Configuring C++ toolchains

Overview

This tutorial uses an example scenario to describe how to configure C++ toolchains for a project. It’s based on an example C++ project that builds error-free using gcc, clang, and msvc.

In this tutorial, you will create a Starlark rule that provides additional configuration for the cc_toolchain so that Bazel can build the application with emscripten. The expected outcome is to run bazel build --config=asmjs //main:helloworld.js on a Linux machine and build the C++ application using emscripten targeting asm.js.

Setting up the build environment

This tutorial assumes you are on Linux on which you have successfully built C++ applications - in other words, we assume that appropriate tooling and libraries have been installed.

Set up your build environment as follows:

  1. If you have not already done so, download and install Bazel 0.23 or later.

  2. Download the example C++ project from GitHub and place it in an empty directory on your local machine.

  3. Add the following cc_binary target to the main/BUILD file:

    cc_binary(
    name = "helloworld.js",
    srcs = ["hello-world.cc"],
)

  4. Create a .bazelrc file at the root of the workspace directory with the following contents to enable the use of the --config flag:

    # Use our custom-configured c++ toolchain.

build:asmjs --crosstool_top=//toolchain:emscripten

# Use --cpu as a differentiator.

build:asmjs --cpu=asmjs

# Use the default Bazel C++ toolchain to build the tools used during the
# build.

build:asmjs --host_crosstool_top=@bazel_tools//tools/cpp:toolchain

In this example, we are using the --cpu flag as a differentiator, since emscripten can target both asmjs and Web assembly. We are not configuring a Web assembly toolchain, however. Since Bazel uses many internal tools written in C++, such as process-wrapper, we are specifying a “sane” C++ toolchain for the host platform.

Configuring the C++ toolchain

To configure the C++ toolchain, repeatedly build the application and eliminate each error one by one as described below.

Note: This tutorial assumes you’re using Bazel 0.23 or later. If you’re using an older release of Bazel, look for the “Configuring CROSSTOOL” tutorial.

  1. Run the build with the following command:

    bazel build --config=asmjs //main:helloworld.js

    Because you specified --crosstool_top=//toolchain:emscripten in the .bazelrc file, Bazel throws the following error:

    No such package `toolchain`: BUILD file not found on package path.

    In the workspace directory, create the toolchain directory for the package and an empty BUILD file inside the toolchain directory.

  2. Run the build again. Because the toolchain package does not yet define the emscripten target, Bazel throws the following error:

    No such target '//toolchain:emscripten': target 'emscripten' not declared in
package 'toolchain' defined by .../toolchain/BUILD

    In the toolchain/BUILD file, define an empty filegroup as follows:

    package(default_visibility = ['//visibility:public'])

filegroup(name = "emscripten")

  3. Run the build again. Bazel throws the following error:

    '//toolchain:emscripten' does not have mandatory providers: 'ToolchainInfo'

    Bazel discovered that the --crosstool_top flag points to a rule that doesn’t provide the necessary ToolchainInfo provider. So we need to point --crosstool_top to a rule that does provide ToolchainInfo - that is the cc_toolchain_suite rule. In the toolchain/BUILD file, replace the empty filegroup with the following:

    cc_toolchain_suite(
    name = "emscripten",
    toolchains = {
        "asmjs": ":asmjs_toolchain",
    },
)

    The toolchains attribute automatically maps the --cpu (and also --compiler when specified) values to cc_toolchain. You have not yet defined any cc_toolchain targets and Bazel will complain about that shortly.

  4. Run the build again. Bazel throws the following error:

    Rule '//toolchain:asmjs_toolchain' does not exist

    Now you need to define cc_toolchain targets for every value in the cc_toolchain_suite.toolchains attribute. This is where you specify the files that comprise the toolchain so that Bazel can set up sandboxing. Add the following to the toolchain/BUILD file:

    filegroup(name = "empty")

cc_toolchain(
    name = "asmjs_toolchain",
    toolchain_identifier = "asmjs-toolchain",
    toolchain_config = ":asmjs_toolchain_config",
    all_files = ":empty",
    compiler_files = ":empty",
    dwp_files = ":empty",
    linker_files = ":empty",
    objcopy_files = ":empty",
    strip_files = ":empty",
    supports_param_files = 0,
)

  5. Run the build again. Bazel throws the following error:

    Rule '//toolchain:asmjs-toolchain' does not exist

    Let’s add a “:asmjs_toolchain_config” target to the toolchain/BUILD file:

    filegroup(name = "asmjs_toolchain_config")

