# General starlark utility functions
load("//bazel/platforms:normalize.bzl", "ARCH_NORMALIZE_MAP")
def write_target_impl(ctx):
out = ctx.actions.declare_file(ctx.label.name + ".gen_source_list")
ctx.actions.write(
out,
"//" + ctx.label.package + ":" + ctx.attr.target_name,
)
return [
DefaultInfo(
files = depset([out]),
),
]
write_target = rule(
write_target_impl,
attrs = {
"target_name": attr.string(
doc = "the name of the target to record",
),
},
)
def retry_download_and_extract(ctx, tries, **kwargs):
sleep_time = 1
for attempt in range(tries):
is_retriable = attempt + 1 < tries
result = ctx.download_and_extract(allow_fail = is_retriable, **kwargs)
if result.success:
return result
else:
print("Download failed (Attempt #%s), sleeping for %s seconds then retrying..." % (attempt + 1, sleep_time))
ctx.execute(["sleep", str(sleep_time)])
sleep_time *= 2
def retry_download(ctx, tries, **kwargs):
sleep_time = 1
for attempt in range(tries):
is_retriable = attempt + 1 < tries
result = ctx.download(allow_fail = is_retriable, **kwargs)
if result.success:
return result
else:
print("Download failed (Attempt #%s), sleeping for %s seconds then retrying..." % (attempt + 1, sleep_time))
ctx.execute(["sleep", str(sleep_time)])
sleep_time *= 2
def retry_execute(ctx, tries, arguments, **kwargs):
"""Runs a command with ctx.execute, retrying with exponential backoff.
Returns the exec_result of the last attempt; the caller is responsible
for checking return_code and failing with a useful message.
"""
sleep_time = 1
result = None
for attempt in range(tries):
result = ctx.execute(arguments, **kwargs)
if result.return_code == 0:
return result
if attempt + 1 < tries:
print("Command %s failed (Attempt #%s), sleeping for %s seconds then retrying..." % (arguments[0], attempt + 1, sleep_time))
ctx.execute(["sleep", str(sleep_time)])
sleep_time *= 2
return result
def write_python_pyc_cache_prefix_customization(ctx, customization_file, pycache_dirname = "bazel_pycache"):
"""Write a site/usercustomize module to redirect .pyc writes to /tmp.
Toolchain and runtime Python distributions often live under Bazel-managed paths
(external repositories, runfiles, etc.). If Python writes `__pycache__/*.pyc`
there, it can cause non-hermetic filesystem changes and Bazel cache churn.
This helper writes either `sitecustomize.py` or `usercustomize.py` (callers
choose the filename) to set `sys.pycache_prefix` to a temp directory.
"""
ctx.file(
customization_file,
"""
# Prevent bytecode cache writes under Bazel-managed paths (external/, runfiles/).
# Only `os` and `sys` are used: they are already imported at interpreter startup,
# so nothing new is imported (and cached) before the prefix takes effect. Importing
# tempfile here would pull in modules whose .pyc land in the Bazel-managed tree.
import os
import sys
_tmpdir = os.environ.get("TMPDIR") or os.environ.get("TEMP") or os.environ.get("TMP") or "/tmp"
sys.pycache_prefix = os.path.join(_tmpdir, "{pycache_dirname}")
""".format(pycache_dirname = pycache_dirname),
)
# A toolchain that deliberately matches nothing, for os/arch combinations where the
# mongo toolchain does not exist. It must not resolve, so that the platform's real
# toolchain wins instead.
_UNRESOLVABLE_TOOLCHAIN = """
toolchain(
name = "mongo_toolchain",
exec_compatible_with = ["@platforms//:incompatible"],
target_compatible_with = ["@platforms//:incompatible"],
toolchain = ":cc_mongo_toolchain",
toolchain_type = "@bazel_tools//tools/cpp:toolchain_type",
)
"""
# A toolchain that always matches, backed by the same hollow cc_toolchain. Used only
# when the toolchain download is deliberately skipped (no_c++_toolchain=1) for
# analysis-only invocations such as the resmoke target cquery in
# generate_result_tasks.py.
_RESOLVABLE_TOOLCHAIN = """
toolchain(
name = "mongo_toolchain",
toolchain = ":cc_mongo_toolchain",
toolchain_type = "@bazel_tools//tools/cpp:toolchain_type",
)
"""
