native_toolchain.py

# Copyright 2018 Pants project contributors (see CONTRIBUTORS.md).
# Licensed under the Apache License, Version 2.0 (see LICENSE).

from dataclasses import dataclass
from typing import Any, Tuple, Union

from pants.backend.native.config.environment import (
    Assembler,
    CCompiler,
    CppCompiler,
    CppToolchain,
    CToolchain,
    Linker,
)
from pants.backend.native.subsystems.binaries.binutils import Binutils
from pants.backend.native.subsystems.binaries.gcc import GCC
from pants.backend.native.subsystems.binaries.llvm import LLVM
from pants.backend.native.subsystems.libc_dev import LibcDev
from pants.backend.native.subsystems.native_build_step import ToolchainVariant
from pants.backend.native.subsystems.xcode_cli_tools import XCodeCLITools
from pants.engine.platform import Platform
from pants.engine.rules import RootRule, rule
from pants.engine.selectors import Get
from pants.subsystem.subsystem import Subsystem
from pants.util.enums import match
from pants.util.memo import memoized_property


class NativeToolchain(Subsystem):
    """Abstraction over platform-specific tools to compile and link native code.

    When this subsystem is consumed, Pants will download and unpack archives (if necessary) which
    together provide an appropriate "native toolchain" for the host platform: a compiler and linker,
    usually. This subsystem exposes the toolchain through `@rule`s, which tasks then request during
    setup or execution (synchronously, for now).

    NB: Currently, on OSX, Pants will find and invoke the XCode command-line tools, or error out with
    installation instructions if the XCode tools could not be found.
    """

    options_scope = "native-toolchain"

    @classmethod
    def subsystem_dependencies(cls):
        return super().subsystem_dependencies() + (
            Binutils.scoped(cls),
            GCC.scoped(cls),
            LibcDev.scoped(cls),
            LLVM.scoped(cls),
            XCodeCLITools.scoped(cls),
        )

    @memoized_property
    def _binutils(self):
        return Binutils.scoped_instance(self)

    @memoized_property
    def _gcc(self):
        return GCC.scoped_instance(self)

    @memoized_property
    def _llvm(self):
        return LLVM.scoped_instance(self)

    @memoized_property
    def _xcode_cli_tools(self):
        return XCodeCLITools.scoped_instance(self)

    @memoized_property
    def _libc_dev(self):
        return LibcDev.scoped_instance(self)


@dataclass(frozen=True)
class LibcObjects:
    crti_object_paths: Any


class LinkerWrapperMixin:
    def for_compiler(self, compiler):
        """Return a Linker object which is intended to be compatible with the given `compiler`."""
        return (
            self.linker
            # TODO(#6143): describe why the compiler needs to be first on the PATH!
            .sequence(compiler, exclude_list_fields=["extra_args", "path_entries"])
            .prepend_field("path_entries", compiler.path_entries)
            .copy(exe_filename=compiler.exe_filename)
        )


@dataclass(frozen=True)
class GCCLinker(LinkerWrapperMixin):
    linker: Linker


@dataclass(frozen=True)
class LLVMLinker(LinkerWrapperMixin):
    linker: Linker


@dataclass(frozen=True)
class GCCCToolchain:
    c_toolchain: CToolchain


@dataclass(frozen=True)
class GCCCppToolchain:
    cpp_toolchain: CppToolchain


@dataclass(frozen=True)
class LLVMCToolchain:
    c_toolchain: CToolchain


@dataclass(frozen=True)
class LLVMCppToolchain:
    cpp_toolchain: CppToolchain


@rule
async def select_libc_objects(platform: Platform, native_toolchain: NativeToolchain) -> LibcObjects:
    # We use lambdas here to avoid searching for libc on osx, where it will fail.
    paths = match(
        platform,
        {
            Platform.darwin: lambda: [],
            Platform.linux: lambda: native_toolchain._libc_dev.get_libc_objects(),
        },
    )()
    return LibcObjects(paths)


@rule
async def select_assembler(platform: Platform, native_toolchain: NativeToolchain) -> Assembler:
    if platform == Platform.darwin:
        assembler = await Get(Assembler, XCodeCLITools, native_toolchain._xcode_cli_tools)
    else:
        assembler = await Get(Assembler, Binutils, native_toolchain._binutils)
    return assembler


@dataclass(frozen=True)
class BaseLinker:
    """A Linker which is not specific to any compiler yet.

