.bazelrc

common --enable_platform_specific_config
build --verbose_failures
build --build_tag_filters=-off-by-default
test --test_tag_filters=-off-by-default
# exclude enormous tests by default
build --test_size_filters=-enormous

# use lower test timeouts: https://bazel.build/reference/test-encyclopedia#role-test-runner
# corresponds to small,medium,large,enormous tests (medium is default)
build --test_timeout=1,15,60,240

# Not using bzlmod for dependencies yet
common --noenable_bzlmod

# bazel7 enables Build without the Bytes (BwoB) by default. This significantly speeds up builds
# using the remote cache since less data needs to be fetched. Set remote_cache_eviction_retries
# just in case there are cache evictions during the build.
build --experimental_remote_cache_eviction_retries=3
# Note that we use remote_download_minimal for test builds, which avoids fetching build outputs
# where possible. While several previous BwoB bugs have been fixed, this is slower than it could be
# due to https://github.com/bazelbuild/bazel/issues/20576.
# Windows build is failing on some runs due to FileAccessException errors, Bazel has had several
# similar bugs in the past. Disable BwoB for now.
build:windows --remote_download_all

# Enable webgpu
build --//src/workerd/io:enable_experimental_webgpu=True

# Avoid generating duplicate runfile trees. This will become the default in a future bazel version.
build --nolegacy_external_runfiles

# Flip this early to avoid breaking compatibility once it becomes the default.
build --incompatible_disallow_empty_glob

# Prevents bazel cache invalidation when switching terminals
build --incompatible_strict_action_env

# Dawn tint build flags
build --@dawn//src/tint:tint_build_glsl_writer=False
build --@dawn//src/tint:tint_build_glsl_validator=False
build --@dawn//src/tint:tint_build_hlsl_writer=True
build --@dawn//src/tint:tint_build_ir=False
build --@dawn//src/tint:tint_build_msl_writer=True
build --@dawn//src/tint:tint_build_spv_reader=False
build --@dawn//src/tint:tint_build_spv_writer=True
build --@dawn//src/tint:tint_build_wgsl_reader=True
build --@dawn//src/tint:tint_build_wgsl_writer=True

# Our dependencies (ICU, zlib, etc.) produce a lot of these warnings, so we disable them. Depending
# on the clang version, zlib either produces warnings for -Wdeprecated-non-prototype or does not
# have that option, so disable -Wunknown-warning-option there too.
build --per_file_copt='external@-Wno-error'
build --per_file_copt='external/v8@-Wno-deprecated-declarations'
build --per_file_copt='external/com_googlesource_chromium_icu@-Wno-ambiguous-reversed-operator,-Wno-deprecated-declarations'
build --host_per_file_copt='external/com_googlesource_chromium_icu@-Wno-ambiguous-reversed-operator,-Wno-deprecated-declarations'
build --per_file_copt='external/dawn@-Wno-shorten-64-to-32,-Wno-unneeded-internal-declaration'
build --host_per_file_copt='external/dawn@-Wno-shorten-64-to-32,-Wno-unneeded-internal-declaration'
build --per_file_copt='external/zlib@-Wno-unknown-warning-option,-Wno-deprecated-non-prototype'
build --host_per_file_copt='external/zlib@-Wno-unknown-warning-option,-Wnodeprecated-non-prototype'
# for some reason tcmalloc uses internal msan-enabled annotation in the code, rather than public one
build --per_file_copt=external/com_google_tcmalloc@-DANNOTATE_MEMORY_IS_INITIALIZED=ABSL_ANNOTATE_MEMORY_IS_INITIALIZED
build --per_file_copt=external/com_google_protobuf@-Wno-deprecated-declarations,-Wno-deprecated-pragma
build --host_per_file_copt=external/com_google_protobuf@-Wno-deprecated-declarations,-Wno-deprecated-pragma

