
        # do not require quoting, unless explicitly specified, which is necessary for
        # the csc compiler.
        line = line.replace('\\', '/')
        if mesonlib.is_windows():
            # Support network paths as backslash, otherwise they are interpreted as
            # arguments for compile/link commands when using MSVC
            line = ' '.join(
                (l.replace('//', '\\\\', 1) if l.startswith('//') else l)
                for l in line.split(' ')
            )
        outfile.write(line)

        if use_rspfile:
            if self.rule.rspfile_quote_style is RSPFileSyntax.MSVC:
                qf = cmd_quote
            else:
                qf = gcc_rsp_quote
        else:
            qf = quote_func

        for e in self.elems:
            (name, elems) = e
            should_quote = name not in raw_names
            line = f' {name} = '
            newelems = []
            for i in elems:
                if not should_quote or i == '&&': # Hackety hack hack
                    newelems.append(ninja_quote(i))
                else:
                    newelems.append(ninja_quote(qf(i)))
            line += ' '.join(newelems)
            line += '\n'
            outfile.write(line)
        outfile.write('\n')

    def check_outputs(self):
        for n in self.outfilenames:
            if n in self.all_outputs:
                raise MesonException(f'Multiple producers for Ninja target "{n}". Please rename your targets.')
            self.all_outputs[n] = True

class NinjaBackend(backends.Backend):

    def __init__(self, build: T.Optional[build.Build], interpreter: T.Optional[Interpreter]):
        super().__init__(build, interpreter)
        self.name = 'ninja'
        self.ninja_filename = 'build.ninja'
        self.fortran_deps = {}
        self.all_outputs = {}
        self.introspection_data = {}
        self.created_llvm_ir_rule = PerMachine(False, False)

    def create_target_alias(self, to_target):
        # We need to use aliases for targets that might be used as directory
        # names to workaround a Ninja bug that breaks `ninja -t clean`.
        # This is used for 'reserved' targets such as 'test', 'install',
        # 'benchmark', etc, and also for RunTargets.
        # https://github.com/mesonbuild/meson/issues/1644
        if not to_target.startswith('meson-'):
            raise AssertionError(f'Invalid usage of create_target_alias with {to_target!r}')
        from_target = to_target[len('meson-'):]
        elem = NinjaBuildElement(self.all_outputs, from_target, 'phony', to_target)
        self.add_build(elem)

    def detect_vs_dep_prefix(self, tempfilename):
        '''VS writes its dependency in a locale dependent format.
        Detect the search prefix to use.'''
        # TODO don't hard-code host
        for compiler in self.environment.coredata.compilers.host.values():
            # Have to detect the dependency format

            # IFort on windows is MSVC like, but doesn't have /showincludes
            if isinstance(compiler, FortranCompiler):
                continue
            if isinstance(compiler, PGICCompiler) and mesonlib.is_windows():
                # for the purpose of this function, PGI doesn't act enough like MSVC
                return open(tempfilename, 'a', encoding='utf-8')
            if isinstance(compiler, VisualStudioLikeCompiler):
                break
        else:
            # None of our compilers are MSVC, we're done.
            return open(tempfilename, 'a', encoding='utf-8')
        filename = os.path.join(self.environment.get_scratch_dir(),
                                'incdetect.c')
        with open(filename, 'w', encoding='utf-8') as f:
            f.write(dedent('''\
                #include<stdio.h>
                int dummy;
            '''))

        # The output of cl dependency information is language
        # and locale dependent. Any attempt at converting it to
        # Python strings leads to failure. We _must_ do this detection
        # in raw byte mode and write the result in raw bytes.
        pc = subprocess.Popen(compiler.get_exelist() +
                              ['/showIncludes', '/c', 'incdetect.c'],
                              cwd=self.environment.get_scratch_dir(),
                              stdout=subprocess.PIPE, stderr=subprocess.PIPE)
        (stdout, stderr) = pc.communicate()

        # We want to match 'Note: including file: ' in the line
        # 'Note: including file: d:\MyDir\include\stdio.h', however
        # different locales have different messages with a different
        # number of colons. Match up to the the drive name 'd:\'.
        # When used in cross compilation, the path separator is a
        # forward slash rather than a backslash so handle both; i.e.
        # the path is /MyDir/include/stdio.h.
        # With certain cross compilation wrappings of MSVC, the paths
        # use backslashes, but without the leading drive name, so
        # allow the path to start with any path separator, i.e.
        # \MyDir\include\stdio.h.
        matchre = re.compile(rb"^(.*\s)([a-zA-Z]:\\|[\\\/]).*stdio.h$")

        def detect_prefix(out):
            for line in re.split(rb'\r?\n', out):
                match = matchre.match(line)
                if match:
                    with open(tempfilename, 'ab') as binfile:
                        binfile.write(b'msvc_deps_prefix = ' + match.group(1) + b'\n')
                    return open(tempfilename, 'a', encoding='utf-8')
            return None

        # Some cl wrappers (e.g. Squish Coco) output dependency info
        # to stderr rather than stdout
        result = detect_prefix(stdout) or detect_prefix(stderr)
        if result:
            return result

        raise MesonException('Could not determine vs dep dependency prefix string.')

    def generate(self):
        ninja = environment.detect_ninja_command_and_version(log=True)
        if self.build.need_vsenv:
            builddir = Path(self.environment.get_build_dir())
            try:
                # For prettier printing, reduce to a relative path. If
                # impossible (e.g., because builddir and cwd are on
                # different Windows drives), skip and use the full path.
                builddir = builddir.relative_to(Path.cwd())
            except ValueError:
                pass
            meson_command = mesonlib.join_args(mesonlib.get_meson_command())
            mlog.log()
            mlog.log('Visual Studio environment is needed to run Ninja. It is recommended to use Meson wrapper:')
            mlog.log(f'{meson_command} compile -C {builddir}')
        if ninja is None:
            raise MesonException('Could not detect Ninja v1.8.2 or newer')
        (self.ninja_command, self.ninja_version) = ninja
        outfilename = os.path.join(self.environment.get_build_dir(), self.ninja_filename)
        tempfilename = outfilename + '~'
        with open(tempfilename, 'w', encoding='utf-8') as outfile:
            outfile.write(f'# This is the build file for project "{self.build.get_project()}"\n')
            outfile.write('# It is autogenerated by the Meson build system.\n')
            outfile.write('# Do not edit by hand.\n\n')
            outfile.write('ninja_required_version = 1.8.2\n\n')

            num_pools = self.environment.coredata.options[OptionKey('backend_max_links')].value
            if num_pools > 0:
                outfile.write(f'''pool link_pool
  depth = {num_pools}

''')

        with self.detect_vs_dep_prefix(tempfilename) as outfile:
            self.generate_rules()

            self.build_elements = []
            self.generate_phony()
            self.add_build_comment(NinjaComment('Build rules for targets'))
            for t in ProgressBar(self.build.get_targets().values(), desc='Generating targets'):
                self.generate_target(t)
            self.add_build_comment(NinjaComment('Test rules'))
            self.generate_tests()
            self.add_build_comment(NinjaComment('Install rules'))
            self.generate_install()
            self.generate_dist()
            key = OptionKey('b_coverage')
            if (key in self.environment.coredata.options and
                    self.environment.coredata.options[key].value):
                gcovr_exe, gcovr_version, lcov_exe, genhtml_exe, _ = environment.find_coverage_tools()
                if gcovr_exe or (lcov_exe and genhtml_exe):
                    self.add_build_comment(NinjaComment('Coverage rules'))
                    self.generate_coverage_rules(gcovr_exe, gcovr_version)
                else:
                    # FIXME: since we explicitly opted in, should this be an error?
                    # The docs just say these targets will be created "if possible".
                    mlog.warning('Need gcovr or lcov/genhtml to generate any coverage reports')
            self.add_build_comment(NinjaComment('Suffix'))
            self.generate_utils()
            self.generate_ending()

            self.write_rules(outfile)
            self.write_builds(outfile)

            default = 'default all\n\n'
            outfile.write(default)
        # Only overwrite the old build file after the new one has been
        # fully created.
        os.replace(tempfilename, outfilename)
        mlog.cmd_ci_include(outfilename)  # For CI debugging
        # Refresh Ninja's caches. https://github.com/ninja-build/ninja/pull/1685
        if mesonlib.version_compare(self.ninja_version, '>=1.10.0') and os.path.exists('.ninja_deps'):
            subprocess.call(self.ninja_command + ['-t', 'restat'])
            subprocess.call(self.ninja_command + ['-t', 'cleandead'])
        self.generate_compdb()

    # http://clang.llvm.org/docs/JSONCompilationDatabase.html
    def generate_compdb(self):
        rules = []
        # TODO: Rather than an explicit list here, rules could be marked in the
        # rule store as being wanted in compdb
        for for_machine in MachineChoice:
            for lang in self.environment.coredata.compilers[for_machine]:
                rules += [f"{rule}{ext}" for rule in [self.get_compiler_rule_name(lang, for_machine)]
                          for ext in ['', '_RSP']]
                rules += [f"{rule}{ext}" for rule in [self.get_pch_rule_name(lang, for_machine)]
                          for ext in ['', '_RSP']]
        compdb_options = ['-x'] if mesonlib.version_compare(self.ninja_version, '>=1.9') else []
        ninja_compdb = self.ninja_command + ['-t', 'compdb'] + compdb_options + rules
        builddir = self.environment.get_build_dir()
        try:
            jsondb = subprocess.check_output(ninja_compdb, cwd=builddir)
            with open(os.path.join(builddir, 'compile_commands.json'), 'wb') as f:
                f.write(jsondb)
        except Exception:
            mlog.warning('Could not create compilation database.', fatal=False)

    # Get all generated headers. Any source file might need them so
    # we need to add an order dependency to them.
    def get_generated_headers(self, target):
        if hasattr(target, 'cached_generated_headers'):
            return target.cached_generated_headers
        header_deps = []
        # XXX: Why don't we add deps to CustomTarget headers here?
        for genlist in target.get_generated_sources():
            if isinstance(genlist, (build.CustomTarget, build.CustomTargetIndex)):
                continue
            for src in genlist.get_outputs():
                if self.environment.is_header(src):
                    header_deps.append(self.get_target_generated_dir(target, genlist, src))
        if 'vala' in target.compilers and not isinstance(target, build.Executable):
            vala_header = File.from_built_file(self.get_target_dir(target), target.vala_header)
            header_deps.append(vala_header)
        # Recurse and find generated headers
        for dep in itertools.chain(target.link_targets, target.link_whole_targets):
            if isinstance(dep, (build.StaticLibrary, build.SharedLibrary)):
                header_deps += self.get_generated_headers(dep)
        target.cached_generated_headers = header_deps
        return header_deps

    def get_target_generated_sources(self, target: build.BuildTarget) -> T.MutableMapping[str, File]:
        """
        Returns a dictionary with the keys being the path to the file
        (relative to the build directory) of that type and the value
        being the GeneratorList or CustomTarget that generated it.
        """
        srcs: T.MutableMapping[str, File] = OrderedDict()
        for gensrc in target.get_generated_sources():
            for s in gensrc.get_outputs():
                f = self.get_target_generated_dir(target, gensrc, s)
                srcs[f] = s
        return srcs

    def get_target_sources(self, target: build.BuildTarget) -> T.MutableMapping[str, File]:
        srcs: T.MutableMapping[str, File] = OrderedDict()
        for s in target.get_sources():
            # BuildTarget sources are always mesonlib.File files which are
            # either in the source root, or generated with configure_file and
            # in the build root
            if not isinstance(s, File):
                raise InvalidArguments(f'All sources in target {s!r} must be of type mesonlib.File')
            f = s.rel_to_builddir(self.build_to_src)
            srcs[f] = s
        return srcs

    def get_target_source_can_unity(self, target, source):
        if isinstance(source, File):
            source = source.fname
        if self.environment.is_llvm_ir(source) or \
           self.environment.is_assembly(source):
            return False
        suffix = os.path.splitext(source)[1][1:].lower()
        for lang in backends.LANGS_CANT_UNITY:
            if lang not in target.compilers:
                continue
            if suffix in target.compilers[lang].file_suffixes:
                return False
        return True

    def create_target_source_introspection(self, target: build.Target, comp: compilers.Compiler, parameters, sources, generated_sources):
        '''
        Adds the source file introspection information for a language of a target

        Internal introspection storage formart:
        self.introspection_data = {
            '<target ID>': {
                <id tuple>: {
                    'language: 'lang',
                    'compiler': ['comp', 'exe', 'list'],
                    'parameters': ['UNIQUE', 'parameter', 'list'],
                    'sources': [],
                    'generated_sources': [],
                }
            }
        }
        '''
        tid = target.get_id()
        lang = comp.get_language()
        tgt = self.introspection_data[tid]
        # Find an existing entry or create a new one
        id_hash = (lang, tuple(parameters))
        src_block = tgt.get(id_hash, None)
        if src_block is None:
            # Convert parameters
            if isinstance(parameters, CompilerArgs):
                parameters = parameters.to_native(copy=True)
            parameters = comp.compute_parameters_with_absolute_paths(parameters, self.build_dir)
            # The new entry
            src_block = {
                'language': lang,
                'compiler': comp.get_exelist(),
                'parameters': parameters,
                'sources': [],
                'generated_sources': [],
            }
            tgt[id_hash] = src_block
        # Make source files absolute
        sources = [x.absolute_path(self.source_dir, self.build_dir) if isinstance(x, File) else os.path.normpath(os.path.join(self.build_dir, x))
                   for x in sources]
        generated_sources = [x.absolute_path(self.source_dir, self.build_dir) if isinstance(x, File) else os.path.normpath(os.path.join(self.build_dir, x))
                             for x in generated_sources]
        # Add the source files
        src_block['sources'] += sources
        src_block['generated_sources'] += generated_sources

    def generate_target(self, target):
        try:
            if isinstance(target, build.BuildTarget):
                os.makedirs(self.get_target_private_dir_abs(target))
        except FileExistsError:
            pass
        if isinstance(target, build.CustomTarget):
            self.generate_custom_target(target)
        if isinstance(target, build.RunTarget):
            self.generate_run_target(target)
        compiled_sources = []
        source2object = {}
        name = target.get_id()
        if name in self.processed_targets:
            return
        self.processed_targets.add(name)
        # Initialize an empty introspection source list
        self.introspection_data[name] = {}
        # Generate rules for all dependency targets
        self.process_target_dependencies(target)

        self.generate_shlib_aliases(target, self.get_target_dir(target))

        # If target uses a language that cannot link to C objects,
        # just generate for that language and return.
        if isinstance(target, build.Jar):
            self.generate_jar_target(target)
            return
        if target.uses_rust():
            self.generate_rust_target(target)
            return
        if 'cs' in target.compilers:
            self.generate_cs_target(target)
            return
        if 'swift' in target.compilers:
            self.generate_swift_target(target)
            return

        # Pre-existing target C/C++ sources to be built; dict of full path to
        # source relative to build root and the original File object.
        target_sources: T.MutableMapping[str, File]

        # GeneratedList and CustomTarget sources to be built; dict of the full
        # path to source relative to build root and the generating target/list
        generated_sources: T.MutableMapping[str, File]

        # List of sources that have been transpiled from a DSL (like Vala) into
        # a language that is haneled below, such as C or C++
        transpiled_sources: T.List[str]

        if 'vala' in target.compilers:
            # Sources consumed by valac are filtered out. These only contain
            # C/C++ sources, objects, generated libs, and unknown sources now.
            target_sources, generated_sources, \
                transpiled_sources = self.generate_vala_compile(target)
        elif 'cython' in target.compilers:
            target_sources, generated_sources, \
                transpiled_sources = self.generate_cython_transpile(target)
        else:
            target_sources = self.get_target_sources(target)
            generated_sources = self.get_target_generated_sources(target)
            transpiled_sources = []
        self.scan_fortran_module_outputs(target)
        # Generate rules for GeneratedLists
        self.generate_generator_list_rules(target)

        # Generate rules for building the remaining source files in this target
        outname = self.get_target_filename(target)
        obj_list = []
        is_unity = self.is_unity(target)
        header_deps = []
        unity_src = []
        unity_deps = [] # Generated sources that must be built before compiling a Unity target.
        header_deps += self.get_generated_headers(target)

        if is_unity:
            # Warn about incompatible sources if a unity build is enabled
            langs = set(target.compilers.keys())
            langs_cant = langs.intersection(backends.LANGS_CANT_UNITY)
            if langs_cant:
                langs_are = langs = ', '.join(langs_cant).upper()
                langs_are += ' are' if len(langs_cant) > 1 else ' is'
                msg = f'{langs_are} not supported in Unity builds yet, so {langs} ' \
                      f'sources in the {target.name!r} target will be compiled normally'
                mlog.log(mlog.red('FIXME'), msg)

