bootstrap: Various Step refactors

[jyn514]

This ended up being a bunch of only somewhat related changes, sorry 😓 on the bright side, they're all fairly small and well-commented in the commit messages.

• Document ShouldRun::crates

• Switch Steps from crates to crate_or_deps where possible

  and document why the single remaining place can't switch

• Switch doc::{Std, Rustc} to crate_or_deps

  Previously they were using all_krates and various hacks to determine
  which crates to document. Switch them to crate_or_deps so ShouldRun
  tells them which crate to document instead of having to guess.

  This also makes a few other refactors:

  • Remove the now unused all_krates; new code should only use
    crate_or_deps.
  • Add tests for documenting Std
  • Remove the unnecessary run_cargo_rustdoc_for closure so that we only
    run cargo once
  • Give a more helpful error message when documenting a no_std target
  • Use builder.msg in the Steps instead of builder.info

• Extend msg and description to work with any subcommand

  Previously description only supported Testing and Benchmarking,
  and msg gave weird results for doc (it would say Docing).

• Add a make_run_crates function and use it Rustc and Std

  This fixes the panic from the previous commit.

• Allow checking individual crates

  This is useful for profiling metadata generation.

  This comes very close to removing all_krates, but doesn't quite -
  there's one last usage left in doc. This is fixed in a later commit.

• Give a more helpful error when calling cargo_crates_in_set for an alias

  Before:

  thread 'main' panicked at 'no entry found for key', builder.rs:110:30
  

  After:

  thread 'main' panicked at 'missing crate for path library', check.rs:89:26
  

[rustbot]

r? @clubby789

(rustbot has picked a reviewer for you, use r? to override)

[clubby789]

Not super familiar with the code here but looks broadly good and (aside from a couple of nitpicks) a lot cleaner

[clubby789]

@bors r+

[bors]

📌 Commit e72ad594bfd2965bd8cc9774a64d5ba4fa23e53d has been approved by clubby789

It is now in the queue for this repository.

[jyn514]

Just for later reference - Mark and I discussed about a year ago that this is the eventual direction we want to move bootstrap towards: #95503 (comment), https://rust-lang.zulipchat.com/#narrow/stream/131828-t-compiler/topic/bootstrap.20.2B.20building.20individual.20crates.20.2395503/near/279768255

This keeps the design mentioned there, where ensure(doc::Std) and ensure(check::Std) always checks all crates, and only an explicit x check core is allowed to build a subset of crates. It actually goes a little further in that direction since x test core linkchecker would previously have only documented core, which seems wrong.

I think I've been the reviewer for all PRs involving builder.msg, but I do feel pretty strongly that we should switch to msg wherever possible, I'll open a tracking issue for that in a second.

[bors]

⌛ Testing commit e72ad594bfd2965bd8cc9774a64d5ba4fa23e53d with merge 3bd00e870384ff8b0547fbea6c0c38b8ebc6d850...

[bors]

💔 Test failed - checks-actions

[jyn514]

@bors r=clubby789

[bors]

📌 Commit c28ee60 has been approved by clubby789

It is now in the queue for this repository.

[RalfJung]

"a single call to paths will only ever generate a single call to paths" eh... what does that mean?^^

[jyn514]

oh, this is a blast from the past. probably i meant "a single call to make_run", by analogy with the docs above? i honestly don't remember though, it's been ages.