Lower let-else in MIR #98574

dingxiangfei2009 · 2022-06-27T09:54:07Z

This MR will switch to lower let-else statements in MIR building instead.

To lower let-else in MIR, we build a mini-switch two branches. One branch leads to the matching case, and the other leads to the else block. This arrangement will allow temporary lifetime analysis running as-is so that the temporaries are properly extended according to the same rule applied to regular let statements.

cc #87335

Fix #98672

rust-highfive · 2022-06-27T09:54:18Z

Some changes occured in need_type_info.rs

cc @lcnr

rust-highfive · 2022-06-27T09:54:19Z

r? @wesleywiser

(rust-highfive has picked a reviewer for you, use r? to override)

rustbot · 2022-06-28T16:09:43Z

Some changes occurred in src/tools/clippy

cc @rust-lang/clippy

Some changes occurred in need_type_info.rs

cc @lcnr

src/tools/clippy/clippy_lints/src/attrs.rs

camsteffen · 2022-06-29T13:57:46Z

Is this intended to fix #98672?

dingxiangfei2009 · 2022-06-29T21:28:53Z

@camsteffen Yes, this is an attempt to.

bors · 2022-07-01T04:46:51Z

☔ The latest upstream changes (presumably #98706) made this pull request unmergeable. Please resolve the merge conflicts.

flip1995

Thanks for reverting the Clippy formatting changes! I have a few comments about the Clippy changes left. (Haven't reviewed any other changes.)

src/tools/clippy/clippy_utils/src/hir_utils.rs

src/tools/clippy/clippy_lints/src/returns.rs

src/tools/clippy/clippy_lints/src/needless_late_init.rs

rustbot · 2022-07-02T12:18:00Z

Some changes occurred to MIR optimizations

cc @rust-lang/wg-mir-opt

dingxiangfei2009 · 2022-07-02T12:34:49Z

compiler/rustc_typeck/src/expr_use_visitor.rs

-        self.walk_irrefutable_pat(&expr_place, &pat);
+        if let Some(els) = els {
+            // borrowing because we need to test the descriminant
+            self.borrow_expr(expr, ImmBorrow);


I might be wrong here. It looks necessary to borrow the initializer when it is a use of an upvar. This is one example:

rust/compiler/rustc_typeck/src/check/_match.rs

Line 134 in 683a9c8

let Expectation::IsLast(stmt) = orig_expected else {

Without this, we will have ICE later in building MIR when lowering the pattern bindings. match_pairs are not cleared because the ancestor path check fails and the orig_expected upvar fails to resolve.

dingxiangfei2009 · 2022-07-02T12:41:51Z

compiler/rustc_mir_transform/src/remove_uninit_drops.rs

@@ -102,7 +102,7 @@ fn is_needs_drop_and_init<'tcx>(
    let field_needs_drop_and_init = |(f, f_ty, mpi)| {
        let child = move_path_children_matching(move_data, mpi, |x| x.is_field_to(f));
        let Some(mpi) = child else {
-            return f_ty.needs_drop(tcx, param_env);
+            return Ty::needs_drop(f_ty, tcx, param_env);


Without this this code does not compile because the type of f_ty is not resolved to Ty.

Here is a reason why this compiled with lowering in AST but not anymore with lowering in MIR. With AST lowering, this let...else was transformed to:

if let Some(mpi) = child { is_needs_drop_and_init(tcx, param_env, maybe_inits, move_data, f_ty, mpi) } else { return f_ty.needs_drop(tcx, param_env); }

Since the is_needs_drop_and_init function call was visted first, typeck resolved the type of f_ty to Ty before visiting the return statement.

We no longer visit the expressions in this order with lowering MIR. This time, the return statement is visited first. We need to resolve the method call but f_ty has no type obligation at this point yet.

this change should not be necessary anymore since the order in typeck has been adjusted.

After the adjustment in order, even after moving the check on the else block into check_decl at the tail position, this is unfortunately still necessary as type annotation is needed is still emitted with the unchanged call for the reason possibly presented in the previous comment.

oh 🤦 now I see it. Thanks!