# Increasing the optimization level of some portions of V8 significantly speeds up Python tests
# in GitHub Actions. This is an optional performance hack and shouly only be used in CI.
build:v8-codegen-opt --per_file_copt=v8/src/compiler@-O2
build:v8-codegen-opt --per_file_copt=v8/src/common@-O2
build:v8-codegen-opt --per_file_copt=v8/src/builtins@-O2
build:v8-codegen-opt --per_file_copt=v8/src/heap@-O2
build:v8-codegen-opt --per_file_copt=v8/src/snapshot@-O2
build:v8-codegen-opt --per_file_copt=v8/src/objects@-O2
build:v8-codegen-opt --per_file_copt=v8/src/wasm@-O2
build:v8-codegen-opt --per_file_copt=v8/src/base@-O2

# Disable relaxing all jumps during LLVM codegen under -O0, which previously led to build
# performance improvements but makes code size worse. This will be the default in LLVM19.
# https://maskray.me/blog/2024-04-27-clang-o0-output-branch-displacement-and-size-increase
build:unix --copt="-mno-relax-all" --host_copt="-mno-relax-all"

# Limit transitive header includes within libc++. This improves compliance with IWYU, helps avoid
# errors with downstream projects that implicitly define this already and reduces total include size.
https://libcxx.llvm.org/DesignDocs/HeaderRemovalPolicy.html
build:unix --cxxopt=-D_LIBCPP_REMOVE_TRANSITIVE_INCLUDES
build:unix --host_cxxopt=-D_LIBCPP_REMOVE_TRANSITIVE_INCLUDES

# Need to redefine _WIN32_WINNT to build dawn on Windows
build:windows --per_file_copt='external/dawn@-Wno-macro-redefined'
build:windows --host_per_file_copt='external/dawn@-Wno-macro-redefined'
# V8 now requires Windows 10, Bazel fails to set this properly.
build:windows --per_file_copt=external/v8/src/base/platform/platform-win32.cc@-D_WIN32_WINNT=0x0A00
build:windows --host_per_file_copt=external/v8/src/base/platform/platform-win32.cc@-D_WIN32_WINNT=0x0A00

# typescript configuration
# do more correct type checking
common --@aspect_rules_ts//ts:skipLibCheck=honor_tsconfig
# Use "tsc" as the transpiler when ts_project has no `transpiler` set.
common --@aspect_rules_ts//ts:default_to_tsc_transpiler

# optimized LTO build. you'll need a fairly recent clang for this to work
build:thin-lto -c opt
build:thin-lto --cxxopt='-flto=thin'
build:thin-lto --linkopt='-flto=thin'

# configuration used for performance profiling
build:profile --config=thin-lto
build:profile --copt="-fno-omit-frame-pointer" --copt="-mno-omit-leaf-frame-pointer"
build:profile --config=limited-dbg-info
build:profile --strip=never

# configuration used for performance benchmarking is the same as profiling for consistency
build:benchmark --config=profile

# Define a debug config which is primarily intended for local development.
build:debug -c dbg

# Using simple template names saves around 5% of binary size of workerd.
build:unix --cxxopt='-gsimple-template-names' --host_cxxopt='-gsimple-template-names'
# Enable hidden visibility for inline functions. This can cause problems when comparing pointers to
# inline functions across shared library boundaries, but this is unlikely to be done anywhere
# within workerd, V8 explicitly supports hidden visibility.
build:unix --cxxopt='-fvisibility-inlines-hidden' --host_cxxopt='-fvisibility-inlines-hidden'

# Define a config mode which is fastbuild but with basic debug info.
build:fastdbg -c fastbuild
build:fastdbg --config=limited-dbg-info
build:fastdbg --config=rust-debug
build:fastdbg --strip=never
build:fastdbg --//:dead_strip=False

# provide a limited amount of debug info, sufficient for qualified stack traces.
build:limited-dbg-info --copt='-g1' --copt="-fdebug-info-for-profiling"

# Miscellaneous platform-independent options
build --@capnp-cpp//src/kj:openssl=True --@capnp-cpp//src/kj:zlib=True --@capnp-cpp//src/kj:brotli=True
build --cxxopt="-fbracket-depth=512" --host_cxxopt="-fbracket-depth=512"

