lix/meson.build
eldritch horrors e6cd67591b libexpr: rewrite the parser with pegtl instead of flex/bison
this gives about 20% performance improvements on pure parsing. obviously
it will be less on full eval, but depending on how much parsing is to be
done (e.g. including hackage-packages.nix or not) it's more like 4%-10%.

this has been tested (with thousands of core hours of fuzzing) to ensure
that the ASTs produced by the new parser are exactly the same as the old
one would have produced. error messages will change (sometimes by a lot)
and are not yet perfect, but we would rather leave this as is for later.

test results for running only the parser (excluding the variable binding
code) in a tight loop with inputs and parameters as given are promising:

  - 40% faster on lix's package.nix at 10000 iterations
  - 1.3% faster on nixpkgs all-packages.nix at 1000 iterations
  - equivalent on all of nixpkgs concatenated at 100 iterations
    (excluding invalid files, each file surrounded with parens)

more realistic benchmarks are somewhere in between the extremes, parsing
once again getting the largest uplift. other realistic workloads improve
by a few percentage points as well, notably system builds are 4% faster.

Benchmarks summary (from ./bench/summarize.jq bench/bench-*.json)
old/bin/nix --extra-experimental-features 'nix-command flakes' eval -f bench/nixpkgs/pkgs/development/haskell-modules/hackage-packages.nix
  mean:     0.408s ± 0.025s
            user: 0.355s | system: 0.033s
  median:   0.389s
  range:    0.388s ... 0.442s
  relative: 1

new/bin/nix --extra-experimental-features 'nix-command flakes' eval -f bench/nixpkgs/pkgs/development/haskell-modules/hackage-packages.nix
  mean:     0.332s ± 0.024s
            user: 0.279s | system: 0.033s
  median:   0.314s
  range:    0.313s ... 0.361s
  relative: 0.814

---

old/bin/nix --extra-experimental-features 'nix-command flakes' eval --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'
  mean:     6.133s ± 0.022s
            user: 5.395s | system: 0.437s
  median:   6.128s
  range:    6.099s ... 6.183s
  relative: 1

new/bin/nix --extra-experimental-features 'nix-command flakes' eval --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'
  mean:     5.925s ± 0.025s
            user: 5.176s | system: 0.456s
  median:   5.934s
  range:    5.861s ... 5.943s
  relative: 0.966

---

GC_INITIAL_HEAP_SIZE=10g old/bin/nix eval --extra-experimental-features 'nix-command flakes' --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'
  mean:     4.503s ± 0.027s
            user: 3.731s | system: 0.547s
  median:   4.499s
  range:    4.478s ... 4.541s
  relative: 1

GC_INITIAL_HEAP_SIZE=10g new/bin/nix eval --extra-experimental-features 'nix-command flakes' --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'
  mean:     4.285s ± 0.031s
            user: 3.504s | system: 0.571s
  median:   4.281s
  range:    4.221s ... 4.328s
  relative: 0.951

---

old/bin/nix --extra-experimental-features 'nix-command flakes' search --no-eval-cache github:nixos/nixpkgs/e1fa12d4f6c6fe19ccb59cac54b5b3f25e160870 hello
  mean:     16.475s ± 0.07s
            user: 14.088s | system: 1.572s
  median:   16.495s
  range:    16.351s ... 16.536s
  relative: 1

new/bin/nix --extra-experimental-features 'nix-command flakes' search --no-eval-cache github:nixos/nixpkgs/e1fa12d4f6c6fe19ccb59cac54b5b3f25e160870 hello
  mean:     15.973s ± 0.013s
            user: 13.558s | system: 1.615s
  median:   15.973s
  range:    15.946s ... 15.99s
  relative: 0.97