    This represents Linker objects provided by subsystems, but may need additional information to be
    usable by a specific compiler.
    """

    linker: Linker


# TODO: select the appropriate `Platform` in the `@rule` decl using variants!
@rule
async def select_base_linker(platform: Platform, native_toolchain: NativeToolchain) -> BaseLinker:
    if platform == Platform.darwin:
        # TODO(#5663): turn this into LLVM when lld works.
        linker = await Get(Linker, XCodeCLITools, native_toolchain._xcode_cli_tools)
    else:
        linker = await Get(Linker, Binutils, native_toolchain._binutils)
    base_linker = BaseLinker(linker=linker)
    return base_linker


@rule
async def select_gcc_linker(native_toolchain: NativeToolchain) -> GCCLinker:
    base_linker = await Get(BaseLinker, NativeToolchain, native_toolchain)
    linker = base_linker.linker
    libc_objects = await Get(LibcObjects, NativeToolchain, native_toolchain)
    linker_with_libc = linker.append_field("extra_object_files", libc_objects.crti_object_paths)
    return GCCLinker(linker_with_libc)


@rule
def select_llvm_linker(base_linker: BaseLinker) -> LLVMLinker:
    return LLVMLinker(base_linker.linker)


@dataclass(frozen=True)
class GCCInstallLocationForLLVM:
    """This class is convertible into a list of command line arguments for clang and clang++.

    This is only used on Linux. The option --gcc-toolchain stops clang from searching for another
    gcc on the host system. The option appears to only exist on Linux clang and clang++.
    """

    toolchain_dir: Any

    @property
    def as_clang_argv(self):
        # TODO(#6143): describe exactly what this argument does to the clang/clang++ invocation!
        return [f"--gcc-toolchain={self.toolchain_dir}"]


@rule
def select_gcc_install_location(gcc: GCC) -> GCCInstallLocationForLLVM:
    return GCCInstallLocationForLLVM(gcc.select())


@rule
async def select_llvm_c_toolchain(
    platform: Platform, native_toolchain: NativeToolchain
) -> LLVMCToolchain:
    provided_clang = await Get(CCompiler, LLVM, native_toolchain._llvm)

    if platform == Platform.darwin:
        xcode_clang = await Get(CCompiler, XCodeCLITools, native_toolchain._xcode_cli_tools)
        joined_c_compiler = provided_clang.sequence(xcode_clang)
    else:
        gcc_install = await Get(GCCInstallLocationForLLVM, GCC, native_toolchain._gcc)
        provided_gcc = await Get(CCompiler, GCC, native_toolchain._gcc)
        joined_c_compiler = (
            provided_clang.sequence(provided_gcc).append_field(
                "extra_args", gcc_install.as_clang_argv
            )
            # We need g++'s version of the GLIBCXX library to be able to run.
            .prepend_field("runtime_library_dirs", provided_gcc.runtime_library_dirs)
        )

    working_c_compiler = joined_c_compiler.prepend_field("extra_args", ["-x", "c", "-std=c11"])

    llvm_linker_wrapper = await Get(LLVMLinker, NativeToolchain, native_toolchain)
    working_linker = llvm_linker_wrapper.for_compiler(working_c_compiler)

    return LLVMCToolchain(CToolchain(working_c_compiler, working_linker))


@rule
async def select_llvm_cpp_toolchain(
    platform: Platform, native_toolchain: NativeToolchain
) -> LLVMCppToolchain:
    provided_clangpp = await Get(CppCompiler, LLVM, native_toolchain._llvm)

    # On OSX, we use the libc++ (LLVM) C++ standard library implementation. This is feature-complete
    # for OSX, but not for Linux (see https://libcxx.llvm.org/ for more info).
    if platform == Platform.darwin:
        xcode_clangpp = await Get(CppCompiler, XCodeCLITools, native_toolchain._xcode_cli_tools)
        joined_cpp_compiler = provided_clangpp.sequence(xcode_clangpp)
        extra_llvm_linking_library_dirs: Tuple[str, ...] = tuple()
        linker_extra_args: Tuple[str, ...] = tuple()
    else:
        gcc_install = await Get(GCCInstallLocationForLLVM, GCC, native_toolchain._gcc)
        provided_gpp = await Get(CppCompiler, GCC, native_toolchain._gcc)
        joined_cpp_compiler = (
            provided_clangpp.sequence(provided_gpp)
            # NB: we use g++'s headers on Linux, and therefore their C++ standard
            # library.
            .copy(include_dirs=provided_gpp.include_dirs).append_field(
                "extra_args", gcc_install.as_clang_argv
            )
            # We need g++'s version of the GLIBCXX library to be able to run.
            .prepend_field("runtime_library_dirs", provided_gpp.runtime_library_dirs)
        )
        extra_llvm_linking_library_dirs = (
            provided_gpp.runtime_library_dirs + provided_clangpp.runtime_library_dirs
        )
        # Ensure we use libstdc++, provided by g++, during the linking stage.
        linker_extra_args = ("-stdlib=libstdc++",)

    working_cpp_compiler = joined_cpp_compiler.prepend_field(
        "extra_args",
        [
            "-x",
            "c++",
            "-std=c++11",
            # This flag is intended to avoid using any of the headers from our LLVM distribution's C++
            # stdlib implementation, or any from the host system, and instead, use include dirs from the
            # XCodeCLITools or GCC.
            # TODO(#6143): Determine precisely what this flag does and why it's necessary.
            "-nostdinc++",
        ],
    )

    llvm_linker_wrapper = await Get(LLVMLinker, NativeToolchain, native_toolchain)
    working_linker = (
        llvm_linker_wrapper.for_compiler(working_cpp_compiler)
        .append_field("linking_library_dirs", extra_llvm_linking_library_dirs)
        .prepend_field("extra_args", linker_extra_args)
    )

    return LLVMCppToolchain(CppToolchain(working_cpp_compiler, working_linker))