  6. Run the build again. Bazel throws the following error:

    '//toolchain:asmjs_toolchain_config' does not have mandatory providers:
'CcToolchainConfigInfo'

    CcToolchainConfigInfo is a provider that we use to configure our C++ toolchains. We are going to create a Starlark rule that will provide CcToolchainConfigInfo. Create a toolchain/cc_toolchain_config.bzl file with the following content:

    def _impl(ctx):
    return cc_common.create_cc_toolchain_config_info(
        ctx = ctx,
        toolchain_identifier = "asmjs-toolchain",
        host_system_name = "i686-unknown-linux-gnu",
        target_system_name = "asmjs-unknown-emscripten",
        target_cpu = "asmjs",
        target_libc = "unknown",
        compiler = "emscripten",
        abi_version = "unknown",
        abi_libc_version = "unknown",
    )

cc_toolchain_config = rule(
    implementation = _impl,
    attrs = {},
    provides = [CcToolchainConfigInfo],
)

    cc_common.create_cc_toolchain_config_info() creates the needed provider CcToolchainConfigInfo. Now let’s declare a rule that will make use of the newly implemented cc_toolchain_config rule. Add a load statement to toolchains/BUILD:

    load(":cc_toolchain_config.bzl", "cc_toolchain_config")

    And replace the “asmjs_toolchain_config” filegroup with a declaration of a cc_toolchain_config rule:

    cc_toolchain_config(name = "asmjs_toolchain_config")

  7. Run the build again. Bazel throws the following error:

    .../BUILD:1:1: C++ compilation of rule '//:helloworld.js' failed (Exit 1)
src/main/tools/linux-sandbox-pid1.cc:421:
"execvp(toolchain/DUMMY_GCC_TOOL, 0x11f20e0)": No such file or directory
Target //:helloworld.js failed to build`

    At this point, Bazel has enough information to attempt building the code but it still does not know what tools to use to complete the required build actions. We will modify our Starlark rule implementation to tell Bazel what tools to use. For that, we’ll need the tool_path() constructor from @bazel_tools//tools/cpp:cc_toolchain_config_lib.bzl:

    # toolchain/cc_toolchain_config.bzl:
load("@bazel_tools//tools/cpp:cc_toolchain_config_lib.bzl", "tool_path")

def _impl(ctx):
    tool_paths = [
        tool_path(
            name = "gcc",
            path = "emcc.sh",
        ),
        tool_path(
            name = "ld",
            path = "emcc.sh",
        ),
        tool_path(
            name = "ar",
            path = "/bin/false",
        ),
        tool_path(
            name = "cpp",
            path = "/bin/false",
        ),
        tool_path(
            name = "gcov",
            path = "/bin/false",
        ),
        tool_path(
            name = "nm",
            path = "/bin/false",
        ),
        tool_path(
            name = "objdump",
            path = "/bin/false",
        ),
        tool_path(
            name = "strip",
            path = "/bin/false",
        ),
    ]
    return cc_common.create_cc_toolchain_config_info(
        ctx = ctx,
        toolchain_identifier = "asmjs-toolchain",
        host_system_name = "i686-unknown-linux-gnu",
        target_system_name = "asmjs-unknown-emscripten",
        target_cpu = "asmjs",
        target_libc = "unknown",
        compiler = "emscripten",
        abi_version = "unknown",
        abi_libc_version = "unknown",
        tool_paths = tool_paths,
    )

    You may notice the emcc.sh wrapper script, which delegates to the external emcc.py file. Create the script in the toolchain package directory with the following contents and set its executable bit:

    #!/bin/bash
set -euo pipefail
python external/emscripten_toolchain/emcc.py "$@"

    Paths specified in the tool_paths list are relative to the package where the cc_toolchain_config target is specified.

    The emcc.py file does not yet exist in the workspace directory. To obtain it, you can either check the emscripten toolchain in with your project or pull it from its GitHub repository. This tutorial uses the latter approach. To pull the toolchain from the GitHub repository, add the following http_archive repository definitions to your WORKSPACE file:

    load("@bazel_tools//tools/build_defs/repo:http.bzl", "http_archive")

http_archive(
    name = 'emscripten_toolchain',
    url = 'https://github.com/kripken/emscripten/archive/1.37.22.tar.gz',
    build_file = '//:emscripten-toolchain.BUILD',
    strip_prefix = "emscripten-1.37.22",
)

http_archive(
    name = 'emscripten_clang',
    url = 'https://s3.amazonaws.com/mozilla-games/emscripten/packages/llvm/tag/linux_64bit/emscripten-llvm-e1.37.22.tar.gz',
    build_file = '//:emscripten-clang.BUILD',
    strip_prefix = "emscripten-llvm-e1.37.22",
)

    In the workspace directory root, create the emscripten-toolchain.BUILD and emscripten-clang.BUILD files that expose these repositories as filegroups and establish their visibility across the build.

