deref patterns: implement implicit deref patterns #138528

dianne · 2025-03-15T03:36:59Z

This implements implicit deref patterns (per https://hackmd.io/4qDDMcvyQ-GDB089IPcHGg#Implicit-deref-patterns) and adds tests and an unstable book chapter.

Best reviewed commit-by-commit. Overall there's a lot of additions, but a lot of that is tests, documentation, and simple(?) refactoring.

Tracking issue: #87121

r? @Nadrieril

rustbot · 2025-03-15T03:37:04Z

Some changes occurred in match checking

cc @Nadrieril

dianne · 2025-03-15T03:42:19Z

tests/ui/pattern/deref-patterns/typeck_fail.stderr

The implicit and explicit deref patterns report different expected types when matching on a String since string literal patterns are AdjustMode::Pass (meaning the String doesn't get peeled either). As a fun bonus, I think that means this is compatible with the string_deref_patterns feature, if we don't end up going with a more general solution.

dianne · 2025-03-15T03:54:50Z

compiler/rustc_hir_typeck/src/pat.rs

+                // If an inline const pattern has a library-defined pointer-like type and
+                // `deref_patterns` is enabled, don't implicitly add deref patterns for its ADT.
+                // Nothing in the library that implements `DerefPure` supports structural equality,
+                // but we don't check that until `const_to_pat` in THIR construction. By avoiding
+                // type errors here, we can get a more meaningful error there.
+                ty::Adt(adt, _) => AdjustMode::peel_until_adt(Some(*adt)),


I think this'll be dead if #138492 lands: no other expression patterns can be ty::Adt(..), right? I don't love leaving untested code in, so maybe this case should be removed (and maybe replaced with a debug assert?) if so. It's not too catastrophic if that assumption ends up being wrong: it'd lead to confusing diagnostics for a relatively niche error.

+1 on making this a debug assert

Done here at the bottom of the pat.rs diff (I can't seem to link to an individual line). Ideally I'd like if the way expression patterns were handled was a bit more consistent, but that'd be a job for a separate PR.

compiler/rustc_hir_typeck/src/expr_use_visitor.rs

dianne · 2025-04-03T23:06:26Z

Rebased to resolve a conflict: as of #138492, the inline_const_pat feature gate has been removed, so I've removed¹ a test for the interaction between implicit deref pattern typing and that feature. The implementation of inline const block patterns has been left intact though, so for now I've left in the inline-const-specific case in typeck that the test was for.

I added a // FIXME(inline_const_pat) comment to the test file in its place, in case inline_const_pat is eventually reinstated. ↩

Nadrieril

About the middle 4 commits (the ones that set up ResolvedPat): I think we can do simpler. To start with, #139449 should make peeling simpler overall, and then I'd prefer an enum over dyn Fn callbacks (cf Zulip DMs).

About the rest of the commits, LGTM and I left some comments.

compiler/rustc_mir_build/src/thir/pattern/mod.rs

compiler/rustc_hir_typeck/src/pat.rs

compiler/rustc_middle/src/ty/adjustment.rs

compiler/rustc_hir_typeck/src/expr_use_visitor.rs

mejrs

Thank you so much for pushing this feature! I'm really looking forward to using this. I can't really comment on the implementation but have some thoughts about the unstable book entry.

src/doc/unstable-book/src/language-features/deref-patterns.md

rustbot · 2025-04-08T09:58:32Z

Some changes occurred in src/tools/clippy

cc @rust-lang/clippy

dianne · 2025-04-08T10:01:03Z

Ty for the reviews! I've pushed an update to hopefully address them. Most of the changes I've added as new commits for ease of comparison, but since the ResolvedPat refactoring was kind of severe, I figured it'd be cleanest to edit the commit history so that it can be compared more easily with the master branch. In case it helps with the review, the changes between the two histories is here.

As for ResolvedPat and PeelKind, I'm still not totally happy with them, but I've been struggling to come up with anything simpler.

This isn't ready for merging, since I haven't rebased on top of #139449 yet.

Edit: The clippy change is in 64c7917, which changes the pat_adjustments table in TypeckResults to contain more information.

mejrs

I like this iteration of the docs, just one nit..

src/doc/unstable-book/src/language-features/deref-patterns.md

bors · 2025-04-10T16:20:23Z

☔ The latest upstream changes (presumably #139622) made this pull request unmergeable. Please resolve the merge conflicts.

dianne · 2025-04-11T02:29:02Z

I cleaned up the history a bit while rebasing. The change to make the pat_adjustments table contain the kind of each implicit deref is now the first commit, so everything else can properly depend on it. @rustbot ready

compiler/rustc_middle/src/ty/adjustment.rs

This allows us to better distinguish builtin and overloaded implicit dereferences.

Nadrieril · 2025-04-16T16:49:05Z

compiler/rustc_mir_build/src/thir/pattern/migration.rs

+        let implicit_deref_mutbls = adjustments.iter().map(|adjust| {
+            let &ty::Ref(_, _, mutbl) = adjust.source.kind() else {


In a future PR, we could store the mutability here, which would remove the need for matching on the type here and in clippy.