        # Get a list of all generated headers that will be needed while building
        # this target's sources (generated sources and pre-existing sources).
        # This will be set as dependencies of all the target's sources. At the
        # same time, also deal with generated sources that need to be compiled.
        generated_source_files = []
        for rel_src in generated_sources.keys():
            dirpart, fnamepart = os.path.split(rel_src)
            raw_src = File(True, dirpart, fnamepart)
            if self.environment.is_source(rel_src) and not self.environment.is_header(rel_src):
                if is_unity and self.get_target_source_can_unity(target, rel_src):
                    unity_deps.append(raw_src)
                    abs_src = os.path.join(self.environment.get_build_dir(), rel_src)
                    unity_src.append(abs_src)
                else:
                    generated_source_files.append(raw_src)
            elif self.environment.is_object(rel_src):
                obj_list.append(rel_src)
            elif self.environment.is_library(rel_src) or modules.is_module_library(rel_src):
                pass
            else:
                # Assume anything not specifically a source file is a header. This is because
                # people generate files with weird suffixes (.inc, .fh) that they then include
                # in their source files.
                header_deps.append(raw_src)
        # These are the generated source files that need to be built for use by
        # this target. We create the Ninja build file elements for this here
        # because we need `header_deps` to be fully generated in the above loop.
        for src in generated_source_files:
            if self.environment.is_llvm_ir(src):
                o, s = self.generate_llvm_ir_compile(target, src)
            else:
                o, s = self.generate_single_compile(target, src, True,
                                                    order_deps=header_deps)
            compiled_sources.append(s)
            source2object[s] = o
            obj_list.append(o)

        use_pch = self.environment.coredata.options.get(OptionKey('b_pch'))
        if use_pch and target.has_pch():
            pch_objects = self.generate_pch(target, header_deps=header_deps)
        else:
            pch_objects = []

        # Generate compilation targets for C sources generated from Vala
        # sources. This can be extended to other $LANG->C compilers later if
        # necessary. This needs to be separate for at least Vala
        #
        # Do not try to unity-build the generated c files from vala, as these
        # often contain duplicate symbols and will fail to compile properly
        vala_generated_source_files = []
        for src in transpiled_sources:
            dirpart, fnamepart = os.path.split(src)
            raw_src = File(True, dirpart, fnamepart)
            # Generated targets are ordered deps because the must exist
            # before the sources compiling them are used. After the first
            # compile we get precise dependency info from dep files.
            # This should work in all cases. If it does not, then just
            # move them from orderdeps to proper deps.
            if self.environment.is_header(src):
                header_deps.append(raw_src)
            else:
                # We gather all these and generate compile rules below
                # after `header_deps` (above) is fully generated
                vala_generated_source_files.append(raw_src)
        for src in vala_generated_source_files:
            # Passing 'vala' here signifies that we want the compile
            # arguments to be specialized for C code generated by
            # valac. For instance, no warnings should be emitted.
            o, s = self.generate_single_compile(target, src, 'vala', [], header_deps)
            obj_list.append(o)

        # Generate compile targets for all the pre-existing sources for this target
        for src in target_sources.values():
            if not self.environment.is_header(src):
                if self.environment.is_llvm_ir(src):
                    o, s = self.generate_llvm_ir_compile(target, src)
                    obj_list.append(o)
                elif is_unity and self.get_target_source_can_unity(target, src):
                    abs_src = os.path.join(self.environment.get_build_dir(),
                                           src.rel_to_builddir(self.build_to_src))
                    unity_src.append(abs_src)
                else:
                    o, s = self.generate_single_compile(target, src, False, [], header_deps)
                    obj_list.append(o)
                    compiled_sources.append(s)
                    source2object[s] = o

        obj_list += self.flatten_object_list(target)
        if is_unity:
            for src in self.generate_unity_files(target, unity_src):
                o, s = self.generate_single_compile(target, src, True, unity_deps + header_deps)
                obj_list.append(o)
                compiled_sources.append(s)
                source2object[s] = o
        linker, stdlib_args = self.determine_linker_and_stdlib_args(target)
        if isinstance(target, build.StaticLibrary) and target.prelink:
            final_obj_list = self.generate_prelink(target, obj_list)
        else:
            final_obj_list = obj_list
        elem = self.generate_link(target, outname, final_obj_list, linker, pch_objects, stdlib_args=stdlib_args)
        self.generate_dependency_scan_target(target, compiled_sources, source2object, generated_source_files)
        self.add_build(elem)

    def should_use_dyndeps_for_target(self, target: 'build.BuildTarget') -> bool:
        if mesonlib.version_compare(self.ninja_version, '<1.10.0'):
            return False
        if 'fortran' in target.compilers:
            return True
        if 'cpp' not in target.compilers:
            return False
        # Currently only the preview version of Visual Studio is supported.
        cpp = target.compilers['cpp']
        if cpp.get_id() != 'msvc':
            return False
        cppversion = self.environment.coredata.options[OptionKey('std', machine=target.for_machine, lang='cpp')].value
        if cppversion not in ('latest', 'c++latest', 'vc++latest'):
            return False
        if not mesonlib.current_vs_supports_modules():
            return False
        if mesonlib.version_compare(cpp.version, '<19.28.28617'):
            return False
        return True

    def generate_dependency_scan_target(self, target, compiled_sources, source2object, generated_source_files: T.List[mesonlib.File]):
        if not self.should_use_dyndeps_for_target(target):
            return
        depscan_file = self.get_dep_scan_file_for(target)
        pickle_base = target.name + '.dat'
        pickle_file = os.path.join(self.get_target_private_dir(target), pickle_base).replace('\\', '/')
        pickle_abs = os.path.join(self.get_target_private_dir_abs(target), pickle_base).replace('\\', '/')
        json_abs = os.path.join(self.get_target_private_dir_abs(target), f'{target.name}-deps.json').replace('\\', '/')
        rule_name = 'depscan'
        scan_sources = self.select_sources_to_scan(compiled_sources)

        # Dump the sources as a json list. This avoids potential probllems where
        # the number of sources passed to depscan exceedes the limit imposed by
        # the OS.
        with open(json_abs, 'w', encoding='utf-8') as f:
            json.dump(scan_sources, f)
        elem = NinjaBuildElement(self.all_outputs, depscan_file, rule_name, json_abs)
        elem.add_item('picklefile', pickle_file)
        # Add any generated outputs to the order deps of the scan target, so
        # that those sources are present
        for g in generated_source_files:
            elem.orderdeps.add(g.relative_name())
        scaninfo = TargetDependencyScannerInfo(self.get_target_private_dir(target), source2object)
        with open(pickle_abs, 'wb') as p:
            pickle.dump(scaninfo, p)
        self.add_build(elem)

    def select_sources_to_scan(self, compiled_sources):
        # in practice pick up C++ and Fortran files. If some other language
        # requires scanning (possibly Java to deal with inner class files)
        # then add them here.
        all_suffixes = set(compilers.lang_suffixes['cpp']) | set(compilers.lang_suffixes['fortran'])
        selected_sources = []
        for source in compiled_sources:
            ext = os.path.splitext(source)[1][1:].lower()
            if ext in all_suffixes:
                selected_sources.append(source)
        return selected_sources

    def process_target_dependencies(self, target):
        for t in target.get_dependencies():
            if t.get_id() not in self.processed_targets:
                self.generate_target(t)

    def custom_target_generator_inputs(self, target):
        for s in target.sources:
            if isinstance(s, build.GeneratedList):
                self.generate_genlist_for_target(s, target)

    def unwrap_dep_list(self, target):
        deps = []
        for i in target.get_dependencies():
            # FIXME, should not grab element at zero but rather expand all.
            if isinstance(i, list):
                i = i[0]
            # Add a dependency on all the outputs of this target
            for output in i.get_outputs():
                deps.append(os.path.join(self.get_target_dir(i), output))
        return deps

    def generate_custom_target(self, target):
        self.custom_target_generator_inputs(target)
        (srcs, ofilenames, cmd) = self.eval_custom_target_command(target)
        deps = self.unwrap_dep_list(target)
        deps += self.get_custom_target_depend_files(target)
        if target.build_always_stale:
            deps.append('PHONY')
        if target.depfile is None:
            rulename = 'CUSTOM_COMMAND'
        else:
            rulename = 'CUSTOM_COMMAND_DEP'
        elem = NinjaBuildElement(self.all_outputs, ofilenames, rulename, srcs)
        elem.add_dep(deps)
        for d in target.extra_depends:
            # Add a dependency on all the outputs of this target
            for output in d.get_outputs():
                elem.add_dep(os.path.join(self.get_target_dir(d), output))

        cmd, reason = self.as_meson_exe_cmdline(target.command[0], cmd[1:],
                                                extra_bdeps=target.get_transitive_build_target_deps(),
                                                capture=ofilenames[0] if target.capture else None,
                                                feed=srcs[0] if target.feed else None,
                                                env=target.env,
                                                verbose=target.console)
        if reason:
            cmd_type = f' (wrapped by meson {reason})'
        else:
            cmd_type = ''
        if target.depfile is not None:
            depfile = target.get_dep_outname(elem.infilenames)
            rel_dfile = os.path.join(self.get_target_dir(target), depfile)
            abs_pdir = os.path.join(self.environment.get_build_dir(), self.get_target_dir(target))
            os.makedirs(abs_pdir, exist_ok=True)
            elem.add_item('DEPFILE', rel_dfile)
        if target.console:
            elem.add_item('pool', 'console')
        full_name = Path(target.subdir, target.name).as_posix()
        elem.add_item('COMMAND', cmd)
        elem.add_item('description', f'Generating {full_name} with a custom command{cmd_type}')
        self.add_build(elem)
        self.processed_targets.add(target.get_id())

    def build_run_target_name(self, target):
        if target.subproject != '':
            subproject_prefix = f'{target.subproject}@@'
        else:
            subproject_prefix = ''
        return f'{subproject_prefix}{target.name}'

    def generate_run_target(self, target):
        target_name = self.build_run_target_name(target)
        if not target.command:
            # This is an alias target, it has no command, it just depends on
            # other targets.
            elem = NinjaBuildElement(self.all_outputs, target_name, 'phony', [])
        else:
            target_env = self.get_run_target_env(target)
            _, _, cmd = self.eval_custom_target_command(target)
            meson_exe_cmd, reason = self.as_meson_exe_cmdline(target.command[0], cmd[1:],
                                                              force_serialize=True, env=target_env,
                                                              verbose=True)
            cmd_type = f' (wrapped by meson {reason})'
            internal_target_name = f'meson-{target_name}'
            elem = NinjaBuildElement(self.all_outputs, internal_target_name, 'CUSTOM_COMMAND', [])
            elem.add_item('COMMAND', meson_exe_cmd)
            elem.add_item('description', f'Running external command {target.name}{cmd_type}')
            elem.add_item('pool', 'console')
            # Alias that runs the target defined above with the name the user specified
            self.create_target_alias(internal_target_name)
        deps = self.unwrap_dep_list(target)
        deps += self.get_custom_target_depend_files(target)
        elem.add_dep(deps)
        self.add_build(elem)
        self.processed_targets.add(target.get_id())

    def generate_coverage_command(self, elem, outputs):
        targets = self.build.get_targets().values()
        use_llvm_cov = False
        for target in targets:
            if not hasattr(target, 'compilers'):
                continue
            for compiler in target.compilers.values():
                if compiler.get_id() == 'clang' and not compiler.info.is_darwin():
                    use_llvm_cov = True
                    break
        elem.add_item('COMMAND', self.environment.get_build_command() +
                      ['--internal', 'coverage'] +
                      outputs +
                      [self.environment.get_source_dir(),
                       os.path.join(self.environment.get_source_dir(),
                                    self.build.get_subproject_dir()),
                       self.environment.get_build_dir(),
                       self.environment.get_log_dir()] +
                      (['--use_llvm_cov'] if use_llvm_cov else []))

    def generate_coverage_rules(self, gcovr_exe: T.Optional[str], gcovr_version: T.Optional[str]):
        e = NinjaBuildElement(self.all_outputs, 'meson-coverage', 'CUSTOM_COMMAND', 'PHONY')
        self.generate_coverage_command(e, [])
        e.add_item('description', 'Generates coverage reports')
        self.add_build(e)
        # Alias that runs the target defined above
        self.create_target_alias('meson-coverage')
        self.generate_coverage_legacy_rules(gcovr_exe, gcovr_version)

    def generate_coverage_legacy_rules(self, gcovr_exe: T.Optional[str], gcovr_version: T.Optional[str]):
        e = NinjaBuildElement(self.all_outputs, 'meson-coverage-html', 'CUSTOM_COMMAND', 'PHONY')
        self.generate_coverage_command(e, ['--html'])
        e.add_item('description', 'Generates HTML coverage report')
        self.add_build(e)
        # Alias that runs the target defined above
        self.create_target_alias('meson-coverage-html')

        if gcovr_exe:
            e = NinjaBuildElement(self.all_outputs, 'meson-coverage-xml', 'CUSTOM_COMMAND', 'PHONY')
            self.generate_coverage_command(e, ['--xml'])
            e.add_item('description', 'Generates XML coverage report')
            self.add_build(e)
            # Alias that runs the target defined above
            self.create_target_alias('meson-coverage-xml')

            e = NinjaBuildElement(self.all_outputs, 'meson-coverage-text', 'CUSTOM_COMMAND', 'PHONY')
            self.generate_coverage_command(e, ['--text'])
            e.add_item('description', 'Generates text coverage report')
            self.add_build(e)
            # Alias that runs the target defined above
            self.create_target_alias('meson-coverage-text')

            if mesonlib.version_compare(gcovr_version, '>=4.2'):
                e = NinjaBuildElement(self.all_outputs, 'meson-coverage-sonarqube', 'CUSTOM_COMMAND', 'PHONY')
                self.generate_coverage_command(e, ['--sonarqube'])
                e.add_item('description', 'Generates Sonarqube XML coverage report')
                self.add_build(e)
                # Alias that runs the target defined above
                self.create_target_alias('meson-coverage-sonarqube')

    def generate_install(self):
        self.create_install_data_files()
        elem = NinjaBuildElement(self.all_outputs, 'meson-install', 'CUSTOM_COMMAND', 'PHONY')
        elem.add_dep('all')
        elem.add_item('DESC', 'Installing files.')
        elem.add_item('COMMAND', self.environment.get_build_command() + ['install', '--no-rebuild'])
        elem.add_item('pool', 'console')
        self.add_build(elem)
        # Alias that runs the target defined above
        self.create_target_alias('meson-install')

    def generate_tests(self):
        self.serialize_tests()
        cmd = self.environment.get_build_command(True) + ['test', '--no-rebuild']
        if not self.environment.coredata.get_option(OptionKey('stdsplit')):
            cmd += ['--no-stdsplit']
        if self.environment.coredata.get_option(OptionKey('errorlogs')):
            cmd += ['--print-errorlogs']
        elem = NinjaBuildElement(self.all_outputs, 'meson-test', 'CUSTOM_COMMAND', ['all', 'PHONY'])
        elem.add_item('COMMAND', cmd)
        elem.add_item('DESC', 'Running all tests.')
        elem.add_item('pool', 'console')
        self.add_build(elem)
        # Alias that runs the above-defined meson-test target
        self.create_target_alias('meson-test')