# Minimal cc_toolchain_config_info. Declares no features and no real tools; it exists
# purely so toolchain resolution and cc_toolchain_alias analysis can succeed.
_NOOP_CC_TOOLCHAIN_CONFIG_BZL = """\
\"\"\"Hollow cc toolchain config for the no-op mongo toolchain.\"\"\"
load("@rules_cc//cc/common:cc_common.bzl", "cc_common")
load("@rules_cc//cc/toolchains:cc_toolchain_config_info.bzl", "CcToolchainConfigInfo")
def _impl(ctx):
return cc_common.create_cc_toolchain_config_info(
ctx = ctx,
toolchain_identifier = "mongo_noop_toolchain",
host_system_name = "local",
target_system_name = "local",
target_cpu = "unknown",
target_libc = "unknown",
compiler = "unknown",
)
noop_cc_toolchain_config = rule(
implementation = _impl,
attrs = {},
provides = [CcToolchainConfigInfo],
)
"""
def generate_noop_toolchain(ctx, substitutions, resolvable = False):
"""Generates a mongo toolchain repo with no downloaded payload.
Args:
ctx: the repository rule context.
substitutions: unused; accepted so callers can pass the usual substitution dict.
resolvable: when True, the stub toolchain carries no platform constraints so it
always resolves. Pass True only when the download was skipped on purpose
(no_c++_toolchain=1) for an analysis-only invocation. Leave False when the
mongo toolchain genuinely does not exist for this os/arch, so that the real
platform-specific toolchain (e.g. the apple one) is chosen instead of this stub.
"""
# BUILD file is required for a no-op.
# Keep a stub mongo_toolchain target so unconditional register_toolchains()
# calls don't fail when the toolchain is intentionally skipped/unsupported.
# Create a stub clang-format script so that targets referencing
# //:clang_format (e.g. the format_multirun rule) can still build on
# platforms where the mongo toolchain is unavailable (macOS, etc.).
ctx.file("noop_cc_toolchain_config.bzl", _NOOP_CC_TOOLCHAIN_CONFIG_BZL)
ctx.file(
"clang_format_noop.sh",
"#!/usr/bin/env bash\n# Stub: mongo toolchain clang-format is not available on this platform.\nexit 0\n",
executable = True,
)
ctx.file(
"BUILD.bazel",
"""
# {} not supported on this platform
load("@rules_cc//cc/toolchains:cc_toolchain.bzl", "cc_toolchain")
load("@rules_shell//shell:sh_binary.bzl", "sh_binary")
load("//:noop_cc_toolchain_config.bzl", "noop_cc_toolchain_config")
package(default_visibility = ["//visibility:public"])
filegroup(
name = "all_files",
srcs = [],
)
filegroup(
name = "clang_tidy",
srcs = [],
)
sh_binary(
name = "clang_format",
srcs = ["clang_format_noop.sh"],
)
filegroup(
name = "llvm_symbolizer",
srcs = [],
)
filegroup(
name = "llvm_symbolizer_libs",
srcs = [],
)
cc_toolchain(
name = "cc_mongo_toolchain",
all_files = ":all_files",
ar_files = ":all_files",
compiler_files = ":all_files",
dwp_files = ":all_files",
linker_files = ":all_files",
objcopy_files = ":all_files",
strip_files = ":all_files",
toolchain_config = ":cc_mongo_toolchain_config",
)
noop_cc_toolchain_config(
name = "cc_mongo_toolchain_config",
)
""".format(ctx.attr.version) + (_RESOLVABLE_TOOLCHAIN if resolvable else _UNRESOLVABLE_TOOLCHAIN),
)
def get_toolchain_subs(ctx):
if ctx.attr.os:
os = ctx.attr.os
else:
os = ctx.os.name
if ctx.attr.arch:
arch = ctx.attr.arch
else:
arch = ctx.os.arch
arch = ARCH_NORMALIZE_MAP[arch]
version = ctx.attr.version
distro = get_host_distro_major_version(ctx)
def _toolchain_substitutions(platforms_arch, bazel_toolchain_cpu, arch, distro):
return {
"{platforms_arch}": platforms_arch,
"{bazel_toolchain_cpu}": bazel_toolchain_cpu,
"{exec_bazel_toolchain_cpu}": bazel_toolchain_cpu,
"{target_bazel_toolchain_cpu}": bazel_toolchain_cpu,
"{mongo_toolchain_constraint}": "@//bazel/platforms:use_mongo_toolchain",
"{exec_distro_constraint}": "",