Nadrieril · 2025-04-16T16:50:51Z

compiler/rustc_mir_build/src/thir/pattern/mod.rs

-            && !adjustments.is_empty()
+            && adjustments.iter().any(|adjust| adjust.source.is_ref())


nit: I would prefer to match on the PatAdjust here.

done here. that's definitely better with PatAdjust::BuiltinDeref only being for ty::Ref(..), yeah

Nadrieril · 2025-04-16T17:51:19Z

compiler/rustc_hir_typeck/src/pat.rs

    /// Only peel reference types. This is used for explicit `deref!(_)` patterns, which dereference
    /// any number of `&`/`&mut` references, plus a single smart pointer.
    ExplicitDerefPat,
    /// Implicitly peel any number of references, and if `deref_patterns` is enabled, smart pointer
    /// ADTs. In order to peel only as much as necessary for the pattern to match, the `until_adt`
    /// field contains the ADT def that the pattern is a constructor for, if applicable, so that we
    /// don't peel it. See [`ResolvedPat`] for more information.
-    // TODO: add `ResolvedPat` and `until_adt`.
-    Implicit,
+    Implicit { until_adt: Option<AdtDef<'tcx>> },


nit: AdtDef has too much info, I think this should just be the adt's DefId.

tests/ui/pattern/deref-patterns/implicit-cow-deref.rs

Nadrieril

Looking gooood! I'm very happy with this PR now. I left a few nits that aren't too important, except I agree with your proposal here to check if there can be an expression pattern without inline_const_pats. I'm in favor of doing it in this PR so we don't forget.

Since this uses `pat_adjustments`, I've also tweaked the documentation to mention implicit deref patterns and made sure the pattern migration diagnostic logic accounts for it. I'll adjust `ExprUseVisitor` in a later commit and add some tests there for closure capture inference.

See the doc comment on `ResolvedPat` for more information. This and the next couple commits split resolution apart from checking for path, struct, and tuple struct patterns, in order to find the pattern's type before peeling the scrutinee. This helps us avoid peeling the scrutinee when the pattern could match it. The reason this handles errors from resolution after peeling is for struct and tuple struct patterns: we check their subpatterns even when they fail to resolve, to potentially catch more resolution errors. By doing this after peeling, we're able to use the updated `PatInfo`. I don't know if there's currently any observable difference from using the outdated `PatInfo`, but it could potentially be a source of subtle diagnostic bugs in the future, so I'm opting to peel first.

See the previous commit for details. This doesn't yet extract the struct pat's type's ADT def before peeling, but it should now be possible.

See the previous two commits.

This is the use for the previous commits' refactors; see the messages there for more information.

Implicit deref patterns allow previously ill-typed programs. Make sure they're still ill-typed without the feature gate. I've thrown in a test for `deref!(_)` too, though it seems it refers to `deref_patterns` as a library feature.

- Clarifies the uses of implicit and explicit deref patterns - Includes motivating examples for both syntaxes - Shows the interaction with match ergonomics for reference types - Cross-links with `string_deref_patterns` and `box_patterns` The examples are contrived, but hopefully the intent comes across.

rustbot assigned Nadrieril Mar 15, 2025

rustbot added S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. labels Mar 15, 2025

dianne commented Mar 15, 2025

View reviewed changes

This comment has been minimized.

Sign in to view

dianne force-pushed the implicit-deref-patterns branch from be33911 to dc134ca Compare March 15, 2025 04:34

This comment has been minimized.

Sign in to view

dianne force-pushed the implicit-deref-patterns branch from dc134ca to 6a5b56a Compare March 15, 2025 06:07

dianne commented Mar 27, 2025

View reviewed changes

compiler/rustc_hir_typeck/src/expr_use_visitor.rs Show resolved Hide resolved

dianne force-pushed the implicit-deref-patterns branch from 6a5b56a to 0cbc959 Compare April 3, 2025 23:06

Nadrieril requested changes Apr 6, 2025

View reviewed changes

rustbot added S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels Apr 6, 2025

mejrs reviewed Apr 6, 2025

View reviewed changes

src/doc/unstable-book/src/language-features/deref-patterns.md Outdated Show resolved Hide resolved

src/doc/unstable-book/src/language-features/deref-patterns.md Outdated Show resolved Hide resolved

src/doc/unstable-book/src/language-features/deref-patterns.md Outdated Show resolved Hide resolved

dianne force-pushed the implicit-deref-patterns branch from 0cbc959 to a006927 Compare April 8, 2025 09:58

dianne force-pushed the implicit-deref-patterns branch from a006927 to d412694 Compare April 9, 2025 08:04

mejrs approved these changes Apr 9, 2025

View reviewed changes

src/doc/unstable-book/src/language-features/deref-patterns.md Outdated Show resolved Hide resolved

dianne force-pushed the implicit-deref-patterns branch from d412694 to adefa58 Compare April 9, 2025 22:50

dianne force-pushed the implicit-deref-patterns branch from adefa58 to 9dc6231 Compare April 11, 2025 02:28

rustbot added S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. and removed S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. labels Apr 11, 2025

dianne commented Apr 13, 2025

View reviewed changes

compiler/rustc_middle/src/ty/adjustment.rs Outdated Show resolved Hide resolved

store the kind of pattern adjustments in pat_adjustments

f35eae7

This allows us to better distinguish builtin and overloaded implicit dereferences.