# Additional Rust flags (see https://doc.rust-lang.org/rustc/codegen-options/index.html)
# Need to disable debug-assertions for fastbuild, should be off automatically for opt. As long as
# rust-level LTO is not enabled, LLVM bitcode is not needed so -C embed-bitcode=n provides a free
# improvement to build speed and rust-deps code size.
# Using extra_rustc_flag for non-exec flags so they can be overwritten selectively.
build --@rules_rust//:extra_rustc_flag=-Cembed-bitcode=n
build --@rules_rust//:extra_rustc_flag=-Cpanic=abort
build --@rules_rust//:extra_rustc_flag=-Cdebug-assertions=n
build --@rules_rust//:extra_exec_rustc_flags=-C,panic=abort,-C,debug-assertions=n,-C,embed-bitcode=n

# We default to not enabling debug assertions and unwinding for Rust. For debug builds this is not
# the right setting, but unfortunately we can't set that directly.
build:rust-debug --@rules_rust//:extra_rustc_flag=-Cpanic=unwind
build:rust-debug --@rules_rust//:extra_rustc_flag=-Cdebug-assertions=y
# Use the equivalent of -g/-g2 for Rust here, this is necessary to get qualified stack traces while
# -Zdebug-info-for-profiling is unavailable. Unlike gcc/clang, Rust defaults to -g3 otherwise.
build:rust-debug --@rules_rust//:extra_rustc_flag=-Cdebuginfo=1
build:rust-debug --@rules_rust//:extra_rustc_flag=-Cstrip=none

# Adding -C lto=thin here would improve binary size significantly – disable it for now due to
# compile errors and wrong code generation when bazel and rust use different LLVM versions.
build:thin-lto --@rules_rust//:extra_rustc_flag=-Ccodegen-units=1

# common sanitizers options
build:sanitizer-common --@workerd//src/workerd/server:use_tcmalloc=False
build:sanitizer-common --copt="-fsanitize-link-c++-runtime" --linkopt="-fsanitize-link-c++-runtime"
build:sanitizer-common --copt="-Og"
build:sanitizer-common --copt="-g" --strip=never
build:sanitizer-common --copt="-fno-optimize-sibling-calls"
build:sanitizer-common --copt="-fno-omit-frame-pointer" --copt="-mno-omit-leaf-frame-pointer"

# address sanitizer (https://github.com/google/sanitizers/wiki/AddressSanitizer)
build:asan --config=sanitizer-common
build:asan --copt="-fsanitize=address" --linkopt="-fsanitize=address"
build:asan --test_env=ASAN_OPTIONS=abort_on_error=true
build:asan --test_env=KJ_CLEAN_SHUTDOWN=1

#
# Linux and macOS
#
build:unix --workspace_status_command=./tools/unix/workspace-status.sh

build:unix --cxxopt='-std=c++20' --host_cxxopt='-std=c++20'
build:unix --@capnp-cpp//src/kj:libdl=True

build:unix --action_env=BAZEL_COMPILER=clang
build:unix --action_env=CC=clang
build:unix --action_env=CXX=clang++

build:unix --test_env=LLVM_SYMBOLIZER=llvm-symbolizer

# Warning options.
build:unix --cxxopt='-Wall' --host_cxxopt='-Wall'
build:unix --cxxopt='-Wextra' --host_cxxopt='-Wextra'
build:unix --cxxopt='-Wunused-function' --host_cxxopt='-Wunused-function'
build:unix --cxxopt='-Wunused-lambda-capture' --host_cxxopt='-Wunused-lambda-capture'
build:unix --cxxopt='-Wunused-variable' --host_cxxopt='-Wunused-variable'
build:unix --cxxopt='-Wno-strict-aliasing' --host_cxxopt='-Wno-strict-aliasing'
build:unix --cxxopt='-Wno-sign-compare' --host_cxxopt='-Wno-sign-compare'
build:unix --cxxopt='-Wno-unused-parameter' --host_cxxopt='-Wno-unused-parameter'
build:unix --cxxopt='-Wno-missing-field-initializers' --host_cxxopt='-Wno-missing-field-initializers'
build:unix --cxxopt='-Wno-ignored-qualifiers' --host_cxxopt='-Wno-ignored-qualifiers'

build:linux --config=unix
build:macos --config=unix

# Support macOS 11 as the minimum version. There should be at least a warning when backward
# compatibility is broken as -Wunguarded-availability-new is enabled by default.
# TODO (cleanup): macOS 11.X is now EOL, bump this to macOS 12.5 soon.
build:macos --macos_minimum_os=11.5 --host_macos_minimum_os=11.5