---

Change-Id: Ie66ec2d045dec964632c6541e25f8f0797319ee2
2024-06-25 12:24:58 +00:00

524 lines
18 KiB
Meson

#
# OUTLINE:
#
# The top-level meson.build file (this file) handles general logic for build options,
# generation of config.h (which is put in the build directory, not the source root
# like the previous, autoconf-based build system did), the mechanism for header
# generation, and the few global C++ compiler arguments that are added to all targets in Lix.
#
# src/meson.build coordinates each of Lix's subcomponents (the lib dirs in ./src),
# which each have their own meson.build. Lix's components depend on each other,
# so each of `src/lib{util,store,fetchers,expr,main,cmd}/meson.build` rely on variables
# set in earlier `meson.build` files. Each of these also defines the install targets for
# their headers.
#
# src/meson.build also collects the miscellaneous source files that are in further subdirectories
# that become part of the final Nix command (things like `src/nix-build/*.cc`).
#
# Finally, src/nix/meson.build defines the Nix command itself, relying on all prior meson files.
#
# libstore, libexpr, and libfetchers have some special handling to make static builds work.
# Their use static constructors for dynamic registration of primops, store backends, etc
# gets borked during static link. We can't simply wholesale apply `link_whole :` either,
# because these libraries get linked multiple times since Lix's components are transitively
# dependent. So instead, each of those libraries have two dependency objects:
# liblix{store,expr,fetchers,util} and liblix{store,expr,fetchers,util}_mstatic ("maybe static").
# The _mstatic versions should be used in the `dependencies :` arguments to ALL EXECUTABLES
# but executables ONLY. When we are not building statically (default_library != 'static'),
# they are equivalent. When we are building statically, the _mstatic version will be
# `link_whole :` rather than `link_with :`.
# FIXME: This hack should be removed when https://git.lix.systems/lix-project/lix/issues/359
# is fixed.
#
# Unit tests are setup in tests/unit/meson.build, under the test suite "check".
#
# Functional tests are a bit more complicated. Generally they're defined in
# tests/functional/meson.build, and rely on helper scripts meson/setup-functional-tests.py
# and meson/run-test.py. Scattered around also are configure_file() invocations, which must
# be placed in specific directories' meson.build files to create the right directory tree
# in the build directory.
project('lix', 'cpp',
version : run_command('bash', '-c', 'echo -n $(jq -r .version < ./version.json)$VERSION_SUFFIX', check : true).stdout().strip(),
default_options : [
'cpp_std=c++2a',
# TODO(Qyriad): increase the warning level
'warning_level=1',
'debug=true',
'optimization=2',
'errorlogs=true', # Please print logs for tests that fail
],
)
fs = import('fs')
prefix = get_option('prefix')
# For each of these paths, assume that it is relative to the prefix unless
# it is already an absolute path (which is the default for store-dir, state-dir, and log-dir).
path_opts = [
# Meson built-ins.
'datadir',
'bindir',
'mandir',
'libdir',
'includedir',
'libexecdir',
# Homecooked Lix directories.
'store-dir',
'state-dir',
'log-dir',
'profile-dir',
]
# For your grepping pleasure, this loop sets the following variables that aren't mentioned
# literally above:
# store_dir
# state_dir
# log_dir
# profile_dir
foreach optname : path_opts
varname = optname.replace('-', '_')
path = get_option(optname)
if fs.is_absolute(path)
set_variable(varname, path)
else
set_variable(varname, prefix / path)
endif
endforeach