I guess I'd prefer to annotate the f_ty argument instead of changing this call site.

…li-obk Lower let-else in MIR This MR will switch to lower let-else statements in MIR building instead. To lower let-else in MIR, we build a mini-switch two branches. One branch leads to the matching case, and the other leads to the `else` block. This arrangement will allow temporary lifetime analysis running as-is so that the temporaries are properly extended according to the same rule applied to regular `let` statements. cc rust-lang#87335 Fix rust-lang#98672

Rollup of 5 pull requests Successful merges: - rust-lang#98574 (Lower let-else in MIR) - rust-lang#99011 (`UnsafeCell` blocks niches inside its nested type from being available outside) - rust-lang#99030 (diagnostics: error messages when struct literals fail to parse) - rust-lang#99155 (Keep unstable target features for asm feature checking) - rust-lang#99199 (Refactor: remove an unnecessary `span_to_snippet`) Failed merges: r? `@ghost` `@rustbot` modify labels: rollup

…li-obk Lower let-else in MIR This MR will switch to lower let-else statements in MIR building instead. To lower let-else in MIR, we build a mini-switch two branches. One branch leads to the matching case, and the other leads to the `else` block. This arrangement will allow temporary lifetime analysis running as-is so that the temporaries are properly extended according to the same rule applied to regular `let` statements. cc rust-lang#87335 Fix rust-lang#98672

Rollup of 5 pull requests Successful merges: - rust-lang#98574 (Lower let-else in MIR) - rust-lang#99011 (`UnsafeCell` blocks niches inside its nested type from being available outside) - rust-lang#99030 (diagnostics: error messages when struct literals fail to parse) - rust-lang#99155 (Keep unstable target features for asm feature checking) - rust-lang#99199 (Refactor: remove an unnecessary `span_to_snippet`) Failed merges: r? `@ghost` `@rustbot` modify labels: rollup

…=Mark-Simulacrum Add two let else regression tests Adds a regression test for rust-lang#94176, as it was fixed by rust-lang#98574 but doesn't have a regression test. The PR also incorporates a commit from rust-lang#94012 which added a test for an issue discovered in that PR. Originally they have been part of rust-lang#99291, but I've moved them out in the hopes of getting them merged more quickly, as that PR is already open since a month, and so that rust-lang#99291 can focus on the drop order part of things. Closes rust-lang#94176 Closes rust-lang#96961 -- dupe of rust-lang#94176