# Avoid emitting duplicate unwind info where compact unwind info can be used. This reduces the
# object size by ~5% on average, improving disk space usage and link times. The final binary size
# is not affected. Note that this is on-by-default on arm64, but turning it on for all mac builds
# is easier than adding the flag only on x86. See https://reviews.llvm.org/D122258 for detailed
# information on the flag.
build:macos --copt='-femit-dwarf-unwind=no-compact-unwind'

# Bazel doesn't allow setting flags only for a specific build mode like fastbuild, but somehow has
# a flag that does this for ObjC to set "-O0 -DDEBUG=1". Worse yet, options set by this flag cannot
# be overwritten on the command line, which confusingly makes enabling optimization there not have
# an effect. On macOS we build some ObjC as part of Dawn, set this flag to an empty string so we
# can set the right configuration flags.
# TODO(soon): File a bazel issue for the flag so we can drop this.
build:macos --experimental_objc_fastbuild_options=''

# Unfortunately experimental_objc_fastbuild_options does not respect options set before and
# overwrites them. This causes mysterious binary size bloat for fastbuild-based configurations on
# macOS when enabling some optimizations. Turn it in into a no-op.
build:mac --experimental_objc_fastbuild_options=""

# On Linux, always link libc++ statically to avoid compatibility issues with different OS versions.
# macOS links with dynamic libc++ by default, which has good backwards compatibility.
# Drop default link flags, which include libstdc++ for Linux
build:linux --features=-default_link_libs --host_features=-default_link_libs
build:linux --cxxopt='-stdlib=libc++' --host_cxxopt='-stdlib=libc++'
build:linux --linkopt='-stdlib=libc++' --host_linkopt='-stdlib=libc++'
build:linux --linkopt='-l:libc++.a' --linkopt='-lm' --linkopt='-static-libgcc'
# We don't expect to distribute host tools, no need to statically link libgcc.
# TODO(cleanup): Ideally we'd also use shared libc++ here, we'd just need to install the
# libunwind-15-dev and libc++abi-15-dev packages on CI to have all of shared libc++ available.
build:linux --host_linkopt='-l:libc++.a' --host_linkopt='-lm'

# On Linux, enable PIC. In macos pic is the default, and the objc_library rule does not work
# correctly if we use this flag since it will not find the object files to include
# https://github.com/bazelbuild/bazel/issues/12439#issuecomment-914449079
build:linux --force_pic

# On Linux, garbage collection sections and optimize binary size. These do not apply to the macOS
# toolchain.
build:linux --linkopt="-Wl,--gc-sections"
build:linux --copt="-ffunction-sections" --host_copt="-ffunction-sections"
build:linux --copt="-fdata-sections" --host_copt="-fdata-sections"

#
# Windows
#

# See https://bazel.build/configure/windows#symlink
startup --windows_enable_symlinks
build:windows --workspace_status_command=./tools/windows/workspace-status.cmd
build:windows --enable_runfiles
# We use LLVM's MSVC-compatible compiler driver to compile our code on Windows,
# as opposed to using MSVC directly. This enables us to use the "same" compiler
# frontend on Linux, macOS, and Windows, massively reducing the effort required
# to compile workerd on Windows. Notably, this provides proper support for
# `#pragma once` when using symlinked virtual includes, `__atomic_*` functions,
# a standards-compliant preprocessor, support for GNU statement expressions
# used by some KJ macros, and understands the `.c++` extension by default.
# As of bazel 7, incompatible_enable_cc_toolchain_resolution is enabled by default, so we need to
# define a platform for the clang-cl build.
build:windows --extra_toolchains=@local_config_cc//:cc-toolchain-x64_windows-clang-cl
build:windows --extra_execution_platforms=//:x64_windows-clang-cl