# sysconfdir doesn't get anything installed to directly, and is only used to
# tell Lix where to look for nix.conf, so it doesn't get appended to prefix.
sysconfdir = get_option('sysconfdir')
if not fs.is_absolute(sysconfdir)
sysconfdir = '/' / sysconfdir
endif
is_static = get_option('default_library') == 'static'
# All of this has to go before the rest of the dependency checking,
# so that internal-api-docs can be built with -Denable-build=false
enable_docs = get_option('enable-docs')
enable_internal_api_docs = get_option('internal-api-docs')
doxygen = find_program('doxygen', required : enable_internal_api_docs, native : true)
bash = find_program('bash', native : true)
rapidcheck_meson = dependency('rapidcheck', required : enable_internal_api_docs)
if enable_internal_api_docs.enabled()
message('subdiring()')
subdir('doc/internal-api')
endif
if not get_option('enable-build')
subdir_done()
endif
enable_tests = get_option('enable-tests')
tests_args = []
if get_option('tests-color')
tests_args += '--gtest_color=yes'
endif
if get_option('tests-brief')
tests_args += '--gtest_brief=1'
endif
cxx = meson.get_compiler('cpp')
# clangd breaks when GCC is using precompiled headers lmao
# https://git.lix.systems/lix-project/lix/issues/374
should_pch = get_option('enable-pch-std')
summary('PCH C++ stdlib', should_pch, bool_yn : true)
if should_pch
# Unlike basically everything else that takes a file, Meson requires the arguments to
# cpp_pch : to be strings and doesn't accept files(). So absolute path it is.
cpp_pch = [meson.project_source_root() / 'src/pch/precompiled-headers.hh']
else
cpp_pch = []
endif
# Translate some historical and Mesony CPU names to Lixy CPU names.
# FIXME(Qyriad): the 32-bit x86 code is not tested right now, because cross compilation for Lix
# to those architectures is currently broken for other reasons, namely:
# - nixos-23.11's x86_64-linux -> i686-linux glibc does not build (also applies to cppnix)
# - nixpkgs-unstable (as of 2024/04)'s boehmgc is not compatible with our patches
# It's also broken in cppnix, though.
host_cpu = host_machine.cpu_family()
if host_cpu in ['x86', 'i686', 'i386']
# Meson considers 32-bit x86 CPUs to be "x86", and does not consider 64-bit
# x86 CPUs to be "x86" (instead using "x86_64", which needs no translation).
host_cpu = 'i686'
elif host_cpu == 'amd64'
# This should not be needed under normal circumstances, but someone could pass a --cross-file
# that sets the cpu_family to this.
host_cpu = 'x86_64'
elif host_cpu in ['armv6', 'armv7']
host_cpu += 'l'
endif
host_system = host_machine.cpu_family() + '-' + host_machine.system()
message('canonical Nix system name:', host_system)
is_linux = host_machine.system() == 'linux'
is_darwin = host_machine.system() == 'darwin'
is_freebsd = host_machine.system() == 'freebsd'
is_x64 = host_machine.cpu_family() == 'x86_64'
# Per-platform arguments that you should probably pass to shared_module() invocations.
# Something like add_project_arguments() can't be scoped on only shared modules, so this
# variable is here instead.
# This corresponds to the $(1)_ALLOW_UNDEFINED option from the Make buildsystem.
# Mostly this is load-bearing on the plugin tests defined in tests/functional/plugins/meson.build.
shared_module_link_args = []
if is_darwin
shared_module_link_args += ['-undefined', 'suppress', '-flat_namespace']
elif is_linux
# -Wl,-z,defs is the equivalent, but a comment in the Make buildsystem says that breaks
# Clang sanitizers on Linux.
# FIXME(Qyriad): is that true?
endif
configdata = { }
#
# Dependencies
#
boehm = dependency('bdw-gc', required : get_option('gc'), version : '>=8.2.6')
configdata += {
'HAVE_BOEHMGC': boehm.found().to_int(),
}
boost = dependency('boost', required : true, modules : ['context', 'coroutine', 'container'])