        # And then benchmarks.
        cmd = self.environment.get_build_command(True) + [
            'test', '--benchmark', '--logbase',
            'benchmarklog', '--num-processes=1', '--no-rebuild']
        elem = NinjaBuildElement(self.all_outputs, 'meson-benchmark', 'CUSTOM_COMMAND', ['all', 'PHONY'])
        elem.add_item('COMMAND', cmd)
        elem.add_item('DESC', 'Running benchmark suite.')
        elem.add_item('pool', 'console')
        self.add_build(elem)
        # Alias that runs the above-defined meson-benchmark target
        self.create_target_alias('meson-benchmark')

    def generate_rules(self):
        self.rules = []
        self.ruledict = {}

        self.add_rule_comment(NinjaComment('Rules for module scanning.'))
        self.generate_scanner_rules()
        self.add_rule_comment(NinjaComment('Rules for compiling.'))
        self.generate_compile_rules()
        self.add_rule_comment(NinjaComment('Rules for linking.'))
        self.generate_static_link_rules()
        self.generate_dynamic_link_rules()
        self.add_rule_comment(NinjaComment('Other rules'))
        # Ninja errors out if you have deps = gcc but no depfile, so we must
        # have two rules for custom commands.
        self.add_rule(NinjaRule('CUSTOM_COMMAND', ['$COMMAND'], [], '$DESC',
                                extra='restat = 1'))
        self.add_rule(NinjaRule('CUSTOM_COMMAND_DEP', ['$COMMAND'], [], '$DESC',
                                deps='gcc', depfile='$DEPFILE',
                                extra='restat = 1'))
        self.add_rule(NinjaRule('COPY_FILE', self.environment.get_build_command() + ['--internal', 'copy'],
                                ['$in', '$out'], 'Copying $in to $out'))

        c = self.environment.get_build_command() + \
            ['--internal',
             'regenerate',
             self.environment.get_source_dir(),
             self.environment.get_build_dir(),
             '--backend',
             'ninja']
        self.add_rule(NinjaRule('REGENERATE_BUILD',
                                c, [],
                                'Regenerating build files.',
                                extra='generator = 1'))

    def add_rule_comment(self, comment):
        self.rules.append(comment)

    def add_build_comment(self, comment):
        self.build_elements.append(comment)

    def add_rule(self, rule):
        if rule.name in self.ruledict:
            raise MesonException(f'Tried to add rule {rule.name} twice.')
        self.rules.append(rule)
        self.ruledict[rule.name] = rule

    def add_build(self, build):
        self.build_elements.append(build)

        if build.rulename != 'phony':
            # reference rule
            if build.rulename in self.ruledict:
                build.rule = self.ruledict[build.rulename]
            else:
                mlog.warning(f"build statement for {build.outfilenames} references non-existent rule {build.rulename}")

    def write_rules(self, outfile):
        for b in self.build_elements:
            if isinstance(b, NinjaBuildElement):
                b.count_rule_references()

        for r in self.rules:
            r.write(outfile)

    def write_builds(self, outfile):
        for b in ProgressBar(self.build_elements, desc='Writing build.ninja'):
            b.write(outfile)

    def generate_phony(self):
        self.add_build_comment(NinjaComment('Phony build target, always out of date'))
        elem = NinjaBuildElement(self.all_outputs, 'PHONY', 'phony', '')
        self.add_build(elem)

    def generate_jar_target(self, target: build.Jar):
        fname = target.get_filename()
        outname_rel = os.path.join(self.get_target_dir(target), fname)
        src_list = target.get_sources()
        resources = target.get_java_resources()
        class_list = []
        compiler = target.compilers['java']
        c = 'c'
        m = 'm'
        e = ''
        f = 'f'
        main_class = target.get_main_class()
        if main_class != '':
            e = 'e'

        # Add possible java generated files to src list
        generated_sources = self.get_target_generated_sources(target)
        gen_src_list = []
        for rel_src in generated_sources.keys():
            dirpart, fnamepart = os.path.split(rel_src)
            raw_src = File(True, dirpart, fnamepart)
            if rel_src.endswith('.java'):
                gen_src_list.append(raw_src)

        compile_args = self.determine_single_java_compile_args(target, compiler)
        for src in src_list + gen_src_list:
            plain_class_path = self.generate_single_java_compile(src, target, compiler, compile_args)
            class_list.append(plain_class_path)
        class_dep_list = [os.path.join(self.get_target_private_dir(target), i) for i in class_list]
        manifest_path = os.path.join(self.get_target_private_dir(target), 'META-INF', 'MANIFEST.MF')
        manifest_fullpath = os.path.join(self.environment.get_build_dir(), manifest_path)
        os.makedirs(os.path.dirname(manifest_fullpath), exist_ok=True)
        with open(manifest_fullpath, 'w', encoding='utf-8') as manifest:
            if any(target.link_targets):
                manifest.write('Class-Path: ')
                cp_paths = [os.path.join(self.get_target_dir(l), l.get_filename()) for l in target.link_targets]
                manifest.write(' '.join(cp_paths))
            manifest.write('\n')
        jar_rule = 'java_LINKER'
        commands = [c + m + e + f]
        commands.append(manifest_path)
        if e != '':
            commands.append(main_class)
        commands.append(self.get_target_filename(target))
        # Java compilation can produce an arbitrary number of output
        # class files for a single source file. Thus tell jar to just
        # grab everything in the final package.
        commands += ['-C', self.get_target_private_dir(target), '.']
        elem = NinjaBuildElement(self.all_outputs, outname_rel, jar_rule, [])
        elem.add_dep(class_dep_list)
        if resources:
            # Copy all resources into the root of the jar.
            elem.add_orderdep(self.__generate_sources_structure(Path(self.get_target_private_dir(target)), resources)[0])
        elem.add_item('ARGS', commands)
        self.add_build(elem)
        # Create introspection information
        self.create_target_source_introspection(target, compiler, compile_args, src_list, gen_src_list)

    def generate_cs_resource_tasks(self, target):
        args = []
        deps = []
        for r in target.resources:
            rel_sourcefile = os.path.join(self.build_to_src, target.subdir, r)
            if r.endswith('.resources'):
                a = '-resource:' + rel_sourcefile
            elif r.endswith('.txt') or r.endswith('.resx'):
                ofilebase = os.path.splitext(os.path.basename(r))[0] + '.resources'
                ofilename = os.path.join(self.get_target_private_dir(target), ofilebase)
                elem = NinjaBuildElement(self.all_outputs, ofilename, "CUSTOM_COMMAND", rel_sourcefile)
                elem.add_item('COMMAND', ['resgen', rel_sourcefile, ofilename])
                elem.add_item('DESC', f'Compiling resource {rel_sourcefile}')
                self.add_build(elem)
                deps.append(ofilename)
                a = '-resource:' + ofilename
            else:
                raise InvalidArguments(f'Unknown resource file {r}.')
            args.append(a)
        return args, deps

    def generate_cs_target(self, target: build.BuildTarget):
        buildtype = self.get_option_for_target(OptionKey('buildtype'), target)
        fname = target.get_filename()
        outname_rel = os.path.join(self.get_target_dir(target), fname)
        src_list = target.get_sources()
        compiler = target.compilers['cs']
        rel_srcs = [os.path.normpath(s.rel_to_builddir(self.build_to_src)) for s in src_list]
        deps = []
        commands = compiler.compiler_args(target.extra_args.get('cs', []))
        commands += compiler.get_buildtype_args(buildtype)
        commands += compiler.get_optimization_args(self.get_option_for_target(OptionKey('optimization'), target))
        commands += compiler.get_debug_args(self.get_option_for_target(OptionKey('debug'), target))
        if isinstance(target, build.Executable):
            commands.append('-target:exe')
        elif isinstance(target, build.SharedLibrary):
            commands.append('-target:library')
        else:
            raise MesonException('Unknown C# target type.')
        (resource_args, resource_deps) = self.generate_cs_resource_tasks(target)
        commands += resource_args
        deps += resource_deps
        commands += compiler.get_output_args(outname_rel)
        for l in target.link_targets:
            lname = os.path.join(self.get_target_dir(l), l.get_filename())
            commands += compiler.get_link_args(lname)
            deps.append(lname)
        if '-g' in commands:
            outputs = [outname_rel, outname_rel + '.mdb']
        else:
            outputs = [outname_rel]
        generated_sources = self.get_target_generated_sources(target)
        generated_rel_srcs = []
        for rel_src in generated_sources.keys():
            if rel_src.lower().endswith('.cs'):
                generated_rel_srcs.append(os.path.normpath(rel_src))
            deps.append(os.path.normpath(rel_src))

        for dep in target.get_external_deps():
            commands.extend_direct(dep.get_link_args())
        commands += self.build.get_project_args(compiler, target.subproject, target.for_machine)
        commands += self.build.get_global_args(compiler, target.for_machine)

        elem = NinjaBuildElement(self.all_outputs, outputs, self.get_compiler_rule_name('cs', target.for_machine), rel_srcs + generated_rel_srcs)
        elem.add_dep(deps)
        elem.add_item('ARGS', commands)
        self.add_build(elem)

        self.generate_generator_list_rules(target)
        self.create_target_source_introspection(target, compiler, commands, rel_srcs, generated_rel_srcs)

    def determine_single_java_compile_args(self, target, compiler):
        args = []
        args += compiler.get_buildtype_args(self.get_option_for_target(OptionKey('buildtype'), target))
        args += self.build.get_global_args(compiler, target.for_machine)
        args += self.build.get_project_args(compiler, target.subproject, target.for_machine)
        args += target.get_java_args()
        args += compiler.get_output_args(self.get_target_private_dir(target))
        args += target.get_classpath_args()
        curdir = target.get_subdir()
        sourcepath = os.path.join(self.build_to_src, curdir) + os.pathsep
        sourcepath += os.path.normpath(curdir) + os.pathsep
        for i in target.include_dirs:
            for idir in i.get_incdirs():
                sourcepath += os.path.join(self.build_to_src, i.curdir, idir) + os.pathsep
        args += ['-sourcepath', sourcepath]
        return args

    def generate_single_java_compile(self, src, target, compiler, args):
        deps = [os.path.join(self.get_target_dir(l), l.get_filename()) for l in target.link_targets]
        generated_sources = self.get_target_generated_sources(target)
        for rel_src in generated_sources.keys():
            if rel_src.endswith('.java'):
                deps.append(rel_src)
        rel_src = src.rel_to_builddir(self.build_to_src)
        plain_class_path = src.fname[:-4] + 'class'
        rel_obj = os.path.join(self.get_target_private_dir(target), plain_class_path)
        element = NinjaBuildElement(self.all_outputs, rel_obj, self.compiler_to_rule_name(compiler), rel_src)
        element.add_dep(deps)
        element.add_item('ARGS', args)
        self.add_build(element)
        return plain_class_path

    def generate_java_link(self):
        rule = 'java_LINKER'
        command = ['jar', '$ARGS']
        description = 'Creating JAR $out'
        self.add_rule(NinjaRule(rule, command, [], description))

    def determine_dep_vapis(self, target):
        """
        Peek into the sources of BuildTargets we're linking with, and if any of
        them was built with Vala, assume that it also generated a .vapi file of
        the same name as the BuildTarget and return the path to it relative to
        the build directory.
        """
        result = OrderedSet()
        for dep in itertools.chain(target.link_targets, target.link_whole_targets):
            if not dep.is_linkable_target():
                continue
            for i in dep.sources:
                if hasattr(i, 'fname'):
                    i = i.fname
                if i.split('.')[-1] in compilers.lang_suffixes['vala']:
                    vapiname = dep.vala_vapi
                    fullname = os.path.join(self.get_target_dir(dep), vapiname)
                    result.add(fullname)
                    break
        return list(result)

    def split_vala_sources(self, t: build.BuildTarget) -> \
            T.Tuple[T.MutableMapping[str, File], T.MutableMapping[str, File],
                    T.Tuple[T.MutableMapping[str, File], T.MutableMapping]]:
        """
        Splits the target's sources into .vala, .gs, .vapi, and other sources.
        Handles both pre-existing and generated sources.

        Returns a tuple (vala, vapi, others) each of which is a dictionary with
        the keys being the path to the file (relative to the build directory)
        and the value being the object that generated or represents the file.
        """
        vala: T.MutableMapping[str, File] = OrderedDict()
        vapi: T.MutableMapping[str, File] = OrderedDict()
        others: T.MutableMapping[str, File] = OrderedDict()
        othersgen: T.MutableMapping[str, File] = OrderedDict()
        # Split pre-existing sources
        for s in t.get_sources():
            # BuildTarget sources are always mesonlib.File files which are
            # either in the source root, or generated with configure_file and
            # in the build root
            if not isinstance(s, File):
                raise InvalidArguments(f'All sources in target {t!r} must be of type mesonlib.File, not {s!r}')
            f = s.rel_to_builddir(self.build_to_src)
            if s.endswith(('.vala', '.gs')):
                srctype = vala
            elif s.endswith('.vapi'):
                srctype = vapi
            else:
                srctype = others
            srctype[f] = s
        # Split generated sources
        for gensrc in t.get_generated_sources():
            for s in gensrc.get_outputs():
                f = self.get_target_generated_dir(t, gensrc, s)
                if s.endswith(('.vala', '.gs')):
                    srctype = vala
                elif s.endswith('.vapi'):
                    srctype = vapi
                # Generated non-Vala (C/C++) sources. Won't be used for
                # generating the Vala compile rule below.
                else:
                    srctype = othersgen
                # Duplicate outputs are disastrous
                if f in srctype and srctype[f] is not gensrc:
                    msg = 'Duplicate output {0!r} from {1!r} {2!r}; ' \
                          'conflicts with {0!r} from {4!r} {3!r}' \
                          ''.format(f, type(gensrc).__name__, gensrc.name,
                                    srctype[f].name, type(srctype[f]).__name__)
                    raise InvalidArguments(msg)
                # Store 'somefile.vala': GeneratedList (or CustomTarget)
                srctype[f] = gensrc
        return vala, vapi, (others, othersgen)

    def generate_vala_compile(self, target: build.BuildTarget) -> \
            T.Tuple[T.MutableMapping[str, File], T.MutableMapping[str, File], T.List[str]]:
        """Vala is compiled into C. Set up all necessary build steps here."""
        (vala_src, vapi_src, other_src) = self.split_vala_sources(target)
        extra_dep_files = []
        if not vala_src:
            raise InvalidArguments(f'Vala library {target.name!r} has no Vala or Genie source files.')

        valac = target.compilers['vala']
        c_out_dir = self.get_target_private_dir(target)
        # C files generated by valac
        vala_c_src: T.List[str] = []
        # Files generated by valac
        valac_outputs: T.List = []
        # All sources that are passed to valac on the commandline
        all_files = list(vapi_src)
        # Passed as --basedir
        srcbasedir = os.path.join(self.build_to_src, target.get_subdir())
        for (vala_file, gensrc) in vala_src.items():
            all_files.append(vala_file)
            # Figure out where the Vala compiler will write the compiled C file
            #
            # If the Vala file is in a subdir of the build dir (in our case
            # because it was generated/built by something else), and is also
            # a subdir of --basedir (because the builddir is in the source
            # tree, and the target subdir is the source root), the subdir
            # components from the source root till the private builddir will be
            # duplicated inside the private builddir. Otherwise, just the
            # basename will be used.
            #
            # If the Vala file is outside the build directory, the paths from
            # the --basedir till the subdir will be duplicated inside the
            # private builddir.
            if isinstance(gensrc, (build.CustomTarget, build.GeneratedList)) or gensrc.is_built:
                vala_c_file = os.path.splitext(os.path.basename(vala_file))[0] + '.c'
                # Check if the vala file is in a subdir of --basedir
                abs_srcbasedir = os.path.join(self.environment.get_source_dir(), target.get_subdir())
                abs_vala_file = os.path.join(self.environment.get_build_dir(), vala_file)
                if PurePath(os.path.commonpath((abs_srcbasedir, abs_vala_file))) == PurePath(abs_srcbasedir):
                    vala_c_subdir = PurePath(abs_vala_file).parent.relative_to(abs_srcbasedir)
                    vala_c_file = os.path.join(str(vala_c_subdir), vala_c_file)
            else:
                path_to_target = os.path.join(self.build_to_src, target.get_subdir())
                if vala_file.startswith(path_to_target):
                    vala_c_file = os.path.splitext(os.path.relpath(vala_file, path_to_target))[0] + '.c'
                else:
                    vala_c_file = os.path.splitext(os.path.basename(vala_file))[0] + '.c'
            # All this will be placed inside the c_out_dir
            vala_c_file = os.path.join(c_out_dir, vala_c_file)
            vala_c_src.append(vala_c_file)
            valac_outputs.append(vala_c_file)