"{target_distro_constraint}": "",
"{target_triple}": "",
"{extra_target_settings}": "",
"{extra_loads}": "",
"{extra_toolchains}": "",
"{host_linker_files}": "[]",
"{host_linker_bin_dirs}": "[]",
"{host_linker_resource_dir}": "\"\"",
"{host_linker_tool_path}": "\"\"",
"{host_linker_toolchain_repo_dir}": "\"\"",
"{host_ar_tool_path}": "\"\"",
"{host_dwp_tool_path}": "\"\"",
"{host_objcopy_tool_path}": "\"\"",
"{host_strip_tool_path}": "\"\"",
"{host_ar_files}": "\":all_files\"",
"{host_dwp_files}": "\":all_files\"",
"{host_objcopy_files}": "\":all_files\"",
"{host_strip_files}": "\":all_files\"",
"{toolchain_repo_name}": "mongo_toolchain_{version}".format(version = version),
"{arch}": arch,
"{version}": version,
"{distro}": distro,
}
if os != "linux":
substitutions = _toolchain_substitutions("arm64", arch, arch, distro)
generate_noop_toolchain(ctx, substitutions)
ctx.report_progress("mongo toolchain not supported on " + os + " and " + arch)
if arch == "aarch64":
substitutions = _toolchain_substitutions("arm64", arch, arch, distro)
elif arch == "x86_64":
substitutions = _toolchain_substitutions("x86_64", "x86_64", arch, distro)
elif arch == "ppc64le":
substitutions = _toolchain_substitutions("ppc64le", "ppc64le", arch, distro)
elif arch == "s390x":
substitutions = _toolchain_substitutions("s390x", arch, arch, distro)
else:
substitutions = _toolchain_substitutions("none", arch, arch, distro)
generate_noop_toolchain(ctx, substitutions)
ctx.report_progress("mongo toolchain not supported on " + os + " and " + arch)
substitutions["{toolchain_repo_dir}"] = "external/" + ctx.name
return distro, arch, substitutions
def _get_amazon_linux_2023_minor_version(repository_ctx):
"""Returns the minor release number of Amazon Linux 2023, e.g. 3 for 2023.3, or None."""
result = repository_ctx.execute([
"sed",
"-n",
"s/Amazon Linux release 2023\\.\\([0-9]*\\)\\..*/\\1/p",
"/etc/system-release",
])
if result.return_code != 0 or not result.stdout.strip():
return None
return result.stdout.strip()
def get_host_distro_major_version(repository_ctx):
_DISTRO_PATTERN_MAP = {
"Ubuntu 18*": "ubuntu18",
"Ubuntu 20*": "ubuntu20",
"Ubuntu 22*": "ubuntu22",
"Pop!_OS 22*": "ubuntu22",
"Ubuntu 24*": "ubuntu24",
"Ubuntu 26*": "ubuntu26",
"Amazon Linux 2": "amazon_linux_2",
"Amazon Linux 2023": "amazon_linux_2023",
"Debian GNU/Linux 10": "debian10",
"Debian GNU/Linux 12": "debian12",
"Debian GNU/Linux 13": "debian13",
"Red Hat Enterprise Linux 8*": "rhel8",
"Red Hat Enterprise Linux 9*": "rhel9",
"Red Hat Enterprise Linux 10*": "rhel10",
"Fedora*": "rhel10",
"SLES 15*": "suse15",
"SLES 16*": "suse16",
}
if repository_ctx.os.name != "linux":
return None
result = repository_ctx.execute([
"sed",
"-n",
"/^\\(NAME\\|VERSION_ID\\)=/{s/[^=]*=//;s/\"//g;p}",
"/etc/os-release",
])
if result.return_code != 0:
print("Failed to determine system distro, parsing os-release failed with the error: " + result.stderr)
return None
distro_seq = result.stdout.splitlines()
if len(distro_seq) != 2:
print("Failed to determine system distro, parsing os-release returned: " + result.stdout)
return None
distro_str = "{distro_name} {distro_version}".format(
distro_name = distro_seq[0],
distro_version = distro_seq[1],
)
for distro_pattern, simplified_name in _DISTRO_PATTERN_MAP.items():
if "*" in distro_pattern:
prefix_suffix = distro_pattern.split("*")
if distro_str.startswith(prefix_suffix[0]) and distro_str.endswith(prefix_suffix[1]):
return simplified_name
elif distro_str == distro_pattern:
if simplified_name == "amazon_linux_2023":
minor = _get_amazon_linux_2023_minor_version(repository_ctx)
if minor == "3":
return "amazon_linux_2023_3"
return simplified_name
return None