@rule
async def select_gcc_c_toolchain(
    platform: Platform, native_toolchain: NativeToolchain
) -> GCCCToolchain:
    provided_gcc = await Get(CCompiler, GCC, native_toolchain._gcc)

    if platform == Platform.darwin:
        # GCC needs access to some headers that are only provided by the XCode toolchain
        # currently (e.g. "_stdio.h"). These headers are unlikely to change across versions, so this is
        # probably safe.
        xcode_clang = await Get(CCompiler, XCodeCLITools, native_toolchain._xcode_cli_tools)
        joined_c_compiler = provided_gcc.sequence(xcode_clang)
    else:
        joined_c_compiler = provided_gcc

    # GCC needs an assembler, so we provide that (platform-specific) tool here.
    assembler = await Get(Assembler, NativeToolchain, native_toolchain)
    working_c_compiler = joined_c_compiler.sequence(assembler).prepend_field(
        "extra_args", ["-x", "c", "-std=c11"]
    )

    gcc_linker_wrapper = await Get(GCCLinker, NativeToolchain, native_toolchain)
    working_linker = gcc_linker_wrapper.for_compiler(working_c_compiler)

    return GCCCToolchain(CToolchain(working_c_compiler, working_linker))


@rule
async def select_gcc_cpp_toolchain(
    platform: Platform, native_toolchain: NativeToolchain
) -> GCCCppToolchain:
    provided_gpp = await Get(CppCompiler, GCC, native_toolchain._gcc)

    if platform == Platform.darwin:
        # GCC needs access to some headers that are only provided by the XCode toolchain
        # currently (e.g. "_stdio.h"). These headers are unlikely to change across versions, so this is
        # probably safe.
        # TODO: we should be providing all of these (so we can eventually phase out XCodeCLITools
        # entirely).
        xcode_clangpp = await Get(CppCompiler, XCodeCLITools, native_toolchain._xcode_cli_tools)
        joined_cpp_compiler = provided_gpp.sequence(xcode_clangpp)
    else:
        joined_cpp_compiler = provided_gpp

    # GCC needs an assembler, so we provide that (platform-specific) tool here.
    assembler = await Get(Assembler, NativeToolchain, native_toolchain)
    working_cpp_compiler = joined_cpp_compiler.sequence(assembler).prepend_field(
        "extra_args",
        [
            "-x",
            "c++",
            "-std=c++11",
            # This flag is intended to avoid using any of the headers from our LLVM distribution's C++
            # stdlib implementation, or any from the host system, and instead, use include dirs from the
            # XCodeCLITools or GCC.
            # TODO(#6143): Determine precisely what this flag does and why it's necessary.
            "-nostdinc++",
        ],
    )

    gcc_linker_wrapper = await Get(GCCLinker, NativeToolchain, native_toolchain)
    working_linker = gcc_linker_wrapper.for_compiler(working_cpp_compiler)

    return GCCCppToolchain(CppToolchain(working_cpp_compiler, working_linker))


@dataclass(frozen=True)
class ToolchainVariantRequest:
    toolchain: NativeToolchain
    variant: ToolchainVariant


@rule
async def select_c_toolchain(toolchain_variant_request: ToolchainVariantRequest) -> CToolchain:
    use_gcc = toolchain_variant_request.variant == ToolchainVariant.gnu
    toolchain_resolved: Union[GCCCToolchain, LLVMCToolchain]
    if use_gcc:
        toolchain_resolved = await Get(
            GCCCToolchain, NativeToolchain, toolchain_variant_request.toolchain
        )
    else:
        toolchain_resolved = await Get(
            LLVMCToolchain, NativeToolchain, toolchain_variant_request.toolchain
        )
    return toolchain_resolved.c_toolchain


@rule
async def select_cpp_toolchain(toolchain_variant_request: ToolchainVariantRequest) -> CppToolchain:
    use_gcc = toolchain_variant_request.variant == ToolchainVariant.gnu
    toolchain_resolved: Union[GCCCppToolchain, LLVMCppToolchain]
    if use_gcc:
        toolchain_resolved = await Get(
            GCCCppToolchain, NativeToolchain, toolchain_variant_request.toolchain
        )
    else:
        toolchain_resolved = await Get(
            LLVMCppToolchain, NativeToolchain, toolchain_variant_request.toolchain
        )
    return toolchain_resolved.cpp_toolchain


def create_native_toolchain_rules():
    return [
        select_libc_objects,
        select_assembler,
        select_base_linker,
        select_gcc_linker,
        select_llvm_linker,
        select_gcc_install_location,
        select_llvm_c_toolchain,
        select_llvm_cpp_toolchain,
        select_gcc_c_toolchain,
        select_gcc_cpp_toolchain,
        select_c_toolchain,
        select_cpp_toolchain,
        RootRule(NativeToolchain),
        RootRule(ToolchainVariantRequest),
    ]