        args = self.generate_basic_compiler_args(target, valac)
        args += valac.get_colorout_args(self.environment.coredata.options.get(OptionKey('b_colorout')).value)
        # Tell Valac to output everything in our private directory. Sadly this
        # means it will also preserve the directory components of Vala sources
        # found inside the build tree (generated sources).
        args += ['--directory', c_out_dir]
        args += ['--basedir', srcbasedir]
        if target.is_linkable_target():
            # Library name
            args += ['--library', target.name]
            # Outputted header
            hname = os.path.join(self.get_target_dir(target), target.vala_header)
            args += ['--header', hname]
            if self.is_unity(target):
                # Without this the declarations will get duplicated in the .c
                # files and cause a build failure when all of them are
                # #include-d in one .c file.
                # https://github.com/mesonbuild/meson/issues/1969
                args += ['--use-header']
            valac_outputs.append(hname)
            # Outputted vapi file
            vapiname = os.path.join(self.get_target_dir(target), target.vala_vapi)
            # Force valac to write the vapi and gir files in the target build dir.
            # Without this, it will write it inside c_out_dir
            args += ['--vapi', os.path.join('..', target.vala_vapi)]
            valac_outputs.append(vapiname)
            target.outputs += [target.vala_header, target.vala_vapi]
            target.install_tag += ['devel', 'devel']
            # Install header and vapi to default locations if user requests this
            if len(target.install_dir) > 1 and target.install_dir[1] is True:
                target.install_dir[1] = self.environment.get_includedir()
            if len(target.install_dir) > 2 and target.install_dir[2] is True:
                target.install_dir[2] = os.path.join(self.environment.get_datadir(), 'vala', 'vapi')
            # Generate GIR if requested
            if isinstance(target.vala_gir, str):
                girname = os.path.join(self.get_target_dir(target), target.vala_gir)
                args += ['--gir', os.path.join('..', target.vala_gir)]
                valac_outputs.append(girname)
                target.outputs.append(target.vala_gir)
                target.install_tag.append('devel')
                # Install GIR to default location if requested by user
                if len(target.install_dir) > 3 and target.install_dir[3] is True:
                    target.install_dir[3] = os.path.join(self.environment.get_datadir(), 'gir-1.0')
        # Detect gresources and add --gresources arguments for each
        for gensrc in other_src[1].values():
            if isinstance(gensrc, modules.GResourceTarget):
                gres_xml, = self.get_custom_target_sources(gensrc)
                args += ['--gresources=' + gres_xml]
        extra_args = []

        for a in target.extra_args.get('vala', []):
            if isinstance(a, File):
                relname = a.rel_to_builddir(self.build_to_src)
                extra_dep_files.append(relname)
                extra_args.append(relname)
            else:
                extra_args.append(a)
        dependency_vapis = self.determine_dep_vapis(target)
        extra_dep_files += dependency_vapis
        args += extra_args
        element = NinjaBuildElement(self.all_outputs, valac_outputs,
                                    self.compiler_to_rule_name(valac),
                                    all_files + dependency_vapis)
        element.add_item('ARGS', args)
        element.add_dep(extra_dep_files)
        self.add_build(element)
        self.create_target_source_introspection(target, valac, args, all_files, [])
        return other_src[0], other_src[1], vala_c_src

    def generate_cython_transpile(self, target: build.BuildTarget) -> \
            T.Tuple[T.MutableMapping[str, File], T.MutableMapping[str, File], T.List[str]]:
        """Generate rules for transpiling Cython files to C or C++

        XXX: Currently only C is handled.
        """
        static_sources: T.MutableMapping[str, File] = OrderedDict()
        generated_sources: T.MutableMapping[str, File] = OrderedDict()
        cython_sources: T.List[str] = []

        cython = target.compilers['cython']

        opt_proxy = self.get_options_for_target(target)

        args: T.List[str] = []
        args += cython.get_always_args()
        args += cython.get_buildtype_args(self.get_option_for_target(OptionKey('buildtype'), target))
        args += cython.get_debug_args(self.get_option_for_target(OptionKey('debug'), target))
        args += cython.get_optimization_args(self.get_option_for_target(OptionKey('optimization'), target))
        args += cython.get_option_compile_args(opt_proxy)
        args += self.build.get_global_args(cython, target.for_machine)
        args += self.build.get_project_args(cython, target.subproject, target.for_machine)

        ext = opt_proxy[OptionKey('language', machine=target.for_machine, lang='cython')].value

        for src in target.get_sources():
            if src.endswith('.pyx'):
                output = os.path.join(self.get_target_private_dir(target), f'{src}.{ext}')
                args = args.copy()
                args += cython.get_output_args(output)
                element = NinjaBuildElement(
                    self.all_outputs, [output],
                    self.compiler_to_rule_name(cython),
                    [src.absolute_path(self.environment.get_source_dir(), self.environment.get_build_dir())])
                element.add_item('ARGS', args)
                self.add_build(element)
                # TODO: introspection?
                cython_sources.append(output)
            else:
                static_sources[src.rel_to_builddir(self.build_to_src)] = src

        for gen in target.get_generated_sources():
            for ssrc in gen.get_outputs():
                if isinstance(gen, GeneratedList):
                    ssrc = os.path.join(self.get_target_private_dir(target), ssrc)
                else:
                    ssrc = os.path.join(gen.get_subdir(), ssrc)
                if ssrc.endswith('.pyx'):
                    args = args.copy()
                    output = os.path.join(self.get_target_private_dir(target), f'{ssrc}.{ext}')
                    args += cython.get_output_args(output)
                    element = NinjaBuildElement(
                        self.all_outputs, [output],
                        self.compiler_to_rule_name(cython),
                        [ssrc])
                    element.add_item('ARGS', args)
                    self.add_build(element)
                    # TODO: introspection?
                    cython_sources.append(output)
                else:
                    generated_sources[ssrc] = mesonlib.File.from_built_file(gen.get_subdir(), ssrc)

        return static_sources, generated_sources, cython_sources

    def _generate_copy_target(self, src: 'mesonlib.FileOrString', output: Path) -> None:
        """Create a target to copy a source file from one location to another."""
        if isinstance(src, File):
            instr = src.absolute_path(self.environment.source_dir, self.environment.build_dir)
        else:
            instr = src
        elem = NinjaBuildElement(self.all_outputs, [str(output)], 'COPY_FILE', [instr])
        elem.add_orderdep(instr)
        self.add_build(elem)

    def __generate_sources_structure(self, root: Path, structured_sources: build.StructuredSources) -> T.Tuple[T.List[str], T.Optional[str]]:
        first_file: T.Optional[str] = None
        orderdeps: T.List[str] = []
        for path, files in structured_sources.sources.items():
            for file in files:
                if isinstance(file, File):
                    out = root / path / Path(file.fname).name
                    orderdeps.append(str(out))
                    self._generate_copy_target(file, out)
                    if first_file is None:
                        first_file = str(out)
                else:
                    for f in file.get_outputs():
                        out = root / path / f
                        orderdeps.append(str(out))
                        self._generate_copy_target(str(Path(file.subdir) / f), out)
                        if first_file is None:
                            first_file = str(out)
        return orderdeps, first_file

    def generate_rust_target(self, target: build.BuildTarget) -> None:
        rustc = target.compilers['rust']
        # Rust compiler takes only the main file as input and
        # figures out what other files are needed via import
        # statements and magic.
        base_proxy = self.get_options_for_target(target)
        args = rustc.compiler_args()
        # Compiler args for compiling this target
        args += compilers.get_base_compile_args(base_proxy, rustc)
        self.generate_generator_list_rules(target)

        # dependencies need to cause a relink, they're not just for odering
        deps = [
            os.path.join(t.subdir, t.get_filename())
            for t in itertools.chain(target.link_targets, target.link_whole_targets)
        ]

        orderdeps: T.List[str] = []

        main_rust_file = None
        if target.structured_sources:
            if target.structured_sources.needs_copy():
                _ods, main_rust_file = self.__generate_sources_structure(Path(
                    self.get_target_private_dir(target)) / 'structured', target.structured_sources)
                orderdeps.extend(_ods)
            else:
                # The only way to get here is to have only files in the "root"
                # positional argument, which are all generated into the same
                # directory
                g = target.structured_sources.first_file()

                if isinstance(g, File):
                    main_rust_file = g.rel_to_builddir(self.build_to_src)
                elif isinstance(g, GeneratedList):
                    main_rust_file = os.path.join(self.get_target_private_dir(target), g.get_outputs()[0])
                else:
                    main_rust_file = os.path.join(g.get_subdir(), g.get_outputs()[0])

                for f in target.structured_sources.as_list():
                    if isinstance(f, File):
                        orderdeps.append(f.rel_to_builddir(self.build_to_src))
                    else:
                        orderdeps.extend([os.path.join(self.build_to_src, f.subdir, s)
                                          for s in f.get_outputs()])

        for i in target.get_sources():
            if not rustc.can_compile(i):
                raise InvalidArguments(f'Rust target {target.get_basename()} contains a non-rust source file.')
            if main_rust_file is None:
                main_rust_file = i.rel_to_builddir(self.build_to_src)
        for g in target.get_generated_sources():
            for i in g.get_outputs():
                if not rustc.can_compile(i):
                    raise InvalidArguments(f'Rust target {target.get_basename()} contains a non-rust source file.')
                if isinstance(g, GeneratedList):
                    fname = os.path.join(self.get_target_private_dir(target), i)
                else:
                    fname = os.path.join(g.get_subdir(), i)
                if main_rust_file is None:
                    main_rust_file = fname
                orderdeps.append(fname)
        if main_rust_file is None:
            raise RuntimeError('A Rust target has no Rust sources. This is weird. Also a bug. Please report')
        target_name = os.path.join(target.subdir, target.get_filename())
        if isinstance(target, build.Executable):
            cratetype = 'bin'
        elif hasattr(target, 'rust_crate_type'):
            cratetype = target.rust_crate_type
        elif isinstance(target, build.SharedLibrary):
            cratetype = 'dylib'
        elif isinstance(target, build.StaticLibrary):
            cratetype = 'rlib'
        else:
            raise InvalidArguments('Unknown target type for rustc.')
        args.extend(['--crate-type', cratetype])

        # If we're dynamically linking, add those arguments
        #
        # Rust is super annoying, calling -C link-arg foo does not work, it has
        # to be -C link-arg=foo
        if cratetype in {'bin', 'dylib'}:
            args.extend(rustc.get_linker_always_args())

        args += self.generate_basic_compiler_args(target, rustc, False)
        # This matches rustc's default behavior.
        args += ['--crate-name', target.name.replace('-', '_')]
        depfile = os.path.join(target.subdir, target.name + '.d')
        args += ['--emit', f'dep-info={depfile}', '--emit', 'link']
        args += target.get_extra_args('rust')
        args += rustc.get_output_args(os.path.join(target.subdir, target.get_filename()))
        linkdirs = mesonlib.OrderedSet()
        external_deps = target.external_deps.copy()
        for d in itertools.chain(target.link_targets, target.link_whole_targets):
            linkdirs.add(d.subdir)
            if d.uses_rust():
                # specify `extern CRATE_NAME=OUTPUT_FILE` for each Rust
                # dependency, so that collisions with libraries in rustc's
                # sysroot don't cause ambiguity
                args += ['--extern', '{}={}'.format(d.name, os.path.join(d.subdir, d.filename))]
            elif d.typename == 'static library':
                # Rustc doesn't follow Meson's convention that static libraries
                # are called .a, and import libraries are .lib, so we have to
                # manually handle that.
                if rustc.linker.id in {'link', 'lld-link'}:
                    args += ['-C', f'link-arg={self.get_target_filename_for_linking(d)}']
                else:
                    args += ['-l', f'static={d.name}']
                external_deps.extend(d.external_deps)
            else:
                # Rust uses -l for non rust dependencies, but we still need to
                # add dylib=foo
                args += ['-l', f'dylib={d.name}']
        for e in external_deps:
            for a in e.get_link_args():
                if a.endswith(('.dll', '.so', '.dylib')):
                    dir_, lib = os.path.split(a)
                    linkdirs.add(dir_)
                    lib, ext = os.path.splitext(lib)
                    if lib.startswith('lib'):
                        lib = lib[3:]
                    args.extend(['-l', f'dylib={lib}'])
                elif a.startswith('-L'):
                    args.append(a)
                elif a.startswith('-l'):
                    _type = 'static' if e.static else 'dylib'
                    args.extend(['-l', f'{_type}={a[2:]}'])
        for d in linkdirs:
            if d == '':
                d = '.'
            args += ['-L', d]
        has_shared_deps = any(isinstance(dep, build.SharedLibrary) for dep in target.get_dependencies())
        if isinstance(target, build.SharedLibrary) or has_shared_deps:
            # add prefer-dynamic if any of the Rust libraries we link
            # against are dynamic, otherwise we'll end up with
            # multiple implementations of crates
            args += ['-C', 'prefer-dynamic']

            # build the usual rpath arguments as well...

            # Set runtime-paths so we can run executables without needing to set
            # LD_LIBRARY_PATH, etc in the environment. Doesn't work on Windows.
            if has_path_sep(target.name):
                # Target names really should not have slashes in them, but
                # unfortunately we did not check for that and some downstream projects
                # now have them. Once slashes are forbidden, remove this bit.
                target_slashname_workaround_dir = os.path.join(os.path.dirname(target.name),
                                                               self.get_target_dir(target))
            else:
                target_slashname_workaround_dir = self.get_target_dir(target)
            rpath_args, target.rpath_dirs_to_remove = (
                rustc.build_rpath_args(self.environment,
                                       self.environment.get_build_dir(),
                                       target_slashname_workaround_dir,
                                       self.determine_rpath_dirs(target),
                                       target.build_rpath,
                                       target.install_rpath))
            # ... but then add rustc's sysroot to account for rustup
            # installations
            for rpath_arg in rpath_args:
                args += ['-C', 'link-arg=' + rpath_arg + ':' + os.path.join(rustc.get_sysroot(), 'lib')]
        compiler_name = self.get_compiler_rule_name('rust', target.for_machine)
        element = NinjaBuildElement(self.all_outputs, target_name, compiler_name, main_rust_file)
        if orderdeps:
            element.add_orderdep(orderdeps)
        if deps:
            element.add_dep(deps)
        element.add_item('ARGS', args)
        element.add_item('targetdep', depfile)
        element.add_item('cratetype', cratetype)
        self.add_build(element)
        if isinstance(target, build.SharedLibrary):
            self.generate_shsym(target)
        self.create_target_source_introspection(target, rustc, args, [main_rust_file], [])

    @staticmethod
    def get_rule_suffix(for_machine: MachineChoice) -> str:
        return PerMachine('_FOR_BUILD', '')[for_machine]

    @classmethod
    def get_compiler_rule_name(cls, lang: str, for_machine: MachineChoice) -> str:
        return '{}_COMPILER{}'.format(lang, cls.get_rule_suffix(for_machine))

    @classmethod
    def get_pch_rule_name(cls, lang: str, for_machine: MachineChoice) -> str:
        return '{}_PCH{}'.format(lang, cls.get_rule_suffix(for_machine))

    @classmethod
    def compiler_to_rule_name(cls, compiler: Compiler) -> str:
        return cls.get_compiler_rule_name(compiler.get_language(), compiler.for_machine)