# cpuid only makes sense on x86_64
cpuid_required = is_x64 ? get_option('cpuid') : false
cpuid = dependency('libcpuid', 'cpuid', required : cpuid_required)
configdata += {
'HAVE_LIBCPUID': cpuid.found().to_int(),
}
# seccomp only makes sense on Linux
seccomp_required = is_linux ? get_option('seccomp-sandboxing') : false
seccomp = dependency('libseccomp', 'seccomp', required : seccomp_required, version : '>=2.5.5')
if is_linux and not seccomp.found()
warning('Sandbox security is reduced because libseccomp has not been found! Please provide libseccomp if it supports your CPU architecture.')
endif
configdata += {
'HAVE_SECCOMP': seccomp.found().to_int(),
}
libarchive = dependency('libarchive', required : true)
brotli = [
dependency('libbrotlicommon', required : true),
dependency('libbrotlidec', required : true),
dependency('libbrotlienc', required : true),
]
openssl = dependency('libcrypto', 'openssl', required : true)
# FIXME: confirm we actually support such old versions of aws-sdk-cpp
aws_sdk = dependency('aws-cpp-sdk-core', required : false, version : '>=1.8')
aws_sdk_transfer = dependency('aws-cpp-sdk-transfer', required : aws_sdk.found(), fallback : ['aws_sdk', 'aws_cpp_sdk_transfer_dep'])
if aws_sdk.found()
# The AWS pkg-config adds -std=c++11.
# https://github.com/aws/aws-sdk-cpp/issues/2673
aws_sdk = aws_sdk.partial_dependency(
compile_args : false,
includes : true,
link_args : true,
links : true,
sources : true,
)
aws_sdk_transfer = aws_sdk_transfer.partial_dependency(
compile_args : false,
includes : true,
link_args : true,
links : true,
sources : true,
)
endif
aws_s3 = dependency('aws-cpp-sdk-s3', required : aws_sdk.found(), fallback : ['aws_sdk', 'aws_cpp_sdk_s3_dep'])
if aws_s3.found()
# The AWS pkg-config adds -std=c++11.
# https://github.com/aws/aws-sdk-cpp/issues/2673
aws_s3 = aws_s3.partial_dependency(
compile_args : false,
includes : true,
link_args : true,
links : true,
sources : true,
)
endif
configdata += {
'ENABLE_S3': aws_s3.found().to_int(),
}
sqlite = dependency('sqlite3', 'sqlite', version : '>=3.6.19', required : true)
sodium = dependency('libsodium', 'sodium', required : true)
curl = dependency('libcurl', 'curl', required : true)
editline = dependency('libeditline', 'editline', version : '>=1.14', required : true)
lowdown = dependency('lowdown', version : '>=0.9.0', required : true)
# HACK(Qyriad): rapidcheck's pkg-config doesn't include the libs lol
# Note: technically we 'check' for rapidcheck twice, for the internal-api-docs handling above,
# but Meson will cache the result of the first one, and the required : arguments are different.
rapidcheck_meson = dependency('rapidcheck', required : enable_tests)
rapidcheck = declare_dependency(dependencies : rapidcheck_meson, link_args : ['-lrapidcheck'])
gtest = [
dependency('gtest', required : enable_tests),
dependency('gtest_main', required : enable_tests),
dependency('gmock', required : enable_tests),
dependency('gmock_main', required : enable_tests),
]
toml11 = dependency('toml11', version : '>=3.7.0', required : true, method : 'cmake')
pegtl = dependency(
'pegtl',
version : '>=3.2.7',
required : true,
method : 'cmake',
modules : [ 'taocpp::pegtl' ],
)
nlohmann_json = dependency('nlohmann_json', required : true)
# lix-doc is a Rust project provided via buildInputs and unfortunately doesn't have any way to be detected.
# Just declare it manually to resolve this.
#
# FIXME: build this with meson in the future after we drop Make (with which we
# *absolutely* are not going to make it work)
lix_doc = declare_dependency(link_args : [ '-llix_doc' ])
#
# Build-time tools
#
coreutils = find_program('coreutils', native : true)
