[Bazel@Head failure] _rust_host_tools_impl fails on Bazel 10 with unexpected Starlark value: <sort_key>

[kotlaja]

When running downstream builds utilizing rules_rust against Bazel 10 (e.g. HEAD / latest green commits), evaluation of the rust_host_tools module extension unconditionally crashes with an invalid attribute type error.

ERROR: .../external/rules_rust+/rust/extensions.bzl:272:44: Traceback (most recent call last):
	File ".../external/rules_rust+/rust/extensions.bzl", line 272, column 44, in _rust_host_tools_impl
		rust_toolchain_tools_repository(
Error in repository_rule: unexpected Starlark value: <sort_key> (of type sort_key)

How to test it

I used USE_BAZEL_VERSION=last_green bazel build //... inside Bazel Examples, for this repo: https://github.com/bazelbuild/examples/tree/main/rust-examples/05-deps-cargo.

The error can also be seen at the Bazel Downstream.

How I temporarily fixed it in Bazel Examples

By adding patch at MODULE.bazel:

archive_override(
    module_name = "rules_rust",
    urls = ["https://github.com/bazelbuild/rules_rust/releases/download/0.65.0/rules_rust-0.65.0.tar.gz"],
    patch_cmds = ["sed -i.bak 's/for key in dir(host_tools)/for key in [\"name\", \"version\", \"allocator_library\", \"dev_components\", \"edition\", \"rustfmt_version\", \"sha256s\", \"urls\"] if hasattr(host_tools, key)/' rust/extensions.bzl"],
)

[Wyverald]

@fmeum This makes me a bit nervous about backporting bazelbuild/bazel#27883 (for 9.1.0 specifically, looks like it was never cherry-picked: bazelbuild/bazel#27884). I hadn't realized that people might be depending on the return value of dir(a_tag). We could maybe remove the _sort_key field and have a method mctx.tag_sort_key(a_tag) instead. WDYT?

[fmeum]

Me neither. Yes, I think that's safer (and even more discoverable).

[UebelAndre]

I only just got Bazel 9 working and have yet to look at failures on rolling so if folks have context on what's broken I would love some support!