    @classmethod
    def compiler_to_pch_rule_name(cls, compiler: Compiler) -> str:
        return cls.get_pch_rule_name(compiler.get_language(), compiler.for_machine)

    def swift_module_file_name(self, target):
        return os.path.join(self.get_target_private_dir(target),
                            self.target_swift_modulename(target) + '.swiftmodule')

    def target_swift_modulename(self, target):
        return target.name

    def determine_swift_dep_modules(self, target):
        result = []
        for l in target.link_targets:
            if self.is_swift_target(l):
                result.append(self.swift_module_file_name(l))
        return result

    def get_swift_link_deps(self, target):
        result = []
        for l in target.link_targets:
            result.append(self.get_target_filename(l))
        return result

    def split_swift_generated_sources(self, target):
        all_srcs = self.get_target_generated_sources(target)
        srcs = []
        others = []
        for i in all_srcs:
            if i.endswith('.swift'):
                srcs.append(i)
            else:
                others.append(i)
        return srcs, others

    def generate_swift_target(self, target):
        module_name = self.target_swift_modulename(target)
        swiftc = target.compilers['swift']
        abssrc = []
        relsrc = []
        abs_headers = []
        header_imports = []
        for i in target.get_sources():
            if swiftc.can_compile(i):
                rels = i.rel_to_builddir(self.build_to_src)
                abss = os.path.normpath(os.path.join(self.environment.get_build_dir(), rels))
                relsrc.append(rels)
                abssrc.append(abss)
            elif self.environment.is_header(i):
                relh = i.rel_to_builddir(self.build_to_src)
                absh = os.path.normpath(os.path.join(self.environment.get_build_dir(), relh))
                abs_headers.append(absh)
                header_imports += swiftc.get_header_import_args(absh)
            else:
                raise InvalidArguments(f'Swift target {target.get_basename()} contains a non-swift source file.')
        os.makedirs(self.get_target_private_dir_abs(target), exist_ok=True)
        compile_args = swiftc.get_compile_only_args()
        compile_args += swiftc.get_optimization_args(self.get_option_for_target(OptionKey('optimization'), target))
        compile_args += swiftc.get_debug_args(self.get_option_for_target(OptionKey('debug'), target))
        compile_args += swiftc.get_module_args(module_name)
        compile_args += self.build.get_project_args(swiftc, target.subproject, target.for_machine)
        compile_args += self.build.get_global_args(swiftc, target.for_machine)
        for i in reversed(target.get_include_dirs()):
            basedir = i.get_curdir()
            for d in i.get_incdirs():
                if d not in ('', '.'):
                    expdir = os.path.join(basedir, d)
                else:
                    expdir = basedir
                srctreedir = os.path.normpath(os.path.join(self.environment.get_build_dir(), self.build_to_src, expdir))
                sargs = swiftc.get_include_args(srctreedir, False)
                compile_args += sargs
        link_args = swiftc.get_output_args(os.path.join(self.environment.get_build_dir(), self.get_target_filename(target)))
        link_args += self.build.get_project_link_args(swiftc, target.subproject, target.for_machine)
        link_args += self.build.get_global_link_args(swiftc, target.for_machine)
        rundir = self.get_target_private_dir(target)
        out_module_name = self.swift_module_file_name(target)
        in_module_files = self.determine_swift_dep_modules(target)
        abs_module_dirs = self.determine_swift_dep_dirs(target)
        module_includes = []
        for x in abs_module_dirs:
            module_includes += swiftc.get_include_args(x, False)
        link_deps = self.get_swift_link_deps(target)
        abs_link_deps = [os.path.join(self.environment.get_build_dir(), x) for x in link_deps]
        for d in target.link_targets:
            reldir = self.get_target_dir(d)
            if reldir == '':
                reldir = '.'
            link_args += ['-L', os.path.normpath(os.path.join(self.environment.get_build_dir(), reldir))]
        (rel_generated, _) = self.split_swift_generated_sources(target)
        abs_generated = [os.path.join(self.environment.get_build_dir(), x) for x in rel_generated]
        # We need absolute paths because swiftc needs to be invoked in a subdir
        # and this is the easiest way about it.
        objects = [] # Relative to swift invocation dir
        rel_objects = [] # Relative to build.ninja
        for i in abssrc + abs_generated:
            base = os.path.basename(i)
            oname = os.path.splitext(base)[0] + '.o'
            objects.append(oname)
            rel_objects.append(os.path.join(self.get_target_private_dir(target), oname))

        rulename = self.get_compiler_rule_name('swift', target.for_machine)

        # Swiftc does not seem to be able to emit objects and module files in one go.
        elem = NinjaBuildElement(self.all_outputs, rel_objects, rulename, abssrc)
        elem.add_dep(in_module_files + rel_generated)
        elem.add_dep(abs_headers)
        elem.add_item('ARGS', compile_args + header_imports + abs_generated + module_includes)
        elem.add_item('RUNDIR', rundir)
        self.add_build(elem)
        elem = NinjaBuildElement(self.all_outputs, out_module_name,
                                 self.get_compiler_rule_name('swift', target.for_machine),
                                 abssrc)
        elem.add_dep(in_module_files + rel_generated)
        elem.add_item('ARGS', compile_args + abs_generated + module_includes + swiftc.get_mod_gen_args())
        elem.add_item('RUNDIR', rundir)
        self.add_build(elem)
        if isinstance(target, build.StaticLibrary):
            elem = self.generate_link(target, self.get_target_filename(target),
                                      rel_objects, self.build.static_linker[target.for_machine])
            self.add_build(elem)
        elif isinstance(target, build.Executable):
            elem = NinjaBuildElement(self.all_outputs, self.get_target_filename(target), rulename, [])
            elem.add_dep(rel_objects)
            elem.add_dep(link_deps)
            elem.add_item('ARGS', link_args + swiftc.get_std_exe_link_args() + objects + abs_link_deps)
            elem.add_item('RUNDIR', rundir)
            self.add_build(elem)
        else:
            raise MesonException('Swift supports only executable and static library targets.')
        # Introspection information
        self.create_target_source_introspection(target, swiftc, compile_args + header_imports + module_includes, relsrc, rel_generated)

    def _rsp_options(self, tool: T.Union['Compiler', 'StaticLinker', 'DynamicLinker']) -> T.Dict[str, T.Union[bool, RSPFileSyntax]]:
        """Helper method to get rsp options.

        rsp_file_syntax() is only guaranteed to be implemented if
        can_linker_accept_rsp() returns True.
        """
        options = dict(rspable=tool.can_linker_accept_rsp())
        if options['rspable']:
            options['rspfile_quote_style'] = tool.rsp_file_syntax()
        return options

    def generate_static_link_rules(self):
        num_pools = self.environment.coredata.options[OptionKey('backend_max_links')].value
        if 'java' in self.environment.coredata.compilers.host:
            self.generate_java_link()
        for for_machine in MachineChoice:
            static_linker = self.build.static_linker[for_machine]
            if static_linker is None:
                continue
            rule = 'STATIC_LINKER{}'.format(self.get_rule_suffix(for_machine))
            cmdlist = []
            args = ['$in']
            # FIXME: Must normalize file names with pathlib.Path before writing
            #        them out to fix this properly on Windows. See:
            # https://github.com/mesonbuild/meson/issues/1517
            # https://github.com/mesonbuild/meson/issues/1526
            if isinstance(static_linker, ArLinker) and not mesonlib.is_windows():
                # `ar` has no options to overwrite archives. It always appends,
                # which is never what we want. Delete an existing library first if
                # it exists. https://github.com/mesonbuild/meson/issues/1355
                cmdlist = execute_wrapper + [c.format('$out') for c in rmfile_prefix]
            cmdlist += static_linker.get_exelist()
            cmdlist += ['$LINK_ARGS']
            cmdlist += NinjaCommandArg.list(static_linker.get_output_args('$out'), Quoting.none)
            description = 'Linking static target $out'
            if num_pools > 0:
                pool = 'pool = link_pool'
            else:
                pool = None

            options = self._rsp_options(static_linker)
            self.add_rule(NinjaRule(rule, cmdlist, args, description, **options, extra=pool))

    def generate_dynamic_link_rules(self):
        num_pools = self.environment.coredata.options[OptionKey('backend_max_links')].value
        for for_machine in MachineChoice:
            complist = self.environment.coredata.compilers[for_machine]
            for langname, compiler in complist.items():
                if langname in {'java', 'vala', 'rust', 'cs', 'cython'}:
                    continue
                rule = '{}_LINKER{}'.format(langname, self.get_rule_suffix(for_machine))
                command = compiler.get_linker_exelist()
                args = ['$ARGS'] + NinjaCommandArg.list(compiler.get_linker_output_args('$out'), Quoting.none) + ['$in', '$LINK_ARGS']
                description = 'Linking target $out'
                if num_pools > 0:
                    pool = 'pool = link_pool'
                else:
                    pool = None

                options = self._rsp_options(compiler)
                self.add_rule(NinjaRule(rule, command, args, description, **options, extra=pool))

        args = self.environment.get_build_command() + \
            ['--internal',
             'symbolextractor',
             self.environment.get_build_dir(),
             '$in',
             '$IMPLIB',
             '$out']
        symrule = 'SHSYM'
        symcmd = args + ['$CROSS']
        syndesc = 'Generating symbol file $out'
        synstat = 'restat = 1'
        self.add_rule(NinjaRule(symrule, symcmd, [], syndesc, extra=synstat))

    def generate_java_compile_rule(self, compiler):
        rule = self.compiler_to_rule_name(compiler)
        command = compiler.get_exelist() + ['$ARGS', '$in']
        description = 'Compiling Java object $in'
        self.add_rule(NinjaRule(rule, command, [], description))

    def generate_cs_compile_rule(self, compiler: 'CsCompiler') -> None:
        rule = self.compiler_to_rule_name(compiler)
        command = compiler.get_exelist()
        args = ['$ARGS', '$in']
        description = 'Compiling C Sharp target $out'
        self.add_rule(NinjaRule(rule, command, args, description,
                                rspable=mesonlib.is_windows(),
                                rspfile_quote_style=compiler.rsp_file_syntax()))

    def generate_vala_compile_rules(self, compiler):
        rule = self.compiler_to_rule_name(compiler)
        command = compiler.get_exelist() + ['$ARGS', '$in']
        description = 'Compiling Vala source $in'
        self.add_rule(NinjaRule(rule, command, [], description, extra='restat = 1'))

    def generate_cython_compile_rules(self, compiler: 'Compiler') -> None:
        rule = self.compiler_to_rule_name(compiler)
        command = compiler.get_exelist() + ['$ARGS', '$in']
        description = 'Compiling Cython source $in'
        self.add_rule(NinjaRule(rule, command, [], description, extra='restat = 1'))

    def generate_rust_compile_rules(self, compiler):
        rule = self.compiler_to_rule_name(compiler)
        command = compiler.get_exelist() + ['$ARGS', '$in']
        description = 'Compiling Rust source $in'
        depfile = '$targetdep'
        depstyle = 'gcc'
        self.add_rule(NinjaRule(rule, command, [], description, deps=depstyle,
                                depfile=depfile))

    def generate_swift_compile_rules(self, compiler):
        rule = self.compiler_to_rule_name(compiler)
        full_exe = self.environment.get_build_command() + [
            '--internal',
            'dirchanger',
            '$RUNDIR',
        ]
        invoc = full_exe + compiler.get_exelist()
        command = invoc + ['$ARGS', '$in']
        description = 'Compiling Swift source $in'
        self.add_rule(NinjaRule(rule, command, [], description))

    def use_dyndeps_for_fortran(self) -> bool:
        '''Use the new Ninja feature for scanning dependencies during build,
        rather than up front. Remove this and all old scanning code once Ninja
        minimum version is bumped to 1.10.'''
        return mesonlib.version_compare(self.ninja_version, '>=1.10.0')

    def generate_fortran_dep_hack(self, crstr: str) -> None:
        if self.use_dyndeps_for_fortran():
            return
        rule = f'FORTRAN_DEP_HACK{crstr}'
        if mesonlib.is_windows():
            cmd = ['cmd', '/C']
        else:
            cmd = ['true']
        self.add_rule_comment(NinjaComment('''Workaround for these issues:
https://groups.google.com/forum/#!topic/ninja-build/j-2RfBIOd_8
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=47485'''))
        self.add_rule(NinjaRule(rule, cmd, [], 'Dep hack', extra='restat = 1'))

    def generate_llvm_ir_compile_rule(self, compiler):
        if self.created_llvm_ir_rule[compiler.for_machine]:
            return
        rule = self.get_compiler_rule_name('llvm_ir', compiler.for_machine)
        command = compiler.get_exelist()
        args = ['$ARGS'] + NinjaCommandArg.list(compiler.get_output_args('$out'), Quoting.none) + compiler.get_compile_only_args() + ['$in']
        description = 'Compiling LLVM IR object $in'

        options = self._rsp_options(compiler)

        self.add_rule(NinjaRule(rule, command, args, description, **options))
        self.created_llvm_ir_rule[compiler.for_machine] = True

    def generate_compile_rule_for(self, langname, compiler):
        if langname == 'java':
            self.generate_java_compile_rule(compiler)
            return
        if langname == 'cs':
            if self.environment.machines.matches_build_machine(compiler.for_machine):
                self.generate_cs_compile_rule(compiler)
            return
        if langname == 'vala':
            self.generate_vala_compile_rules(compiler)
            return
        if langname == 'rust':
            self.generate_rust_compile_rules(compiler)
            return
        if langname == 'swift':
            if self.environment.machines.matches_build_machine(compiler.for_machine):
                self.generate_swift_compile_rules(compiler)
            return
        if langname == 'cython':
            self.generate_cython_compile_rules(compiler)
            return
        crstr = self.get_rule_suffix(compiler.for_machine)
        if langname == 'fortran':
            self.generate_fortran_dep_hack(crstr)
        rule = self.get_compiler_rule_name(langname, compiler.for_machine)
        depargs = NinjaCommandArg.list(compiler.get_dependency_gen_args('$out', '$DEPFILE'), Quoting.none)
        command = compiler.get_exelist()
        args = ['$ARGS'] + depargs + NinjaCommandArg.list(compiler.get_output_args('$out'), Quoting.none) + compiler.get_compile_only_args() + ['$in']
        description = f'Compiling {compiler.get_display_language()} object $out'
        if isinstance(compiler, VisualStudioLikeCompiler):
            deps = 'msvc'
            depfile = None
        else:
            deps = 'gcc'
            depfile = '$DEPFILE'
        options = self._rsp_options(compiler)
        self.add_rule(NinjaRule(rule, command, args, description, **options,
                                deps=deps, depfile=depfile))

    def generate_pch_rule_for(self, langname, compiler):
        if langname != 'c' and langname != 'cpp':
            return
        rule = self.compiler_to_pch_rule_name(compiler)
        depargs = compiler.get_dependency_gen_args('$out', '$DEPFILE')

        if isinstance(compiler, VisualStudioLikeCompiler):
            output = []
        else:
            output = NinjaCommandArg.list(compiler.get_output_args('$out'), Quoting.none)
        command = compiler.get_exelist() + ['$ARGS'] + depargs + output + compiler.get_compile_only_args() + ['$in']
        description = 'Precompiling header $in'
        if isinstance(compiler, VisualStudioLikeCompiler):
            deps = 'msvc'
            depfile = None
        else:
            deps = 'gcc'
            depfile = '$DEPFILE'
        self.add_rule(NinjaRule(rule, command, [], description, deps=deps,
                                depfile=depfile))

    def generate_scanner_rules(self):
        rulename = 'depscan'
        if rulename in self.ruledict:
            # Scanning command is the same for native and cross compilation.
            return
        command = self.environment.get_build_command() + \
            ['--internal', 'depscan']
        args = ['$picklefile', '$out', '$in']
        description = 'Module scanner.'
        rule = NinjaRule(rulename, command, args, description)
        self.add_rule(rule)

    def generate_compile_rules(self):
        for for_machine in MachineChoice:
            clist = self.environment.coredata.compilers[for_machine]
            for langname, compiler in clist.items():
                if compiler.get_id() == 'clang':
                    self.generate_llvm_ir_compile_rule(compiler)
                self.generate_compile_rule_for(langname, compiler)
                self.generate_pch_rule_for(langname, compiler)

    def generate_generator_list_rules(self, target):
        # CustomTargets have already written their rules and
        # CustomTargetIndexes don't actually get generated, so write rules for
        # GeneratedLists here
        for genlist in target.get_generated_sources():
            if isinstance(genlist, (build.CustomTarget, build.CustomTargetIndex)):
                continue
            self.generate_genlist_for_target(genlist, target)

    def replace_paths(self, target, args, override_subdir=None):
        if override_subdir:
            source_target_dir = os.path.join(self.build_to_src, override_subdir)
        else:
            source_target_dir = self.get_target_source_dir(target)
        relout = self.get_target_private_dir(target)
        args = [x.replace("@SOURCE_DIR@", self.build_to_src).replace("@BUILD_DIR@", relout)
                for x in args]
        args = [x.replace("@CURRENT_SOURCE_DIR@", source_target_dir) for x in args]
        args = [x.replace("@SOURCE_ROOT@", self.build_to_src).replace("@BUILD_ROOT@", '.')
                for x in args]
        args = [x.replace('\\', '/') for x in args]
        return args

    def generate_genlist_for_target(self, genlist, target):
        generator = genlist.get_generator()
        subdir = genlist.subdir
        exe = generator.get_exe()
        exe_arr = self.build_target_to_cmd_array(exe)
        infilelist = genlist.get_inputs()
        outfilelist = genlist.get_outputs()
        extra_dependencies = self.get_custom_target_depend_files(genlist)
        for i, curfile in enumerate(infilelist):
            if len(generator.outputs) == 1:
                sole_output = os.path.join(self.get_target_private_dir(target), outfilelist[i])
            else:
                sole_output = f'{curfile}'
            infilename = curfile.rel_to_builddir(self.build_to_src)
            base_args = generator.get_arglist(infilename)
            outfiles = genlist.get_outputs_for(curfile)
            outfiles = [os.path.join(self.get_target_private_dir(target), of) for of in outfiles]
            if generator.depfile is None:
                rulename = 'CUSTOM_COMMAND'
                args = base_args
            else:
                rulename = 'CUSTOM_COMMAND_DEP'
                depfilename = generator.get_dep_outname(infilename)
                depfile = os.path.join(self.get_target_private_dir(target), depfilename)
                args = [x.replace('@DEPFILE@', depfile) for x in base_args]
            args = [x.replace("@INPUT@", infilename).replace('@OUTPUT@', sole_output)
                    for x in args]
            args = self.replace_outputs(args, self.get_target_private_dir(target), outfilelist)
            # We have consumed output files, so drop them from the list of remaining outputs.
            if len(generator.outputs) > 1:
                outfilelist = outfilelist[len(generator.outputs):]
            args = self.replace_paths(target, args, override_subdir=subdir)
            cmdlist = exe_arr + self.replace_extra_args(args, genlist)
            cmdlist, reason = self.as_meson_exe_cmdline(cmdlist[0], cmdlist[1:],
                                                        capture=outfiles[0] if generator.capture else None)
            abs_pdir = os.path.join(self.environment.get_build_dir(), self.get_target_dir(target))
            os.makedirs(abs_pdir, exist_ok=True)

            elem = NinjaBuildElement(self.all_outputs, outfiles, rulename, infilename)
            elem.add_dep([self.get_target_filename(x) for x in generator.depends])
            if generator.depfile is not None:
                elem.add_item('DEPFILE', depfile)
            if len(extra_dependencies) > 0:
                elem.add_dep(extra_dependencies)

            if len(generator.outputs) == 1:
                what = f'{sole_output!r}'
            else:
                # since there are multiple outputs, we log the source that caused the rebuild
                what = f'from {sole_output!r}.'
            if reason:
                reason = f' (wrapped by meson {reason})'
            elem.add_item('DESC', f'Generating {what}{reason}.')

