bootstrap: Overhaul matching of command-line selectors to steps

[Zalathar]

The existing bootstrap code for matching command-line selectors (paths) to steps has to jump through hoops to reconstruct the relative order of command-line arguments, and to keep track of which selectors didn't end up matching any steps. It also makes an awkward distinction between test-suite paths and other kinds of path/alias.

We can simplify the matching process by iterating over the command-line selectors in an outer loop, and retaining that ordering in the resulting collection of work items as much as possible.

Specifically, we keep an ordered list of steps to run, and an ordered list of selected anchors to pass to each step. After deduplicating those lists, the resulting run order is mostly consistent with command-line order, while batching each step's work as much as possible.

---

This new implementation should give mostly the same outcomes as the previous implementation, with a few known differences:

• The order of selected anchors passed to each step is more consistent with command-line order in some cases (e.g. see the core+alloc+stdarch snapshot test).
• Test-suite paths are no longer processed in a separate earlier pass, so invocations like ./x test compiler ui will now run the compiler-crate unit tests before the UI tests.
• If multiple test paths are passed for the same test suite, snapshot tests no longer see multiple invocations of the underlying test step. (Step-caching would already prevent the duplicate invocations from actually running.)

---

There are further tweaks or enhancements that could be made, but I decided to stop here to stick to preserve existing behaviour as much as possible, outside of the differences that naturally fall out of the simpler implementation.

[rustbot]

r? @jieyouxu

rustbot has assigned @jieyouxu.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

Why was this reviewer chosen?

The reviewer was selected based on:

• Owners of files modified in this PR: bootstrap
• bootstrap expanded to 6 candidates
• Random selection from Mark-Simulacrum, clubby789, jieyouxu

[Zalathar]

[RUSTC-TIMING] serde_derive test:false 3.843
error: you are explicitly cloning with `.map()`
   --> src/bootstrap/src/core/builder/cli_paths.rs:139:23
    |
139 |         let anchors = step_anchors[&step_ix].iter().map(|p| PathSet::clone(p)).collect::<Vec<_>>();
    |                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ help: consider calling the dedicated `cloned` method: `step_anchors[&step_ix].iter().cloned()`
    |
    = help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#map_clone
    = note: `-D clippy::map-clone` implied by `-D warnings`
    = help: to override `-D warnings` add `#[allow(clippy::map_clone)]`

This is a clippy false-positive. Replacing the map-clone with .cloned() is incorrect, for two reasons:

• It calls <&PathSet>::clone rather than <PathSet>::clone, changing the output and causing a type error.
• Replacing the map-clone with .copied().cloned() would work, but loses the type-safety of explicitly specifying the type to be cloned.

[Kobzol]

Thank you, this logic is much easier to follow than the one before. Left some comments.

r? kobzol

View changes since this review

[Kobzol]

Thanks. You can r=me once PR CI is green.

[Zalathar]

• This will need a rebase to account for the changes in Fix handling of relative paths starting with a dot in bootstrap #160970.

[rustbot]

This PR was rebased onto a different main commit. Here's a range-diff highlighting what actually changed.

Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers.

[Zalathar]

Rebased with only trivial changes. I'll wait for PR CI before reapproving.

[Zalathar]

@bors r=Kobzol

[rust-bors]

📌 Commit aa512ce has been approved by Kobzol

It is now in the queue for this repository.