Tracking Issue for minicore (core prelude stubs test auxiliary)

[jieyouxu]

Context

We have a bunch of #![no_core] ui/codegen/assembly tests that roll their own core prelude stubs because they need to build (but not run) on a cross-compiled target. But this means:

• A bunch of duplicated implementations of core prelude stubs scattered in different tests.
• A heavy burden on a contributor of possibly having to update many tests if a new required lang item wants to be added.
• Prelude stubs can obfuscate the intent of the test (more things that the reader has to read and mentally filter out).

Instead of continuing to add more copies of core prelude stubs, let's centralize such core prelude stubs into a shared test auxiliary called minicore. minicore is for core prelude stubs specifically. std prelude which can include liballoc items and such is beyond the scope of minicore, because core is usable by more tests. This does not preclude adding another ministd or minialloc for stubs of std/liballoc respectively.

See #130375 for more discussions.

Steps

• 1. Open an MCP to add minicore test auxiliary.
  • MCP is accepted: Support tests to use minicore std/core prelude stubs compiler-team#786
  • Note that the MCP title specified std/core prelude stubs, but we revised the scope to core stubs only.
• 2. Implement initial test infra support for minicore.
  • Add minicore test auxiliary and support //@ add-core-stubs directive in ui/assembly/codegen tests #130693
• 3. Adjust rustc-dev-guide docs.
  • Describe minicore test auxiliary and directive rustc-dev-guide#2097
• 4. Figure out the logistics on how to absorb core stubs into minicore, then use minicore to simplify existing #![no_core] tests.
  • Currently, I think it's better to just have the infra in place to allow newer and changed tests to use minicore.rs, it seems like unnecessary churn to also force existing tests to use minicore.rs atm.

Unresolved questions

• What do we do with existing #![no_core] tests that could benefit from minicore?

[scottmcm]

Great news! I've been doing some MIR work that uses lang items, and ended up just making it work a bit worse instead of dealing with all the no-core tests. If there was a centralized place to deal with problems like this, that'd be great 🎉

[jieyouxu]

Initial test infra PR #130693 has now merged.

[RalfJung]

Thanks a lot for adding this! I only now finally got around to play with it, and it seems to work great. :)

I admit I am a bit confused by the naming -- it's called "core-stubs" but then the import is called "minicore". Are these two
different things, or are we using two different names to refer to the same thing?