            if isinstance(exe, build.BuildTarget):
                elem.add_dep(self.get_target_filename(exe))
            elem.add_item('COMMAND', cmdlist)
            self.add_build(elem)

    def scan_fortran_module_outputs(self, target):
        """
        Find all module and submodule made available in a Fortran code file.
        """
        if self.use_dyndeps_for_fortran():
            return
        compiler = None
        # TODO other compilers
        for lang, c in self.environment.coredata.compilers.host.items():
            if lang == 'fortran':
                compiler = c
                break
        if compiler is None:
            self.fortran_deps[target.get_basename()] = {}
            return

        modre = re.compile(FORTRAN_MODULE_PAT, re.IGNORECASE)
        submodre = re.compile(FORTRAN_SUBMOD_PAT, re.IGNORECASE)
        module_files = {}
        submodule_files = {}
        for s in target.get_sources():
            # FIXME, does not work for Fortran sources generated by
            # custom_target() and generator() as those are run after
            # the configuration (configure_file() is OK)
            if not compiler.can_compile(s):
                continue
            filename = s.absolute_path(self.environment.get_source_dir(),
                                       self.environment.get_build_dir())
            # Fortran keywords must be ASCII.
            with open(filename, encoding='ascii', errors='ignore') as f:
                for line in f:
                    modmatch = modre.match(line)
                    if modmatch is not None:
                        modname = modmatch.group(1).lower()
                        if modname in module_files:
                            raise InvalidArguments(
                                f'Namespace collision: module {modname} defined in '
                                'two files {module_files[modname]} and {s}.')
                        module_files[modname] = s
                    else:
                        submodmatch = submodre.match(line)
                        if submodmatch is not None:
                            # '_' is arbitrarily used to distinguish submod from mod.
                            parents = submodmatch.group(1).lower().split(':')
                            submodname = parents[0] + '_' + submodmatch.group(2).lower()

                            if submodname in submodule_files:
                                raise InvalidArguments(
                                    'Namespace collision: submodule {submodname} defined in '
                                    'two files {submodule_files[submodname]} and {s}.')
                            submodule_files[submodname] = s

        self.fortran_deps[target.get_basename()] = {**module_files, **submodule_files}

    def get_fortran_deps(self, compiler: FortranCompiler, src: Path, target) -> T.List[str]:
        """
        Find all module and submodule needed by a Fortran target
        """
        if self.use_dyndeps_for_fortran():
            return []

        dirname = Path(self.get_target_private_dir(target))
        tdeps = self.fortran_deps[target.get_basename()]
        srcdir = Path(self.source_dir)

        mod_files = _scan_fortran_file_deps(src, srcdir, dirname, tdeps, compiler)
        return mod_files

    def get_no_stdlib_link_args(self, target, linker):
        if hasattr(linker, 'language') and linker.language in self.build.stdlibs[target.for_machine]:
            return linker.get_no_stdlib_link_args()
        return []

    def get_compile_debugfile_args(self, compiler, target, objfile):
        # The way MSVC uses PDB files is documented exactly nowhere so
        # the following is what we have been able to decipher via
        # reverse engineering.
        #
        # Each object file gets the path of its PDB file written
        # inside it.  This can be either the final PDB (for, say,
        # foo.exe) or an object pdb (for foo.obj). If the former, then
        # each compilation step locks the pdb file for writing, which
        # is a bottleneck and object files from one target can not be
        # used in a different target. The latter seems to be the
        # sensible one (and what Unix does) but there is a catch.  If
        # you try to use precompiled headers MSVC will error out
        # because both source and pch pdbs go in the same file and
        # they must be the same.
        #
        # This means:
        #
        # - pch files must be compiled anew for every object file (negating
        #   the entire point of having them in the first place)
        # - when using pch, output must go to the target pdb
        #
        # Since both of these are broken in some way, use the one that
        # works for each target. This unfortunately means that you
        # can't combine pch and object extraction in a single target.
        #
        # PDB files also lead to filename collisions. A target foo.exe
        # has a corresponding foo.pdb. A shared library foo.dll _also_
        # has pdb file called foo.pdb. So will a static library
        # foo.lib, which clobbers both foo.pdb _and_ the dll file's
        # export library called foo.lib (by default, currently we name
        # them libfoo.a to avoidt this issue). You can give the files
        # unique names such as foo_exe.pdb but VC also generates a
        # bunch of other files which take their names from the target
        # basename (i.e. "foo") and stomp on each other.
        #
        # CMake solves this problem by doing two things. First of all
        # static libraries do not generate pdb files at
        # all. Presumably you don't need them and VC is smart enough
        # to look up the original data when linking (speculation, not
        # tested). The second solution is that you can only have
        # target named "foo" as an exe, shared lib _or_ static
        # lib. This makes filename collisions not happen. The downside
        # is that you can't have an executable foo that uses a shared
        # library libfoo.so, which is a common idiom on Unix.
        #
        # If you feel that the above is completely wrong and all of
        # this is actually doable, please send patches.

        if target.has_pch():
            tfilename = self.get_target_filename_abs(target)
            return compiler.get_compile_debugfile_args(tfilename, pch=True)
        else:
            return compiler.get_compile_debugfile_args(objfile, pch=False)

    def get_link_debugfile_name(self, linker, target, outname):
        return linker.get_link_debugfile_name(outname)

    def get_link_debugfile_args(self, linker, target, outname):
        return linker.get_link_debugfile_args(outname)

    def generate_llvm_ir_compile(self, target, src):
        base_proxy = self.get_options_for_target(target)
        compiler = get_compiler_for_source(target.compilers.values(), src)
        commands = compiler.compiler_args()
        # Compiler args for compiling this target
        commands += compilers.get_base_compile_args(base_proxy, compiler)
        if isinstance(src, File):
            if src.is_built:
                src_filename = os.path.join(src.subdir, src.fname)
            else:
                src_filename = src.fname
        elif os.path.isabs(src):
            src_filename = os.path.basename(src)
        else:
            src_filename = src
        obj_basename = self.canonicalize_filename(src_filename)
        rel_obj = os.path.join(self.get_target_private_dir(target), obj_basename)
        rel_obj += '.' + self.environment.machines[target.for_machine].get_object_suffix()
        commands += self.get_compile_debugfile_args(compiler, target, rel_obj)
        if isinstance(src, File) and src.is_built:
            rel_src = src.fname
        elif isinstance(src, File):
            rel_src = src.rel_to_builddir(self.build_to_src)
        else:
            raise InvalidArguments(f'Invalid source type: {src!r}')
        # Write the Ninja build command
        compiler_name = self.get_compiler_rule_name('llvm_ir', compiler.for_machine)
        element = NinjaBuildElement(self.all_outputs, rel_obj, compiler_name, rel_src)
        element.add_item('ARGS', commands)
        self.add_build(element)
        return (rel_obj, rel_src)

    @lru_cache(maxsize=None)
    def generate_inc_dir(self, compiler: 'Compiler', d: str, basedir: str, is_system: bool) -> \
            T.Tuple['ImmutableListProtocol[str]', 'ImmutableListProtocol[str]']:
        # Avoid superfluous '/.' at the end of paths when d is '.'
        if d not in ('', '.'):
            expdir = os.path.normpath(os.path.join(basedir, d))
        else:
            expdir = basedir
        srctreedir = os.path.normpath(os.path.join(self.build_to_src, expdir))
        sargs = compiler.get_include_args(srctreedir, is_system)
        # There may be include dirs where a build directory has not been
        # created for some source dir. For example if someone does this:
        #
        # inc = include_directories('foo/bar/baz')
        #
        # But never subdir()s into the actual dir.
        if os.path.isdir(os.path.join(self.environment.get_build_dir(), expdir)):
            bargs = compiler.get_include_args(expdir, is_system)
        else:
            bargs = []
        return (sargs, bargs)

    def _generate_single_compile(self, target: build.BuildTarget, compiler: 'Compiler',
                                 is_generated: bool = False) -> 'CompilerArgs':
        commands = self._generate_single_compile_base_args(target, compiler)
        commands += self._generate_single_compile_target_args(target, compiler, is_generated)
        return commands

    def _generate_single_compile_base_args(self, target: build.BuildTarget, compiler: 'Compiler') -> 'CompilerArgs':
        base_proxy = self.get_options_for_target(target)
        # Create an empty commands list, and start adding arguments from
        # various sources in the order in which they must override each other
        commands = compiler.compiler_args()
        # Start with symbol visibility.
        commands += compiler.gnu_symbol_visibility_args(target.gnu_symbol_visibility)
        # Add compiler args for compiling this target derived from 'base' build
        # options passed on the command-line, in default_options, etc.
        # These have the lowest priority.
        commands += compilers.get_base_compile_args(base_proxy,
                                                    compiler)
        return commands

    @lru_cache(maxsize=None)
    def _generate_single_compile_target_args(self, target: build.BuildTarget, compiler: 'Compiler',
                                             is_generated: bool = False) -> 'ImmutableListProtocol[str]':
        # The code generated by valac is usually crap and has tons of unused
        # variables and such, so disable warnings for Vala C sources.
        no_warn_args = (is_generated == 'vala')
        # Add compiler args and include paths from several sources; defaults,
        # build options, external dependencies, etc.
        commands = self.generate_basic_compiler_args(target, compiler, no_warn_args)
        # Add custom target dirs as includes automatically, but before
        # target-specific include directories.
        if target.implicit_include_directories:
            commands += self.get_custom_target_dir_include_args(target, compiler)
        # Add include dirs from the `include_directories:` kwarg on the target
        # and from `include_directories:` of internal deps of the target.
        #
        # Target include dirs should override internal deps include dirs.
        # This is handled in BuildTarget.process_kwargs()
        #
        # Include dirs from internal deps should override include dirs from
        # external deps and must maintain the order in which they are specified.
        # Hence, we must reverse the list so that the order is preserved.
        for i in reversed(target.get_include_dirs()):
            basedir = i.get_curdir()
            # We should iterate include dirs in reversed orders because
            # -Ipath will add to begin of array. And without reverse
            # flags will be added in reversed order.
            for d in reversed(i.get_incdirs()):
                # Add source subdir first so that the build subdir overrides it
                (compile_obj, includeargs) = self.generate_inc_dir(compiler, d, basedir, i.is_system)
                commands += compile_obj
                commands += includeargs
            for d in i.get_extra_build_dirs():
                commands += compiler.get_include_args(d, i.is_system)
        # Add per-target compile args, f.ex, `c_args : ['-DFOO']`. We set these
        # near the end since these are supposed to override everything else.
        commands += self.escape_extra_args(target.get_extra_args(compiler.get_language()))

        # D specific additional flags
        if compiler.language == 'd':
            commands += compiler.get_feature_args(target.d_features, self.build_to_src)

        # Add source dir and build dir. Project-specific and target-specific
        # include paths must override per-target compile args, include paths
        # from external dependencies, internal dependencies, and from
        # per-target `include_directories:`
        #
        # We prefer headers in the build dir over the source dir since, for
        # instance, the user might have an srcdir == builddir Autotools build
        # in their source tree. Many projects that are moving to Meson have
        # both Meson and Autotools in parallel as part of the transition.
        if target.implicit_include_directories:
            commands += self.get_source_dir_include_args(target, compiler)
        if target.implicit_include_directories:
            commands += self.get_build_dir_include_args(target, compiler)
        # Finally add the private dir for the target to the include path. This
        # must override everything else and must be the final path added.
        commands += compiler.get_include_args(self.get_target_private_dir(target), False)
        return commands

    def generate_single_compile(self, target, src, is_generated=False, header_deps=None, order_deps=None):
        """
        Compiles C/C++, ObjC/ObjC++, Fortran, and D sources
        """
        header_deps = header_deps if header_deps is not None else []
        order_deps = order_deps if order_deps is not None else []

        if isinstance(src, str) and src.endswith('.h'):
            raise AssertionError(f'BUG: sources should not contain headers {src!r}')

        compiler = get_compiler_for_source(target.compilers.values(), src)
        commands = self._generate_single_compile_base_args(target, compiler)

        # Include PCH header as first thing as it must be the first one or it will be
        # ignored by gcc https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100462
        if self.environment.coredata.options.get(OptionKey('b_pch')) and is_generated != 'pch':
            commands += self.get_pch_include_args(compiler, target)

        commands += self._generate_single_compile_target_args(target, compiler, is_generated)
        commands = commands.compiler.compiler_args(commands)

        # Create introspection information
        if is_generated is False:
            self.create_target_source_introspection(target, compiler, commands, [src], [])
        else:
            self.create_target_source_introspection(target, compiler, commands, [], [src])

        build_dir = self.environment.get_build_dir()
        if isinstance(src, File):
            rel_src = src.rel_to_builddir(self.build_to_src)
            if os.path.isabs(rel_src):
                # Source files may not be from the source directory if they originate in source-only libraries,
                # so we can't assert that the absolute path is anywhere in particular.
                if src.is_built:
                    assert rel_src.startswith(build_dir)
                    rel_src = rel_src[len(build_dir) + 1:]
        elif is_generated:
            raise AssertionError(f'BUG: broken generated source file handling for {src!r}')
        else:
            raise InvalidArguments(f'Invalid source type: {src!r}')
        obj_basename = self.object_filename_from_source(target, src)
        rel_obj = os.path.join(self.get_target_private_dir(target), obj_basename)
        dep_file = compiler.depfile_for_object(rel_obj)

        # Add MSVC debug file generation compile flags: /Fd /FS
        commands += self.get_compile_debugfile_args(compiler, target, rel_obj)

