rustdoc: Unindent doc fragments on `Attributes` construction
`Attributes` can be constructed at arbitrary points, even after the `unindent_comments` pass.
`Attributes` that are constructed too late end up unindented.
All doc fragments need to be eventually indented before use, so there are no reasons to not do this immediately during their construction.
Fixes https://rust-lang.zulipchat.com/#narrow/stream/266220-rustdoc/topic/.60unindent_comments.60.20cannot.20work.20as.20a.20separate.20pass.
I'm not sure how to make a minimized reproduction, but unindenting the fragments during their construction should fix the issue.. by construction, and I also verified that all doc strings now hit the `resolver_caches.markdown_links` cache in https://github.com/rust-lang/rust/pull/94857.
Rollup of 5 pull requests
Successful merges:
- #90630 (Create real parser for search queries)
- #96193 ([fuchsia] Add implementation for `current_exe`)
- #96196 (Remove assertion that all paths in `ShouldRun` exist)
- #96228 (Fix locations for intrinsics impls and change to links)
- #96236 (Add an explicit `Span` field to `OutlivesConstraint`)
Failed merges:
r? `@ghost`
`@rustbot` modify labels: rollup
Create real parser for search queries
You can test it [here](https://rustdoc.crud.net/imperio/improve-rustdoc-search/std/index.html).
This PR adds a real parser for the query engine in rustdoc. The parser is quite simple but it allows to makes query handling much easier. I added a new testsuite to ensure it works as expected and ran fuzzing checks on it for a few hours without problems.
So about the parser: as you can see in the screenshot, it handles recursive generics parsing. It also allows to set which item should use exact matching by adding double-quotes around it (look for `exact_search` in the screenshot).
Now about the query engine itself: I simplified it a lot thanks to the parsed query. It behaves mostly the same when there is only one argument, but is much more powerful when there are more than one.
When making this change, we also removed the support for multi-query.
PS: A big part of the PR is tests and test-related code. :)
r? `@camelid`
- Cache doc link resolutions obtained early
- Cache markdown links retrieved from doc strings early
- Rename and restructure the code in early doc link resolution to be closer to #94857
Stop using CRATE_DEF_INDEX outside of metadata encoding.
`CRATE_DEF_ID` and `CrateNum::as_def_id` are almost always what we want. We should not manipulate raw `DefIndex` outside of metadata encoding.
* Forbid generics without a path (so "<p>" is forbidden).
* Change `handleSingleArg` so that it takes `results_others`, `results_in_args` and `results_returned` as arguments instead of using the "global" variables.
* Change `createQueryElement` so that it returns the newly created element instead of taking `elems` as argument.
* Improve documentation
rustdoc: Rename `def_id` into `item_id` when the type is `ItemId` for readability
As `@notriddle` mentioned in https://github.com/rust-lang/rust/pull/96091, the field name is inaccurate. This PR fixes it by renaming it accordingly to its real type.
r? `@notriddle`
clarify doc(cfg) wording
The current "This is supported" wording implies that it's possible to
still use the item on other configurations, but in an unsupported way.
Changing this to "Available" removes this ambiguity.
remove find_use_placement
A more robust solution to finding where to place use suggestions was added in #94584.
The algorithm uses the AST to find the span for the suggestion so we pass this span
down to the HIR during lowering and use it instead of calling `find_use_placement`
Fixes#94941
The current "This is supported" wording implies that it's possible to
still use the item on other configurations, but in an unsupported way.
Changing this to "Available" removes this ambiguity.
errors: lazily load fallback fluent bundle
Addresses (hopefully) https://github.com/rust-lang/rust/pull/95667#issuecomment-1094794087.
Loading the fallback bundle in compilation sessions that won't go on to emit any errors unnecessarily degrades compile time performance, so lazily create the Fluent bundle when it is first required.
r? `@ghost` (just for perf initially)
