Fix and slightly improve AstValidator
#120698
Closed
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
To ensure that `in_new_context` sets resets all "important" fields.
|
r? @TaKO8Ki (rustbot has picked a reviewer for you, use r? to override) |
rustbot
added
S-waiting-on-review
|
☔ The latest upstream changes (presumably #120550) made this pull request unmergeable. Please resolve the merge conflicts. |
fmease
added
S-waiting-on-author
GuillaumeGomez
added a commit
to GuillaumeGomez/rust
that referenced
this pull request
Mar 7, 2024
…piler-errors AST validation: Improve handling of inherent impls nested within functions and anon consts Minimal fix for issue rust-lang#121607 extracted from PR rust-lang#120698 for ease of backporting and since I'd like to improve PR rust-lang#120698 in such a way that it makes AST validator truly robust against such sort of regressions (AST validator is generally *beyond* footgun-y atm). The current version of PR rust-lang#120698 sort of does that already but there's still room for improvement. Fixes rust-lang#89342. Fixes [after beta-backport] rust-lang#121607. Partially addresses rust-lang#119924 (rust-lang#120698 aims to fully fix it). --- ### Explainer The last commit of PR rust-lang#119505 regressed issue rust-lang#121607. Previously we would reject visibilities on associated items with `visibility_not_permitted` if we were in a trait (by checking the parameter `ctxt` of `visit_assoc_item` which was 100% accurate) or if we were in a trait impl (by checking a flag called `in_trait_impl` tracked in `AstValidator` which was/is only accurate if the visitor methods correctly updated it which isn't actually the case giving rise to the old open issue rust-lang#89342). In PR rust-lang#119505, I moved even more state into the `AstValidator` by generalizing the flag `in_trait_impl` to `trait_or_trait_impl` to be able to report more precise diagnostics (modeling *Trait | TraitImpl*). However since we/I didn't update `trait_or_trait_impl` in all places to reflect reality (similar to us not updating `in_trait_impl` before), this lead to rust-lang#121607 (comment) getting wrongfully rejected. Since PR rust-lang#119505 we reject visibilities if the “globally tracked” (wrt. to `AstValidator`) `outer_trait_or_trait_impl` is `Some`. Crucially, when visiting an inherent impl, I never reset `outer_trait_or_trait_impl` back to `None` leading us to believe that `bar` in the stack [`trait Foo` > `fn foo` > `impl Bar` > `pub fn bar`] (from the MCVE) was an inherent associated item (we saw `trait Foo` but not `impl Bar` before it). The old open issue rust-lang#89342 is caused by the aforementioned issue of us never updating `in_trait_impl` prior to my PR rust-lang#119505 / `outer_trait_or_trait` after my PR. Stack: [`impl Default for Foo` > `{` > `impl Foo` > `pub const X`] (we only saw `impl Default for Foo` but not the `impl Foo` before it). --- This PR is only meant to be a *hot fix*. I plan on completely *rewriting* `AstValidator` from the ground up to not rely on “globally tracked” state like this or at least make it close to impossible to forget updating it when descending into nested items (etc.). Other visitors do a way better job at that (e.g. AST lowering). I actually plan on experimenting with moving more and more logic from `AstValidator` into the AST lowering pass/stage/visitor to follow the [Parse, don't validate](https://lexi-lambda.github.io/blog/2019/11/05/parse-don-t-validate/) “pattern”. --- r? `@compiler-errors`
rust-timer
added a commit
to rust-lang-ci/rust
that referenced
this pull request
Mar 8, 2024
Rollup merge of rust-lang#122004 - fmease:astvalidator-min-fix, r=compiler-errors AST validation: Improve handling of inherent impls nested within functions and anon consts Minimal fix for issue rust-lang#121607 extracted from PR rust-lang#120698 for ease of backporting and since I'd like to improve PR rust-lang#120698 in such a way that it makes AST validator truly robust against such sort of regressions (AST validator is generally *beyond* footgun-y atm). The current version of PR rust-lang#120698 sort of does that already but there's still room for improvement. Fixes rust-lang#89342. Fixes [after beta-backport] rust-lang#121607. Partially addresses rust-lang#119924 (rust-lang#120698 aims to fully fix it). --- ### Explainer The last commit of PR rust-lang#119505 regressed issue rust-lang#121607. Previously we would reject visibilities on associated items with `visibility_not_permitted` if we were in a trait (by checking the parameter `ctxt` of `visit_assoc_item` which was 100% accurate) or if we were in a trait impl (by checking a flag called `in_trait_impl` tracked in `AstValidator` which was/is only accurate if the visitor methods correctly updated it which isn't actually the case giving rise to the old open issue rust-lang#89342). In PR rust-lang#119505, I moved even more state into the `AstValidator` by generalizing the flag `in_trait_impl` to `trait_or_trait_impl` to be able to report more precise diagnostics (modeling *Trait | TraitImpl*). However since we/I didn't update `trait_or_trait_impl` in all places to reflect reality (similar to us not updating `in_trait_impl` before), this lead to rust-lang#121607 (comment) getting wrongfully rejected. Since PR rust-lang#119505 we reject visibilities if the “globally tracked” (wrt. to `AstValidator`) `outer_trait_or_trait_impl` is `Some`. Crucially, when visiting an inherent impl, I never reset `outer_trait_or_trait_impl` back to `None` leading us to believe that `bar` in the stack [`trait Foo` > `fn foo` > `impl Bar` > `pub fn bar`] (from the MCVE) was an inherent associated item (we saw `trait Foo` but not `impl Bar` before it). The old open issue rust-lang#89342 is caused by the aforementioned issue of us never updating `in_trait_impl` prior to my PR rust-lang#119505 / `outer_trait_or_trait` after my PR. Stack: [`impl Default for Foo` > `{` > `impl Foo` > `pub const X`] (we only saw `impl Default for Foo` but not the `impl Foo` before it). --- This PR is only meant to be a *hot fix*. I plan on completely *rewriting* `AstValidator` from the ground up to not rely on “globally tracked” state like this or at least make it close to impossible to forget updating it when descending into nested items (etc.). Other visitors do a way better job at that (e.g. AST lowering). I actually plan on experimenting with moving more and more logic from `AstValidator` into the AST lowering pass/stage/visitor to follow the [Parse, don't validate](https://lexi-lambda.github.io/blog/2019/11/05/parse-don-t-validate/) “pattern”. --- r? `@compiler-errors`
|
@fmease any updates on this? thanks |
|
Closing this as inactive. Feel free to reöpen this pr or create a new pr if you get the time to work on this. Thanks |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Fixes #89342.
Fixes #119924 (incl. #119924 (comment)).
Fixes [after beta-backport] #121607.
Draft status:
Contextstruct / methodin_new_context, thinking about ways to structure this stuff bettertests/ui/parser/semantic/but sth. liketests/ui/ast-validation/visit_attributewhich was missing before