dianne force-pushed the implicit-deref-patterns branch from 9dc6231 to 86ee4ef Compare April 13, 2025 00:49

Nadrieril reviewed Apr 16, 2025

View reviewed changes

tests/ui/pattern/deref-patterns/implicit-cow-deref.rs Outdated Show resolved Hide resolved

Nadrieril reviewed Apr 16, 2025

View reviewed changes

tests/ui/pattern/deref-patterns/implicit-cow-deref.rs Show resolved Hide resolved

Nadrieril approved these changes Apr 16, 2025

View reviewed changes

dianne and others added 12 commits April 16, 2025 14:42

pattern typing for immutable implicit deref patterns

e4b7b3d

register DerefMut bounds for implicit mutable derefs

91d0b57

refactor struct pattern checking to get info for peeling

9196048

See the previous commit for details. This doesn't yet extract the struct pat's type's ADT def before peeling, but it should now be possible.

refactor tuple struct pattern checking to get info for peeling

1f40e9a

See the previous two commits.

don't peel ADTs the pattern could match

923d95c

This is the use for the previous commits' refactors; see the messages there for more information.

respect the tcx's recursion limit when peeling

977c9ab

upvar inference for implicit deref patterns

ff0d4bc

add a feature gate test

4c4b61b

Implicit deref patterns allow previously ill-typed programs. Make sure they're still ill-typed without the feature gate. I've thrown in a test for `deref!(_)` too, though it seems it refers to `deref_patterns` as a library feature.

Make cow_of_cow test a teeny bit more explicit

3b91b7a

dianne force-pushed the implicit-deref-patterns branch from fc51373 to 3b91b7a Compare April 16, 2025 22:44

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

deref patterns: implement implicit deref patterns #138528

deref patterns: implement implicit deref patterns #138528

dianne commented Mar 15, 2025

rustbot commented Mar 15, 2025

dianne Mar 15, 2025 •

edited

Loading

dianne Mar 15, 2025

Nadrieril Apr 16, 2025

dianne Apr 16, 2025

This comment has been minimized.

This comment has been minimized.

dianne commented Apr 3, 2025

Nadrieril left a comment •

edited

Loading

mejrs left a comment

rustbot commented Apr 8, 2025

dianne commented Apr 8, 2025 •

edited

Loading

mejrs left a comment

bors commented Apr 10, 2025

dianne commented Apr 11, 2025

Nadrieril Apr 16, 2025

Nadrieril Apr 16, 2025 •

edited

Loading

dianne Apr 16, 2025

Nadrieril Apr 16, 2025

dianne Apr 16, 2025

Nadrieril left a comment •

edited

Loading

		let implicit_deref_mutbls = adjustments.iter().map(\|adjust\| {
		let &ty::Ref(_, _, mutbl) = adjust.source.kind() else {

		&& !adjustments.is_empty()
		&& adjustments.iter().any(\|adjust\| adjust.source.is_ref())

deref patterns: implement implicit deref patterns #138528

Are you sure you want to change the base?

deref patterns: implement implicit deref patterns #138528

Conversation

dianne commented Mar 15, 2025

rustbot commented Mar 15, 2025

dianne Mar 15, 2025 • edited Loading

Choose a reason for hiding this comment

dianne Mar 15, 2025

Choose a reason for hiding this comment

Nadrieril Apr 16, 2025

Choose a reason for hiding this comment

dianne Apr 16, 2025

Choose a reason for hiding this comment

This comment has been minimized.

This comment has been minimized.

dianne commented Apr 3, 2025

Footnotes

Nadrieril left a comment • edited Loading

Choose a reason for hiding this comment

mejrs left a comment

Choose a reason for hiding this comment

rustbot commented Apr 8, 2025

dianne commented Apr 8, 2025 • edited Loading

mejrs left a comment

Choose a reason for hiding this comment

bors commented Apr 10, 2025

dianne commented Apr 11, 2025

Nadrieril Apr 16, 2025

Choose a reason for hiding this comment

Nadrieril Apr 16, 2025 • edited Loading

Choose a reason for hiding this comment

dianne Apr 16, 2025

Choose a reason for hiding this comment

Nadrieril Apr 16, 2025

Choose a reason for hiding this comment

dianne Apr 16, 2025

Choose a reason for hiding this comment

Nadrieril left a comment • edited Loading

Choose a reason for hiding this comment

dianne Mar 15, 2025 •

edited

Loading

Nadrieril left a comment •

edited

Loading

dianne commented Apr 8, 2025 •

edited

Loading

Nadrieril Apr 16, 2025 •

edited

Loading

Nadrieril left a comment •

edited

Loading