# The Windows fastbuild bazel configuration is broken in that it necessarily generates PDB debug
# information while the Linux and macOS toolchains only compile with debug information in the dbg
# configuration or when requested with the -g flag. This causes huge increases in compile time and
# disk/cache space usage – a single test may come with a 490MB PDB file.
# In an optional configuration, use the opt configuration and manually disable optimizations as a
# workaround.

build:windows_no_dbg -c opt
build:windows_no_dbg --copt='-O0'
build:windows_no_dbg --copt='/Od'
build:windows_no_dbg --linkopt='/INCREMENTAL:NO'
build:windows_no_dbg --features=-smaller_binary

# Mitigate the large size impact of Windows debug binaries somewhat by applying string merging and
# linker garbage collection.
build:windows_dbg --config=debug
build:windows_dbg --copt='/Gy' --copt='/Gw'
build:windows_dbg --linkopt='/OPT:REF'
build:windows_dbg --linkopt='/OPT:LLDTAILMERGE' --linkopt='/OPT:SAFEICF'

# This hides inline symbols in classes that are marked to be exported, similar to
# -fvisibility-inlines-hidden on Unix systems (https://blog.llvm.org/2018/11/30-faster-windows-builds-with-clang-cl_14.html)
# Currently this only reduces object sizes slightly, larger gains are possible if we compile V8 as
# a shared library.
build:windows --copt='/Zc:dllexportInlines-'

# Release configuration.
build:release -c opt
build:release --@rules_rust//:extra_rustc_flag=-Ccodegen-units=1

# enable -O3 for the release configuration. Based on benchmarking there is little difference in
# performance, but -O3 should generally be expected to be faster for at least parts of the workerd API.
build:release_unix --copt='-O3'
build:release_unix --config=release

build:release_linux --config=release_unix
build:release_linux --linkopt="-Wl,-O2"

build:release_macos --config=release_unix
# Disable generating LC_DATA_IN_CODE and LC_FUNCTION_STARTS binary sections, two rarely used types
# of macOS debug info. These sections are largely undocumented, but are used by LLDB to improve
# debugging on binaries that are otherwise stripped. There should be no need to include this
# data in releases.
build:release_macos --linkopt="-Wl,-no_data_in_code_info"
build:release_macos --linkopt="-Wl,-no_function_starts"

build:release_windows --config=release
# Windows uses /O2 as its preferred optimization setting and enabled by bazel in the opt
# configuration, but for clang-cl this is equivalent to only -O2 and a few other things. -O3 is
# not exposed directly in the clang-cl driver, but we can access regular clang
# flags using the /clang prefix anyway. https://clang.llvm.org/docs/UsersManual.html#the-clang-option
build:release_windows --copt="/clang:-O3"
# clang-cl does not enable strict aliasing by default to match MSVC's approach, unlike clang on
# Unix which turns it on for opt builds.
build:release_windows --copt="-fstrict-aliasing"

build:windows --cxxopt='/std:c++20' --host_cxxopt='/std:c++20'
build:windows --copt='/D_CRT_USE_BUILTIN_OFFSETOF' --host_copt='/D_CRT_USE_BUILTIN_OFFSETOF'
build:windows --copt='/DWIN32_LEAN_AND_MEAN' --host_copt='/DWIN32_LEAN_AND_MEAN'
# The `/std:c++20` argument is unused during boringssl compilation and we don't
# want a warning when compiling each file.
build:windows --copt='-Wno-unused-command-line-argument' --host_copt='-Wno-unused-command-line-argument'

# MSVC disappointingly sets __cplusplus to 199711L by default. Defining /Zc:__cplusplus makes it
# set the correct value. We currently don't check __cplusplus, but some dependencies do.
build:windows --cxxopt='/Zc:__cplusplus' --host_cxxopt='/Zc:__cplusplus'

# enable clang coverage: https://clang.llvm.org/docs/SourceBasedCodeCoverage.html
build:clang-coverage --copt="-fprofile-instr-generate" --linkopt="-fprofile-instr-generate"
build:clang-coverage --copt="-fcoverage-mapping" --linkopt="-fcoverage-mapping"