    In the workspace directory root, make sure that a BUILD file is present. If not, create an empty one.

    touch BUILD

    First create the emscripten-toolchain.BUILD file with the following contents:

    package(default_visibility = ['//visibility:public'])

filegroup(
    name = "all",
    srcs = glob(["**/*"]),
)

    Next, create the emscripten-clang.BUILD file with the following contents:

    package(default_visibility = ['//visibility:public'])`

filegroup(
    name = "all",
    srcs = glob(["**/*"]),
)

    You may notice that the targets simply parse all of the files contained in the archives pulled by the http_archive repository rules. In a real world scenario, you would likely want to be more selective and granular by only parsing the files needed by the build and splitting them by action, such as compilation, linking, and so on. For the sake of simplicity, this tutorial omits this step.

  8. Run the build again. Bazel throws the following error:

    "execvp(toolchain/emcc.sh, 0x12bd0e0)": No such file or directory

    You now need to make Bazel aware of the artifacts you added in the previous step. In particular, the emcc.sh script must also be explicitly listed as a dependency of the corresponding cc_toolchain rule. Modify the toolchain/BUILD file to look as follows:

    package(default_visibility = ["//visibility:public"])

load(":cc_toolchain_config.bzl", "cc_toolchain_config")

cc_toolchain_config(name = "asmjs_toolchain_config")

cc_toolchain_suite(
    name = "emscripten",
    toolchains = {
        "asmjs": ":asmjs_toolchain",
    },
)

filegroup(
    name = "all",
    srcs = [
        "emcc.sh",
        "@emscripten_clang//:all",
        "@emscripten_toolchain//:all",
    ],
)

cc_toolchain(
    name = "asmjs_toolchain",
    toolchain_identifier = "asmjs-toolchain",
    toolchain_config = ":asmjs_toolchain_config",
    all_files = ":all",
    compiler_files = ":all",
    dwp_files = ":empty",
    linker_files = ":all",
    objcopy_files = ":empty",
    strip_files = ":empty",
    supports_param_files = 0,
)

    Congratulations! You are now using the emscripten toolchain to build your C++ sample code. The next steps are optional but are included for completeness.

  9. (Optional) Run the build again. Bazel throws the following error:

     ERROR: .../BUILD:1:1: C++ compilation of rule '//:helloworld.js' failed (Exit 1)

    The next step is to make the toolchain deterministic and hermetic - that is, limit it to only touch files it’s supposed to touch and ensure it doesn’t write temporary data outside the sandbox.

    You also need to ensure the toolchain does not assume the existence of your home directory with its configuration files and that it does not depend on unspecified environment variables.

    For our example project, make the following modifications to the toolchain/BUILD file:

     filegroup(
     name = "all",
     srcs = [
         "emcc.sh",
         "@emscripten_toolchain//:all",
         "@emscripten_clang//:all",
         ":emscripten_cache_content",
     ],
 )

 filegroup(
     name = "emscripten_cache_content",
     srcs = glob(["emscripten_cache/**/*"]),
 )

    Since emscripten caches standard library files, you can save time by not compiling stdlib for every action and also prevent it from storing temporary data in random place, check in the precompiled bitcode files into the toolchain/emscript_cache directory. You can create them by calling the following from the emscripten_clang repository (or let emscripten create them in ~/.emscripten_cache):

     python embuilder.py build dlmalloc libcxx libc gl libcxxabi libcxx_noexcept wasm-libc

    Copy those files to toolchain/emscripten_cache. Also update the emcc.sh script to look as follows:

     #!/bin/bash

 set -euo pipefail

 export LLVM_ROOT='external/emscripten_clang'
 export EMSCRIPTEN_NATIVE_OPTIMIZER='external/emscripten_clang/optimizer'
 export BINARYEN_ROOT='external/emscripten_clang/'
 export NODE_JS=''
 export EMSCRIPTEN_ROOT='external/emscripten_toolchain'
 export SPIDERMONKEY_ENGINE=''
 export EM_EXCLUSIVE_CACHE_ACCESS=1
 export EMCC_SKIP_SANITY_CHECK=1
 export EMCC_WASM_BACKEND=0

 mkdir -p "tmp/emscripten_cache"

 export EM_CACHE="tmp/emscripten_cache"
 export TEMP_DIR="tmp"