dot = find_program('dot', required : false, native : true)
pymod = import('python')
python = pymod.find_installation('python3')
if enable_docs
mdbook = find_program('mdbook', native : true)
endif
# Used to workaround https://github.com/mesonbuild/meson/issues/2320 in src/nix/meson.build.
installcmd = find_program('install', native : true)
enable_embedded_sandbox_shell = get_option('enable-embedded-sandbox-shell')
if enable_embedded_sandbox_shell
# This one goes in config.h
# The path to busybox is passed as a -D flag when compiling libstore.
# Idk why, ask the old buildsystem.
configdata += {
'HAVE_EMBEDDED_SANDBOX_SHELL': 1,
}
endif
sandbox_shell = get_option('sandbox-shell')
# Consider it required if we're on Linux and the user explicitly specified a non-default value.
sandbox_shell_required = sandbox_shell != 'busybox' and host_machine.system() == 'linux'
# NOTE(Qyriad): package.nix puts busybox in buildInputs for Linux.
# Most builds should not require setting this.
busybox = find_program(sandbox_shell, required : sandbox_shell_required, native : false)
if not busybox.found() and host_machine.system() == 'linux' and sandbox_shell_required
warning('busybox not found and other sandbox shell was specified')
warning('a sandbox shell is recommended on Linux -- configure with -Dsandbox-shell=/path/to/shell to set')
endif
# FIXME(Qyriad): the autoconf system checks that busybox has the "standalone" feature, indicating
# that busybox sh won't run busybox applets as builtins (which would break our sandbox).
lsof = find_program('lsof', native : true)
# This is how Nix does generated headers...
# other instances of header generation use a very similar command.
# FIXME(Qyriad): do we really need to use the shell for this?
gen_header_sh = 'echo \'R"__NIX_STR(\' | cat - @INPUT@ && echo \')__NIX_STR"\''
gen_header = generator(
bash,
arguments : [ '-c', gen_header_sh ],
capture : true,
output : '@PLAINNAME@.gen.hh',
)
#
# Configuration
#
run_command('ln', '-s',
meson.project_build_root() / '__nothing_link_target',
meson.project_build_root() / '__nothing_symlink',
check : true,
)
can_link_symlink = run_command('ln',
meson.project_build_root() / '__nothing_symlink',
meson.project_build_root() / '__nothing_hardlink',
check : false,
).returncode() == 0
run_command('rm', '-f',
meson.project_build_root() / '__nothing_symlink',
meson.project_build_root() / '__nothing_hardlink',
check : true,
)
summary('can hardlink to symlink', can_link_symlink, bool_yn : true)
configdata += { 'CAN_LINK_SYMLINK': can_link_symlink.to_int() }
# Check for each of these functions, and create a define like `#define HAVE_LCHOWN 1`.
check_funcs = [
'lchown',
'lutimes',
'pipe2',
'posix_fallocate',
'statvfs',
'strsignal',
'sysconf',
]
foreach funcspec : check_funcs
define_name = 'HAVE_' + funcspec.underscorify().to_upper()
define_value = cxx.has_function(funcspec).to_int()
configdata += {
define_name: define_value,
}
endforeach
config_h = configure_file(
configuration : {
'PACKAGE_NAME': '"' + meson.project_name() + '"',
'PACKAGE_VERSION': '"' + meson.project_version() + '"',
'PACKAGE_TARNAME': '"' + meson.project_name() + '"',
'PACKAGE_STRING': '"' + meson.project_name() + ' ' + meson.project_version() + '"',
'HAVE_STRUCT_DIRENT_D_TYPE': 1, # FIXME: actually check this for solaris
'SYSTEM': '"' + host_system + '"',
} + configdata,
output : 'config.h',
)
install_headers(config_h, subdir : 'lix')
# FIXME: not using the pkg-config module because it creates way too many deps
# while meson migration is in progress, and we want to not include boost here
configure_file(
input : 'src/lix-base.pc.in',
output : 'lix-base.pc',
install_dir : libdir / 'pkgconfig',
