2
0
Fork 0
mirror of https://github.com/bazel-contrib/bazel-lib synced 2024-12-01 07:15:24 +00:00
bazel-lib/lib/private/extension_utils.bzl

105 lines
4.5 KiB
Python
Raw Normal View History

"""Utility functions for bzlmod extensions"""
def _toolchain_repos_bfs(mctx, get_tag_fn, toolchain_name, toolchain_repos_fn, default_repository = None, get_name_fn = None, get_version_fn = None):
"""Create toolchain repositories from bzlmod extensions using a breadth-first resolution strategy.
Toolchains are assumed to have a "default" or canonical repository name so that across
all invocations of the module extension with that name only a single toolchain repository
is created. As such, it is recommended to default the toolchain name in the extension's
tag class attributes so that diverging from the canonical name is a special case.
The resolved toolchain version will be the one invoked closest to the root module, following
Bazel's breadth-first ordering of modules in the dependency graph.
For example, given the module extension usage in a MODULE file:
```starlark
ext = use_extension("@my_lib//lib:extensions.bzl", "ext")
ext.foo_toolchain(version = "1.2.3") # Default `name = "foo"`
use_repo(ext, "foo")
register_toolchains(
"@foo//:all",
)
```
This macro would be used in the module extension implementation as follows:
```starlark
extension_utils.toolchain_repos(
mctx = mctx,
get_tag_fn = lambda tags: tags.foo_toolchain,
toolchain_name = "foo",
toolchain_repos_fn = lambda name, version: register_foo_toolchains(name = name, register = False),
get_version_fn = lambda attr: None,
)
```
Where `register_foo_toolchains` is a typical WORKSPACE macro used to register
the foo toolchain for a particular version, minus the actual registration step
which is done separately in the MODULE file.
This macro enforces that only root MODULEs may use a different name for the toolchain
in case several versions of the toolchain repository is desired.
Args:
mctx: The module context
get_tag_fn: A function that takes in `module.tags` and returns the tag used for the toolchain.
For example, `tag: lambda tags: tags.foo_toolchain`. This is required because `foo_toolchain`
cannot be accessed as a simple string key from `module.tags`.
toolchain_name: Name of the toolchain to use in error messages
toolchain_repos_fn: A function that takes (name, version) and creates a toolchain repository. This lambda
should call a typical reposotiory rule to create toolchains.
default_repository: Default name of the toolchain repository to pass to the repos_fn.
By default, it equals `toolchain_name`.
get_name_fn: A function that extracts the module name from the toolchain tag's attributes. Defaults
to grabbing the `name` attribute.
get_version_fn: A function that extracts the module version from the a tag's attributes. Defaults
to grabbing the `version` attribute. Override this to a lambda that returns `None` if
version isn't used as an attribute.
"""
if default_repository == None:
default_repository = toolchain_name
if get_name_fn == None:
get_name_fn = lambda attr: attr.name
if get_version_fn == None:
get_version_fn = lambda attr: attr.version
registrations = {}
for mod in mctx.modules:
for attr in get_tag_fn(mod.tags):
name = get_name_fn(attr)
version = get_version_fn(attr)
if name != default_repository and not mod.is_root:
fail("Only the root module may provide a name for the {} toolchain.".format(toolchain_name))
if name in registrations.keys():
if name == default_repository:
# Prioritize the root-most registration of the default toolchain version and
# ignore any further registrations (modules are processed breadth-first)
continue
if version == registrations[name]:
# No problem to register a matching toolchain twice
continue
fail("Multiple conflicting {} toolchains declared for name {} ({} and {})".format(
toolchain_name,
name,
version,
registrations[name],
))
else:
registrations[name] = version
for name, version in registrations.items():
toolchain_repos_fn(
name = name,
version = version,
)
extension_utils = struct(
toolchain_repos_bfs = _toolchain_repos_bfs,
)