        # PCH handling
        if self.environment.coredata.options.get(OptionKey('b_pch')):
            pchlist = target.get_pch(compiler.language)
        else:
            pchlist = []
        if not pchlist:
            pch_dep = []
        elif compiler.id == 'intel':
            pch_dep = []
        else:
            arr = []
            i = os.path.join(self.get_target_private_dir(target), compiler.get_pch_name(pchlist[0]))
            arr.append(i)
            pch_dep = arr

        compiler_name = self.compiler_to_rule_name(compiler)
        extra_deps = []
        if compiler.get_language() == 'fortran':
            # Can't read source file to scan for deps if it's generated later
            # at build-time. Skip scanning for deps, and just set the module
            # outdir argument instead.
            # https://github.com/mesonbuild/meson/issues/1348
            if not is_generated:
                abs_src = Path(build_dir) / rel_src
                extra_deps += self.get_fortran_deps(compiler, abs_src, target)
            if not self.use_dyndeps_for_fortran():
                # Dependency hack. Remove once multiple outputs in Ninja is fixed:
                # https://groups.google.com/forum/#!topic/ninja-build/j-2RfBIOd_8
                for modname, srcfile in self.fortran_deps[target.get_basename()].items():
                    modfile = os.path.join(self.get_target_private_dir(target),
                                           compiler.module_name_to_filename(modname))

                    if srcfile == src:
                        crstr = self.get_rule_suffix(target.for_machine)
                        depelem = NinjaBuildElement(self.all_outputs,
                                                    modfile,
                                                    'FORTRAN_DEP_HACK' + crstr,
                                                    rel_obj)
                        self.add_build(depelem)
            commands += compiler.get_module_outdir_args(self.get_target_private_dir(target))

        element = NinjaBuildElement(self.all_outputs, rel_obj, compiler_name, rel_src)
        self.add_header_deps(target, element, header_deps)
        for d in extra_deps:
            element.add_dep(d)
        for d in order_deps:
            if isinstance(d, File):
                d = d.rel_to_builddir(self.build_to_src)
            elif not self.has_dir_part(d):
                d = os.path.join(self.get_target_private_dir(target), d)
            element.add_orderdep(d)
        element.add_dep(pch_dep)
        for i in self.get_fortran_orderdeps(target, compiler):
            element.add_orderdep(i)
        element.add_item('DEPFILE', dep_file)
        element.add_item('ARGS', commands)

        self.add_dependency_scanner_entries_to_element(target, compiler, element, src)
        self.add_build(element)
        assert isinstance(rel_obj, str)
        assert isinstance(rel_src, str)
        return (rel_obj, rel_src.replace('\\', '/'))

    def add_dependency_scanner_entries_to_element(self, target, compiler, element, src):
        if not self.should_use_dyndeps_for_target(target):
            return
        extension = os.path.splitext(src.fname)[1][1:]
        if not (extension.lower() in compilers.lang_suffixes['fortran'] or extension in compilers.lang_suffixes['cpp']):
            return
        dep_scan_file = self.get_dep_scan_file_for(target)
        element.add_item('dyndep', dep_scan_file)
        element.add_orderdep(dep_scan_file)

    def get_dep_scan_file_for(self, target):
        return os.path.join(self.get_target_private_dir(target), 'depscan.dd')

    def add_header_deps(self, target, ninja_element, header_deps):
        for d in header_deps:
            if isinstance(d, File):
                d = d.rel_to_builddir(self.build_to_src)
            elif not self.has_dir_part(d):
                d = os.path.join(self.get_target_private_dir(target), d)
            ninja_element.add_dep(d)

    def has_dir_part(self, fname):
        # FIXME FIXME: The usage of this is a terrible and unreliable hack
        if isinstance(fname, File):
            return fname.subdir != ''
        return has_path_sep(fname)

    # Fortran is a bit weird (again). When you link against a library, just compiling a source file
    # requires the mod files that are output when single files are built. To do this right we would need to
    # scan all inputs and write out explicit deps for each file. That is stoo slow and too much effort so
    # instead just have an ordered dependency on the library. This ensures all required mod files are created.
    # The real deps are then detected via dep file generation from the compiler. This breaks on compilers that
    # produce incorrect dep files but such is life.
    def get_fortran_orderdeps(self, target, compiler):
        if compiler.language != 'fortran':
            return []
        return [
            os.path.join(self.get_target_dir(lt), lt.get_filename())
            for lt in itertools.chain(target.link_targets, target.link_whole_targets)
        ]

    def generate_msvc_pch_command(self, target, compiler, pch):
        header = pch[0]
        pchname = compiler.get_pch_name(header)
        dst = os.path.join(self.get_target_private_dir(target), pchname)

        commands = []
        commands += self.generate_basic_compiler_args(target, compiler)

        if len(pch) == 1:
            # Auto generate PCH.
            source = self.create_msvc_pch_implementation(target, compiler.get_language(), pch[0])
            pch_header_dir = os.path.dirname(os.path.join(self.build_to_src, target.get_source_subdir(), header))
            commands += compiler.get_include_args(pch_header_dir, False)
        else:
            source = os.path.join(self.build_to_src, target.get_source_subdir(), pch[1])

        just_name = os.path.basename(header)
        (objname, pch_args) = compiler.gen_pch_args(just_name, source, dst)
        commands += pch_args
        commands += self._generate_single_compile(target, compiler)
        commands += self.get_compile_debugfile_args(compiler, target, objname)
        dep = dst + '.' + compiler.get_depfile_suffix()
        return commands, dep, dst, [objname], source

    def generate_gcc_pch_command(self, target, compiler, pch):
        commands = self._generate_single_compile(target, compiler)
        if pch.split('.')[-1] == 'h' and compiler.language == 'cpp':
            # Explicitly compile pch headers as C++. If Clang is invoked in C++ mode, it actually warns if
            # this option is not set, and for gcc it also makes sense to use it.
            commands += ['-x', 'c++-header']
        dst = os.path.join(self.get_target_private_dir(target),
                           os.path.basename(pch) + '.' + compiler.get_pch_suffix())
        dep = dst + '.' + compiler.get_depfile_suffix()
        return commands, dep, dst, []  # Gcc does not create an object file during pch generation.

    def generate_pch(self, target, header_deps=None):
        header_deps = header_deps if header_deps is not None else []
        pch_objects = []
        for lang in ['c', 'cpp']:
            pch = target.get_pch(lang)
            if not pch:
                continue
            if not has_path_sep(pch[0]) or not has_path_sep(pch[-1]):
                msg = f'Precompiled header of {target.get_basename()!r} must not be in the same ' \
                      'directory as source, please put it in a subdirectory.'
                raise InvalidArguments(msg)
            compiler = target.compilers[lang]
            if isinstance(compiler, VisualStudioLikeCompiler):
                (commands, dep, dst, objs, src) = self.generate_msvc_pch_command(target, compiler, pch)
                extradep = os.path.join(self.build_to_src, target.get_source_subdir(), pch[0])
            elif compiler.id == 'intel':
                # Intel generates on target generation
                continue
            else:
                src = os.path.join(self.build_to_src, target.get_source_subdir(), pch[0])
                (commands, dep, dst, objs) = self.generate_gcc_pch_command(target, compiler, pch[0])
                extradep = None
            pch_objects += objs
            rulename = self.compiler_to_pch_rule_name(compiler)
            elem = NinjaBuildElement(self.all_outputs, dst, rulename, src)
            if extradep is not None:
                elem.add_dep(extradep)
            self.add_header_deps(target, elem, header_deps)
            elem.add_item('ARGS', commands)
            elem.add_item('DEPFILE', dep)
            self.add_build(elem)
        return pch_objects

    def get_target_shsym_filename(self, target):
        # Always name the .symbols file after the primary build output because it always exists
        targetdir = self.get_target_private_dir(target)
        return os.path.join(targetdir, target.get_filename() + '.symbols')

    def generate_shsym(self, target):
        target_file = self.get_target_filename(target)
        symname = self.get_target_shsym_filename(target)
        elem = NinjaBuildElement(self.all_outputs, symname, 'SHSYM', target_file)
        # The library we will actually link to, which is an import library on Windows (not the DLL)
        elem.add_item('IMPLIB', self.get_target_filename_for_linking(target))
        if self.environment.is_cross_build():
            elem.add_item('CROSS', '--cross-host=' + self.environment.machines[target.for_machine].system)
        self.add_build(elem)

    def get_import_filename(self, target):
        return os.path.join(self.get_target_dir(target), target.import_filename)

    def get_target_type_link_args(self, target, linker):
        commands = []
        if isinstance(target, build.Executable):
            # Currently only used with the Swift compiler to add '-emit-executable'
            commands += linker.get_std_exe_link_args()
            # If export_dynamic, add the appropriate linker arguments
            if target.export_dynamic:
                commands += linker.gen_export_dynamic_link_args(self.environment)
            # If implib, and that's significant on this platform (i.e. Windows using either GCC or Visual Studio)
            if target.import_filename:
                commands += linker.gen_import_library_args(self.get_import_filename(target))
            if target.pie:
                commands += linker.get_pie_link_args()
        elif isinstance(target, build.SharedLibrary):
            if isinstance(target, build.SharedModule):
                options = self.environment.coredata.options
                commands += linker.get_std_shared_module_link_args(options)
            else:
                commands += linker.get_std_shared_lib_link_args()
            # All shared libraries are PIC
            commands += linker.get_pic_args()
            if not isinstance(target, build.SharedModule) or target.force_soname:
                # Add -Wl,-soname arguments on Linux, -install_name on OS X
                commands += linker.get_soname_args(
                    self.environment, target.prefix, target.name, target.suffix,
                    target.soversion, target.darwin_versions)
            # This is only visited when building for Windows using either GCC or Visual Studio
            if target.vs_module_defs and hasattr(linker, 'gen_vs_module_defs_args'):
                commands += linker.gen_vs_module_defs_args(target.vs_module_defs.rel_to_builddir(self.build_to_src))
            # This is only visited when building for Windows using either GCC or Visual Studio
            if target.import_filename:
                commands += linker.gen_import_library_args(self.get_import_filename(target))
        elif isinstance(target, build.StaticLibrary):
            commands += linker.get_std_link_args(not target.should_install())
        else:
            raise RuntimeError('Unknown build target type.')
        return commands

    def get_target_type_link_args_post_dependencies(self, target, linker):
        commands = []
        if isinstance(target, build.Executable):
            # If gui_app is significant on this platform, add the appropriate linker arguments.
            # Unfortunately this can't be done in get_target_type_link_args, because some misguided
            # libraries (such as SDL2) add -mwindows to their link flags.
            m = self.environment.machines[target.for_machine]

            if m.is_windows() or m.is_cygwin():
                if target.gui_app is not None:
                    commands += linker.get_gui_app_args(target.gui_app)
                else:
                    commands += linker.get_win_subsystem_args(target.win_subsystem)
        return commands

    def get_link_whole_args(self, linker, target):
        use_custom = False
        if isinstance(linker, mixins.visualstudio.MSVCCompiler):
            # Expand our object lists manually if we are on pre-Visual Studio 2015 Update 2
            # (incidentally, the "linker" here actually refers to cl.exe)
            if mesonlib.version_compare(linker.version, '<19.00.23918'):
                use_custom = True

        if use_custom:
            objects_from_static_libs: T.List[ExtractedObjects] = []
            for dep in target.link_whole_targets:
                l = dep.extract_all_objects(False)
                objects_from_static_libs += self.determine_ext_objs(l, '')
                objects_from_static_libs.extend(self.flatten_object_list(dep))

            return objects_from_static_libs
        else:
            target_args = self.build_target_link_arguments(linker, target.link_whole_targets)
            return linker.get_link_whole_for(target_args) if target_args else []

    @lru_cache(maxsize=None)
    def guess_library_absolute_path(self, linker, libname, search_dirs, patterns) -> Path:
        for d in search_dirs:
            for p in patterns:
                trial = CCompiler._get_trials_from_pattern(p, d, libname)
                if not trial:
                    continue
                trial = CCompiler._get_file_from_list(self.environment, trial)
                if not trial:
                    continue
                # Return the first result
                return trial

    def guess_external_link_dependencies(self, linker, target, commands, internal):
        # Ideally the linker would generate dependency information that could be used.
        # But that has 2 problems:
        # * currently ld can not create dependency information in a way that ninja can use:
        #   https://sourceware.org/bugzilla/show_bug.cgi?id=22843
        # * Meson optimizes libraries from the same build using the symbol extractor.
        #   Just letting ninja use ld generated dependencies would undo this optimization.
        search_dirs = OrderedSet()
        libs = OrderedSet()
        absolute_libs = []

        build_dir = self.environment.get_build_dir()
        # the following loop sometimes consumes two items from command in one pass
        it = iter(linker.native_args_to_unix(commands))
        for item in it:
            if item in internal and not item.startswith('-'):
                continue

            if item.startswith('-L'):
                if len(item) > 2:
                    path = item[2:]
                else:
                    try:
                        path = next(it)
                    except StopIteration:
                        mlog.warning("Generated linker command has -L argument without following path")
                        break
                if not os.path.isabs(path):
                    path = os.path.join(build_dir, path)
                search_dirs.add(path)
            elif item.startswith('-l'):
                if len(item) > 2:
                    lib = item[2:]
                else:
                    try:
                        lib = next(it)
                    except StopIteration:
                        mlog.warning("Generated linker command has '-l' argument without following library name")
                        break
                libs.add(lib)
            elif os.path.isabs(item) and self.environment.is_library(item) and os.path.isfile(item):
                absolute_libs.append(item)

        guessed_dependencies = []
        # TODO The get_library_naming requirement currently excludes link targets that use d or fortran as their main linker
        try:
            static_patterns = linker.get_library_naming(self.environment, LibType.STATIC, strict=True)
            shared_patterns = linker.get_library_naming(self.environment, LibType.SHARED, strict=True)
            search_dirs = tuple(search_dirs) + tuple(linker.get_library_dirs(self.environment))
            for libname in libs:
                # be conservative and record most likely shared and static resolution, because we don't know exactly
                # which one the linker will prefer
                staticlibs = self.guess_library_absolute_path(linker, libname,
                                                              search_dirs, static_patterns)
                sharedlibs = self.guess_library_absolute_path(linker, libname,
                                                              search_dirs, shared_patterns)
                if staticlibs:
                    guessed_dependencies.append(staticlibs.resolve().as_posix())
                if sharedlibs:
                    guessed_dependencies.append(sharedlibs.resolve().as_posix())
        except (mesonlib.MesonException, AttributeError) as e:
            if 'get_library_naming' not in str(e):
                raise

        return guessed_dependencies + absolute_libs

    def generate_prelink(self, target, obj_list):
        assert isinstance(target, build.StaticLibrary)
        prelink_name = os.path.join(self.get_target_private_dir(target), target.name + '-prelink.o')
        elem = NinjaBuildElement(self.all_outputs, [prelink_name], 'CUSTOM_COMMAND', obj_list)

        prelinker = target.get_prelinker()
        cmd = prelinker.exelist[:]
        cmd += prelinker.get_prelink_args(prelink_name, obj_list)

        cmd = self.replace_paths(target, cmd)
        elem.add_item('COMMAND', cmd)
        elem.add_item('description', f'Prelinking {prelink_name}.')
        self.add_build(elem)
        return [prelink_name]

    def generate_link(self, target: build.BuildTarget, outname, obj_list, linker: T.Union['Compiler', 'StaticLinker'], extra_args=None, stdlib_args=None):
        extra_args = extra_args if extra_args is not None else []
        stdlib_args = stdlib_args if stdlib_args is not None else []
        implicit_outs = []
        if isinstance(target, build.StaticLibrary):
            linker_base = 'STATIC'
        else:
            linker_base = linker.get_language() # Fixme.
        if isinstance(target, build.SharedLibrary):
            self.generate_shsym(target)
        crstr = self.get_rule_suffix(target.for_machine)
        linker_rule = linker_base + '_LINKER' + crstr
        # Create an empty commands list, and start adding link arguments from
        # various sources in the order in which they must override each other
        # starting from hard-coded defaults followed by build options and so on.
        #
        # Once all the linker options have been passed, we will start passing
        # libraries and library paths from internal and external sources.
        commands = linker.compiler_args()
        # First, the trivial ones that are impossible to override.
        #
        # Add linker args for linking this target derived from 'base' build
        # options passed on the command-line, in default_options, etc.
        # These have the lowest priority.
        if isinstance(target, build.StaticLibrary):
            commands += linker.get_base_link_args(self.get_options_for_target(target))
        else:
            commands += compilers.get_base_link_args(self.get_options_for_target(target),
                                                     linker,
                                                     isinstance(target, build.SharedModule))
        # Add -nostdlib if needed; can't be overridden
        commands += self.get_no_stdlib_link_args(target, linker)
        # Add things like /NOLOGO; usually can't be overridden
        commands += linker.get_linker_always_args()
        # Add buildtype linker args: optimization level, etc.
        commands += linker.get_buildtype_linker_args(self.get_option_for_target(OptionKey('buildtype'), target))
        # Add /DEBUG and the pdb filename when using MSVC
        if self.get_option_for_target(OptionKey('debug'), target):
            commands += self.get_link_debugfile_args(linker, target, outname)
            debugfile = self.get_link_debugfile_name(linker, target, outname)
            if debugfile is not None:
                implicit_outs += [debugfile]
        # Add link args specific to this BuildTarget type, such as soname args,
        # PIC, import library generation, etc.
        commands += self.get_target_type_link_args(target, linker)
        # Archives that are copied wholesale in the result. Must be before any
        # other link targets so missing symbols from whole archives are found in those.
        if not isinstance(target, build.StaticLibrary):
            commands += self.get_link_whole_args(linker, target)

        if not isinstance(target, build.StaticLibrary):
            # Add link args added using add_project_link_arguments()
            commands += self.build.get_project_link_args(linker, target.subproject, target.for_machine)
            # Add link args added using add_global_link_arguments()
            # These override per-project link arguments
            commands += self.build.get_global_link_args(linker, target.for_machine)
            # Link args added from the env: LDFLAGS. We want these to override
            # all the defaults but not the per-target link args.
            commands += self.environment.coredata.get_external_link_args(target.for_machine, linker.get_language())