configuration : {
'prefix' : prefix,
'libdir' : libdir,
'includedir' : includedir,
'PACKAGE_VERSION' : meson.project_version(),
},
)
add_project_arguments(
# TODO(Qyriad): Yes this is how the autoconf+Make system did it.
# It would be nice for our headers to be idempotent instead.
'-include', 'config.h',
'-Wno-deprecated-declarations',
'-Wimplicit-fallthrough',
'-Werror=switch',
'-Werror=switch-enum',
'-Wdeprecated-copy',
'-Wignored-qualifiers',
# Enable assertions in libstdc++ by default. Harmless on libc++. Benchmarked
# at ~1% overhead in `nix search`.
#
# FIXME: remove when we get meson 1.4.0 which will default this to on for us:
# https://mesonbuild.com/Release-notes-for-1-4-0.html#ndebug-setting-now-controls-c-stdlib-assertions
'-D_GLIBCXX_ASSERTIONS=1',
language : 'cpp',
)
# We turn off the production UBSan if the slower dev UBSan is requested, to
# give better diagnostics.
if cxx.get_id() in ['gcc', 'clang'] and 'undefined' not in get_option('b_sanitize')
# 2024-03-24: jade benchmarked the default sanitize reporting in clang and got
# a regression of about 10% on hackage-packages.nix with clang. So we are trapping instead.
#
# This has an unmeasurably low overhead in Nix evaluation benchmarks.
#
# N.B. Meson generates a completely nonsense warning here:
# https://github.com/mesonbuild/meson/issues/9822
# Both of these args cannot be written in the default meson configuration.
# b_sanitize=signed-integer-overflow is ignored, and
# -fsanitize-undefined-trap-on-error is not representable.
sanitize_args = ['-fsanitize=signed-integer-overflow', '-fsanitize-undefined-trap-on-error']
add_project_arguments(sanitize_args, language: 'cpp')
add_project_link_arguments(sanitize_args, language: 'cpp')
endif
# Clang's default of -no-shared-libsan on Linux causes link errors; on macOS it defaults to shared.
# GCC defaults to shared libsan so is fine.
if cxx.get_id() == 'clang' and get_option('b_sanitize') != ''
add_project_link_arguments('-shared-libsan', language : 'cpp')
endif
add_project_link_arguments('-pthread', language : 'cpp')
if cxx.get_linker_id() in ['ld.bfd', 'ld.gold']
add_project_link_arguments('-Wl,--no-copy-dt-needed-entries', language : 'cpp')
endif
if is_freebsd
# FreeBSD's `environ` is defined in `crt1.o`, not `libc.so`,
# so the linker thinks it's undefined
add_project_link_arguments('-Wl,-z,undefs', language: 'cpp')
endif
# Generate Chromium tracing files for each compiled file, which enables
# maintainers/buildtime_report.sh BUILD-DIR to simply work in clang builds.
#
# They can also be manually viewed at https://ui.perfetto.dev
if get_option('profile-build').require(meson.get_compiler('cpp').get_id() == 'clang').enabled()
add_project_arguments('-ftime-trace', language: 'cpp')
endif
if cxx.get_id() in ['clang', 'gcc']
add_project_arguments([
# Meson uses out of source builds, conventionally usually in a subdirectory
# of the source tree (e.g. meson setup ./build). This means that unlike in
# the previous Make buildsystem, all compilation sources are passed as a relative
# parent, e.g. `cc -o src/libexpr/nixexpr.cc.o ../src/libexpr/nixexpr.cc`.
# These paths show up when debugging, and in asserts, which look both look strange
# and confuse debuggers.
# So let's just tell GCC and Clang that ../src really means src.
'-ffile-prefix-map=../src=src',
],
language : 'cpp',
)
endif
if is_darwin
configure_file(
input : 'misc/launchd/org.nixos.nix-daemon.plist.in',
output : 'org.nixos.nix-daemon.plist',
copy : true,
install_dir : prefix / 'Library/LaunchDaemons',
)
endif
subdir('src')
subdir('scripts')
subdir('misc')
if enable_docs
subdir('doc/manual')
endif
if enable_tests
subdir('tests/unit')
subdir('tests/functional')
endif