 # Prepare the cache content so emscripten doesn't keep rebuilding it
 cp -r toolchain/emscripten_cache/* tmp/emscripten_cache

 # Run emscripten to compile and link
 python external/emscripten_toolchain/emcc.py "$@"

 # Remove the first line of .d file
 find . -name "*.d" -exec sed -i '2d' {} \;

    Bazel can now properly compile the sample C++ code in hello-world.cc.

  10. (Optional) Run the build again. Bazel throws the following error:

    ..../BUILD:1:1: undeclared inclusion(s) in rule '//:helloworld.js':
this rule is missing dependency declarations for the following files included by 'helloworld.cc':
'.../external/emscripten_toolchain/system/include/libcxx/stdio.h'
'.../external/emscripten_toolchain/system/include/libcxx/__config'
'.../external/emscripten_toolchain/system/include/libc/stdio.h'
'.../external/emscripten_toolchain/system/include/libc/features.h'
'.../external/emscripten_toolchain/system/include/libc/bits/alltypes.h'

    At this point you have successfully compiled the example C++ code. The error above occurs because Bazel uses a .d file produced by the compiler to verify that all includes have been declared and to prune action inputs.

    In the .d file, Bazel discovered that our source code references system headers that have not been explicitly declared in the BUILD file. This in and of itself is not a problem and you can easily fix this by adding the target folders as -isystem directories. For this, you’ll need to add a feature to the CcToolchainConfigInfo. Modify toolchain/cc_toolchain_config.bzl to look like this:

    load("@bazel_tools//tools/build_defs/cc:action_names.bzl", "ACTION_NAMES")
load(
    "@bazel_tools//tools/cpp:cc_toolchain_config_lib.bzl",
    "feature",
    "flag_group",
    "flag_set",
    "tool_path",
)

def _impl(ctx):
    tool_paths = [
        tool_path(
            name = "gcc",
            path = "emcc.sh",
        ),
        tool_path(
            name = "ld",
            path = "emcc.sh",
        ),
        tool_path(
            name = "ar",
            path = "/bin/false",
        ),
        tool_path(
            name = "cpp",
            path = "/bin/false",
        ),
        tool_path(
            name = "gcov",
            path = "/bin/false",
        ),
        tool_path(
            name = "nm",
            path = "/bin/false",
        ),
        tool_path(
            name = "objdump",
            path = "/bin/false",
        ),
        tool_path(
            name = "strip",
            path = "/bin/false",
        ),
    ]
    toolchain_include_directories_feature = feature(
        name = "toolchain_include_directories",
        enabled = True,
        flag_sets = [
            flag_set(
                actions = [
                    ACTION_NAMES.assemble,
                    ACTION_NAMES.preprocess_assemble,
                    ACTION_NAMES.linkstamp_compile,
                    ACTION_NAMES.c_compile,
                    ACTION_NAMES.cpp_compile,
                    ACTION_NAMES.cpp_header_parsing,
                    ACTION_NAMES.cpp_module_compile,
                    ACTION_NAMES.cpp_module_codegen,
                    ACTION_NAMES.lto_backend,
                    ACTION_NAMES.clif_match,
                ],
                flag_groups = [
                    flag_group(
                        flags = [
                            "-isystem",
                            "external/emscripten_toolchain/system/include/libcxx",
                            "-isystem",
                            "external/emscripten_toolchain/system/include/libc",
                        ],
                    ),
                ],
            ),
        ],
    )

    return cc_common.create_cc_toolchain_config_info(
        ctx = ctx,
        toolchain_identifier = "asmjs-toolchain",
        host_system_name = "i686-unknown-linux-gnu",
        target_system_name = "asmjs-unknown-emscripten",
        target_cpu = "asmjs",
        target_libc = "unknown",
        compiler = "emscripten",
        abi_version = "unknown",
        abi_libc_version = "unknown",
        tool_paths = tool_paths,
        features = [toolchain_include_directories_feature],
    )

cc_toolchain_config = rule(
    implementation = _impl,
    attrs = {},
    provides = [CcToolchainConfigInfo],
)

  11. (Optional) Run the build again. With this final change, the build now completes error-free.