        # Now we will add libraries and library paths from various sources

        # Set runtime-paths so we can run executables without needing to set
        # LD_LIBRARY_PATH, etc in the environment. Doesn't work on Windows.
        if has_path_sep(target.name):
            # Target names really should not have slashes in them, but
            # unfortunately we did not check for that and some downstream projects
            # now have them. Once slashes are forbidden, remove this bit.
            target_slashname_workaround_dir = os.path.join(
                os.path.dirname(target.name),
                self.get_target_dir(target))
        else:
            target_slashname_workaround_dir = self.get_target_dir(target)
        (rpath_args, target.rpath_dirs_to_remove) = (
            linker.build_rpath_args(self.environment,
                                    self.environment.get_build_dir(),
                                    target_slashname_workaround_dir,
                                    self.determine_rpath_dirs(target),
                                    target.build_rpath,
                                    target.install_rpath))
        commands += rpath_args

        # Add link args to link to all internal libraries (link_with:) and
        # internal dependencies needed by this target.
        if linker_base == 'STATIC':
            # Link arguments of static libraries are not put in the command
            # line of the library. They are instead appended to the command
            # line where the static library is used.
            dependencies = []
        else:
            dependencies = target.get_dependencies()
        internal = self.build_target_link_arguments(linker, dependencies)
        commands += internal
        # Only non-static built targets need link args and link dependencies
        if not isinstance(target, build.StaticLibrary):
            # For 'automagic' deps: Boost and GTest. Also dependency('threads').
            # pkg-config puts the thread flags itself via `Cflags:`

            commands += linker.get_target_link_args(target)
            # External deps must be last because target link libraries may depend on them.
            for dep in target.get_external_deps():
                # Extend without reordering or de-dup to preserve `-L -l` sets
                # https://github.com/mesonbuild/meson/issues/1718
                commands.extend_preserving_lflags(linker.get_dependency_link_args(dep))
            for d in target.get_dependencies():
                if isinstance(d, build.StaticLibrary):
                    for dep in d.get_external_deps():
                        commands.extend_preserving_lflags(linker.get_dependency_link_args(dep))

        # Add link args specific to this BuildTarget type that must not be overridden by dependencies
        commands += self.get_target_type_link_args_post_dependencies(target, linker)

        # Add link args for c_* or cpp_* build options. Currently this only
        # adds c_winlibs and cpp_winlibs when building for Windows. This needs
        # to be after all internal and external libraries so that unresolved
        # symbols from those can be found here. This is needed when the
        # *_winlibs that we want to link to are static mingw64 libraries.
        if isinstance(linker, Compiler):
            # The static linker doesn't know what language it is building, so we
            # don't know what option. Fortunately, it doesn't care to see the
            # language-specific options either.
            #
            # We shouldn't check whether we are making a static library, because
            # in the LTO case we do use a real compiler here.
            commands += linker.get_option_link_args(self.environment.coredata.options)

        dep_targets = []
        dep_targets.extend(self.guess_external_link_dependencies(linker, target, commands, internal))

        # Add libraries generated by custom targets
        custom_target_libraries = self.get_custom_target_provided_libraries(target)
        commands += extra_args
        commands += custom_target_libraries
        commands += stdlib_args # Standard library arguments go last, because they never depend on anything.
        dep_targets.extend([self.get_dependency_filename(t) for t in dependencies])
        dep_targets.extend([self.get_dependency_filename(t)
                            for t in target.link_depends])
        elem = NinjaBuildElement(self.all_outputs, outname, linker_rule, obj_list, implicit_outs=implicit_outs)
        elem.add_dep(dep_targets + custom_target_libraries)
        elem.add_item('LINK_ARGS', commands)
        return elem

    def get_dependency_filename(self, t):
        if isinstance(t, build.SharedLibrary):
            return self.get_target_shsym_filename(t)
        elif isinstance(t, mesonlib.File):
            if t.is_built:
                return t.relative_name()
            else:
                return t.absolute_path(self.environment.get_source_dir(),
                                       self.environment.get_build_dir())
        return self.get_target_filename(t)

    def generate_shlib_aliases(self, target, outdir):
        for alias, to, tag in target.get_aliases():
            aliasfile = os.path.join(self.environment.get_build_dir(), outdir, alias)
            try:
                os.remove(aliasfile)
            except Exception:
                pass
            try:
                os.symlink(to, aliasfile)
            except NotImplementedError:
                mlog.debug("Library versioning disabled because symlinks are not supported.")
            except OSError:
                mlog.debug("Library versioning disabled because we do not have symlink creation privileges.")

    def generate_custom_target_clean(self, trees: T.List[str]) -> str:
        e = NinjaBuildElement(self.all_outputs, 'meson-clean-ctlist', 'CUSTOM_COMMAND', 'PHONY')
        d = CleanTrees(self.environment.get_build_dir(), trees)
        d_file = os.path.join(self.environment.get_scratch_dir(), 'cleantrees.dat')
        e.add_item('COMMAND', self.environment.get_build_command() + ['--internal', 'cleantrees', d_file])
        e.add_item('description', 'Cleaning custom target directories')
        self.add_build(e)
        # Alias that runs the target defined above
        self.create_target_alias('meson-clean-ctlist')
        # Write out the data file passed to the script
        with open(d_file, 'wb') as ofile:
            pickle.dump(d, ofile)
        return 'clean-ctlist'

    def generate_gcov_clean(self):
        gcno_elem = NinjaBuildElement(self.all_outputs, 'meson-clean-gcno', 'CUSTOM_COMMAND', 'PHONY')
        gcno_elem.add_item('COMMAND', mesonlib.get_meson_command() + ['--internal', 'delwithsuffix', '.', 'gcno'])
        gcno_elem.add_item('description', 'Deleting gcno files')
        self.add_build(gcno_elem)
        # Alias that runs the target defined above
        self.create_target_alias('meson-clean-gcno')

        gcda_elem = NinjaBuildElement(self.all_outputs, 'meson-clean-gcda', 'CUSTOM_COMMAND', 'PHONY')
        gcda_elem.add_item('COMMAND', mesonlib.get_meson_command() + ['--internal', 'delwithsuffix', '.', 'gcda'])
        gcda_elem.add_item('description', 'Deleting gcda files')
        self.add_build(gcda_elem)
        # Alias that runs the target defined above
        self.create_target_alias('meson-clean-gcda')

    def get_user_option_args(self):
        cmds = []
        for (k, v) in self.environment.coredata.options.items():
            if k.is_project():
                cmds.append('-D' + str(k) + '=' + (v.value if isinstance(v.value, str) else str(v.value).lower()))
        # The order of these arguments must be the same between runs of Meson
        # to ensure reproducible output. The order we pass them shouldn't
        # affect behavior in any other way.
        return sorted(cmds)

    def generate_dist(self):
        elem = NinjaBuildElement(self.all_outputs, 'meson-dist', 'CUSTOM_COMMAND', 'PHONY')
        elem.add_item('DESC', 'Creating source packages')
        elem.add_item('COMMAND', self.environment.get_build_command() + ['dist'])
        elem.add_item('pool', 'console')
        self.add_build(elem)
        # Alias that runs the target defined above
        self.create_target_alias('meson-dist')

    def generate_scanbuild(self):
        if not environment.detect_scanbuild():
            return
        if ('', 'scan-build') in self.build.run_target_names:
            return
        cmd = self.environment.get_build_command() + \
            ['--internal', 'scanbuild', self.environment.source_dir, self.environment.build_dir] + \
            self.environment.get_build_command() + self.get_user_option_args()
        elem = NinjaBuildElement(self.all_outputs, 'meson-scan-build', 'CUSTOM_COMMAND', 'PHONY')
        elem.add_item('COMMAND', cmd)
        elem.add_item('pool', 'console')
        self.add_build(elem)
        # Alias that runs the target defined above
        self.create_target_alias('meson-scan-build')

    def generate_clangtool(self, name, extra_arg=None):
        target_name = 'clang-' + name
        extra_args = []
        if extra_arg:
            target_name += f'-{extra_arg}'
            extra_args.append(f'--{extra_arg}')
        if not os.path.exists(os.path.join(self.environment.source_dir, '.clang-' + name)) and \
                not os.path.exists(os.path.join(self.environment.source_dir, '_clang-' + name)):
            return
        if target_name in self.all_outputs:
            return
        if ('', target_name) in self.build.run_target_names:
            return
        cmd = self.environment.get_build_command() + \
            ['--internal', 'clang' + name, self.environment.source_dir, self.environment.build_dir] + \
            extra_args
        elem = NinjaBuildElement(self.all_outputs, 'meson-' + target_name, 'CUSTOM_COMMAND', 'PHONY')
        elem.add_item('COMMAND', cmd)
        elem.add_item('pool', 'console')
        self.add_build(elem)
        self.create_target_alias('meson-' + target_name)

    def generate_clangformat(self):
        if not environment.detect_clangformat():
            return
        self.generate_clangtool('format')
        self.generate_clangtool('format', 'check')

    def generate_clangtidy(self):
        import shutil
        if not shutil.which('clang-tidy'):
            return
        self.generate_clangtool('tidy')

    def generate_tags(self, tool, target_name):
        import shutil
        if not shutil.which(tool):
            return
        if ('', target_name) in self.build.run_target_names:
            return
        if target_name in self.all_outputs:
            return
        cmd = self.environment.get_build_command() + \
            ['--internal', 'tags', tool, self.environment.source_dir]
        elem = NinjaBuildElement(self.all_outputs, 'meson-' + target_name, 'CUSTOM_COMMAND', 'PHONY')
        elem.add_item('COMMAND', cmd)
        elem.add_item('pool', 'console')
        self.add_build(elem)
        # Alias that runs the target defined above
        self.create_target_alias('meson-' + target_name)

    # For things like scan-build and other helper tools we might have.
    def generate_utils(self):
        self.generate_scanbuild()
        self.generate_clangformat()
        self.generate_clangtidy()
        self.generate_tags('etags', 'TAGS')
        self.generate_tags('ctags', 'ctags')
        self.generate_tags('cscope', 'cscope')
        cmd = self.environment.get_build_command() + ['--internal', 'uninstall']
        elem = NinjaBuildElement(self.all_outputs, 'meson-uninstall', 'CUSTOM_COMMAND', 'PHONY')
        elem.add_item('COMMAND', cmd)
        elem.add_item('pool', 'console')
        self.add_build(elem)
        # Alias that runs the target defined above
        self.create_target_alias('meson-uninstall')

    def generate_ending(self):
        targetlist = []
        for t in self.get_build_by_default_targets().values():
            # Add the first output of each target to the 'all' target so that
            # they are all built
            targetlist.append(os.path.join(self.get_target_dir(t), t.get_outputs()[0]))

        elem = NinjaBuildElement(self.all_outputs, 'all', 'phony', targetlist)
        self.add_build(elem)

        elem = NinjaBuildElement(self.all_outputs, 'meson-clean', 'CUSTOM_COMMAND', 'PHONY')
        elem.add_item('COMMAND', self.ninja_command + ['-t', 'clean'])
        elem.add_item('description', 'Cleaning')
        # Alias that runs the above-defined meson-clean target
        self.create_target_alias('meson-clean')

        # If we have custom targets in this project, add all their outputs to
        # the list that is passed to the `cleantrees.py` script. The script
        # will manually delete all custom_target outputs that are directories
        # instead of files. This is needed because on platforms other than
        # Windows, Ninja only deletes directories while cleaning if they are
        # empty. https://github.com/mesonbuild/meson/issues/1220
        ctlist = []
        for t in self.build.get_targets().values():
            if isinstance(t, build.CustomTarget):
                # Create a list of all custom target outputs
                for o in t.get_outputs():
                    ctlist.append(os.path.join(self.get_target_dir(t), o))
        if ctlist:
            elem.add_dep(self.generate_custom_target_clean(ctlist))

        if OptionKey('b_coverage') in self.environment.coredata.options and \
           self.environment.coredata.options[OptionKey('b_coverage')].value:
            self.generate_gcov_clean()
            elem.add_dep('clean-gcda')
            elem.add_dep('clean-gcno')
        self.add_build(elem)

        deps = self.get_regen_filelist()
        elem = NinjaBuildElement(self.all_outputs, 'build.ninja', 'REGENERATE_BUILD', deps)
        elem.add_item('pool', 'console')
        self.add_build(elem)

        elem = NinjaBuildElement(self.all_outputs, 'reconfigure', 'REGENERATE_BUILD', 'PHONY')
        elem.add_item('pool', 'console')
        self.add_build(elem)

        elem = NinjaBuildElement(self.all_outputs, deps, 'phony', '')
        self.add_build(elem)

    def get_introspection_data(self, target_id: str, target: build.Target) -> T.List[T.Dict[str, T.Union[bool, str, T.List[T.Union[str, T.Dict[str, T.Union[str, T.List[str], bool]]]]]]]:
        if target_id not in self.introspection_data or len(self.introspection_data[target_id]) == 0:
            return super().get_introspection_data(target_id, target)

        result = []
        for i in self.introspection_data[target_id].values():
            result += [i]
        return result


def _scan_fortran_file_deps(src: Path, srcdir: Path, dirname: Path, tdeps, compiler) -> T.List[str]:
    """
    scan a Fortran file for dependencies. Needs to be distinct from target
    to allow for recursion induced by `include` statements.er

    It makes a number of assumptions, including

    * `use`, `module`, `submodule` name is not on a continuation line

    Regex
    -----

    * `incre` works for `#include "foo.f90"` and `include "foo.f90"`
    * `usere` works for legacy and Fortran 2003 `use` statements
    * `submodre` is for Fortran >= 2008 `submodule`
    """

    incre = re.compile(FORTRAN_INCLUDE_PAT, re.IGNORECASE)
    usere = re.compile(FORTRAN_USE_PAT, re.IGNORECASE)
    submodre = re.compile(FORTRAN_SUBMOD_PAT, re.IGNORECASE)

    mod_files = []
    src = Path(src)
    with src.open(encoding='ascii', errors='ignore') as f:
        for line in f:
            # included files
            incmatch = incre.match(line)
            if incmatch is not None:
                incfile = src.parent / incmatch.group(1)
                # NOTE: src.parent is most general, in particular for CMake subproject with Fortran file
                # having an `include 'foo.f'` statement.
                if incfile.suffix.lower()[1:] in compiler.file_suffixes:
                    mod_files.extend(_scan_fortran_file_deps(incfile, srcdir, dirname, tdeps, compiler))
            # modules
            usematch = usere.match(line)
            if usematch is not None:
                usename = usematch.group(1).lower()
                if usename == 'intrinsic':  # this keeps the regex simpler
                    continue
                if usename not in tdeps:
                    # The module is not provided by any source file. This
                    # is due to:
                    #   a) missing file/typo/etc
                    #   b) using a module provided by the compiler, such as
                    #      OpenMP
                    # There's no easy way to tell which is which (that I
                    # know of) so just ignore this and go on. Ideally we
                    # would print a warning message to the user but this is
                    # a common occurrence, which would lead to lots of
                    # distracting noise.
                    continue
                srcfile = srcdir / tdeps[usename].fname  # type: Path
                if not srcfile.is_file():
                    if srcfile.name != src.name:  # generated source file
                        pass
                    else:  # subproject
                        continue
                elif srcfile.samefile(src):  # self-reference
                    continue

                mod_name = compiler.module_name_to_filename(usename)
                mod_files.append(str(dirname / mod_name))
            else:  # submodules
                submodmatch = submodre.match(line)
                if submodmatch is not None:
                    parents = submodmatch.group(1).lower().split(':')
                    assert len(parents) in (1, 2), (
                        'submodule ancestry must be specified as'
                        f' ancestor:parent but Meson found {parents}')

                    ancestor_child = '_'.join(parents)
                    if ancestor_child not in tdeps:
                        raise MesonException("submodule {} relies on ancestor module {} that was not found.".format(submodmatch.group(2).lower(), ancestor_child.split('_')[0]))
                    submodsrcfile = srcdir / tdeps[ancestor_child].fname  # type: Path
                    if not submodsrcfile.is_file():
                        if submodsrcfile.name != src.name:  # generated source file
                            pass
                        else:  # subproject
                            continue
                    elif submodsrcfile.samefile(src):  # self-reference
                        continue
                    mod_name = compiler.module_name_to_filename(ancestor_child)
                    mod_files.append(str(dirname / mod_name))
    return mod_files