…plett Stabilize `let else` :tada: **Stabilizes the `let else` feature, added by [RFC 3137](rust-lang/rfcs#3137 🎉 Reference PR: rust-lang/reference#1156 closes rust-lang#87335 (`let else` tracking issue) FCP: rust-lang#93628 (comment) ---------- ## Stabilization report ### Summary The feature allows refutable patterns in `let` statements if the expression is followed by a diverging `else`: ```Rust fn get_count_item(s: &str) -> (u64, &str) { let mut it = s.split(' '); let (Some(count_str), Some(item)) = (it.next(), it.next()) else { panic!("Can't segment count item pair: '{s}'"); }; let Ok(count) = u64::from_str(count_str) else { panic!("Can't parse integer: '{count_str}'"); }; (count, item) } assert_eq!(get_count_item("3 chairs"), (3, "chairs")); ``` ### Differences from the RFC / Desugaring Outside of desugaring I'm not aware of any differences between the implementation and the RFC. The chosen desugaring has been changed from the RFC's [original](https://rust-lang.github.io/rfcs/3137-let-else.html#reference-level-explanations). You can read a detailed discussion of the implementation history of it in `@cormacrelf` 's [summary](rust-lang#93628 (comment)) in this thread, as well as the [followup](rust-lang#93628 (comment)). Since that followup, further changes have happened to the desugaring, in rust-lang#98574, rust-lang#99518, rust-lang#99954. The later changes were mostly about the drop order: On match, temporaries drop in the same order as they would for a `let` declaration. On mismatch, temporaries drop before the `else` block. ### Test cases In chronological order as they were merged. Added by df9a2e0 (rust-lang#87688): * [`ui/pattern/usefulness/top-level-alternation.rs`](https://github.com/rust-lang/rust/blob/1.58.1/src/test/ui/pattern/usefulness/top-level-alternation.rs) to ensure the unreachable pattern lint visits patterns inside `let else`. Added by 5b95df4 (rust-lang#87688): * [`ui/let-else/let-else-bool-binop-init.rs`](https://github.com/rust-lang/rust/blob/1.58.1/src/test/ui/let-else/let-else-bool-binop-init.rs) to ensure that no lazy boolean expressions (using `&&` or `||`) are allowed in the expression, as the RFC mandates. * [`ui/let-else/let-else-brace-before-else.rs`](https://github.com/rust-lang/rust/blob/1.58.1/src/test/ui/let-else/let-else-brace-before-else.rs) to ensure that no `}` directly preceding the `else` is allowed in the expression, as the RFC mandates. * [`ui/let-else/let-else-check.rs`](https://github.com/rust-lang/rust/blob/1.58.1/src/test/ui/let-else/let-else-check.rs) to ensure that `#[allow(...)]` attributes added to the entire `let` statement apply for the `else` block. * [`ui/let-else/let-else-irrefutable.rs`](https://github.com/rust-lang/rust/blob/1.58.1/src/test/ui/let-else/let-else-irrefutable.rs) to ensure that the `irrefutable_let_patterns` lint fires. * [`ui/let-else/let-else-missing-semicolon.rs`](https://github.com/rust-lang/rust/blob/1.58.1/src/test/ui/let-else/let-else-missing-semicolon.rs) to ensure the presence of semicolons at the end of the `let` statement. * [`ui/let-else/let-else-non-diverging.rs`](https://github.com/rust-lang/rust/blob/1.58.1/src/test/ui/let-else/let-else-non-diverging.rs) to ensure the `else` block diverges. * [`ui/let-else/let-else-run-pass.rs`](https://github.com/rust-lang/rust/blob/1.58.1/src/test/ui/let-else/let-else-run-pass.rs) to ensure the feature works in some simple test case settings. * [`ui/let-else/let-else-scope.rs`](https://github.com/rust-lang/rust/blob/1.58.1/src/test/ui/let-else/let-else-scope.rs) to ensure the bindings created by the outer `let` expression are not available in the `else` block of it. Added by bf7c32a (rust-lang#89965): * [`ui/let-else/issue-89960.rs`](https://github.com/rust-lang/rust/blob/1.58.1/src/test/ui/let-else/issue-89960.rs) as a regression test for the ICE-on-error bug rust-lang#89960 . Later in 102b912 this got removed in favour of more comprehensive tests. Added by 8565419 (rust-lang#89974): * [`ui/let-else/let-else-if.rs`](https://github.com/rust-lang/rust/blob/1.58.1/src/test/ui/let-else/let-else-if.rs) to test for the improved error message that points out that `let else if` is not possible. Added by 9b45713: * [`ui/let-else/let-else-allow-unused.rs`](https://github.com/rust-lang/rust/blob/1.61.0/src/test/ui/let-else/let-else-allow-unused.rs) as a regression test for rust-lang#89807, to ensure that `#[allow(...)]` attributes added to the entire `let` statement apply for bindings created by the `let else` pattern. Added by 61bcd8d (rust-lang#89841): * [`ui/let-else/let-else-non-copy.rs`](https://github.com/rust-lang/rust/blob/1.61.0/src/test/ui/let-else/let-else-non-copy.rs) to ensure that a copy is performed out of non-copy wrapper types. This mirrors `if let` behaviour. The test case bases on rustc internal changes originally meant for rust-lang#89933 but then removed from the PR due to the error prior to the improvements of rust-lang#89841. * [`ui/let-else/let-else-source-expr-nomove-pass.rs `](https://github.com/rust-lang/rust/blob/1.61.0/src/test/ui/let-else/let-else-source-expr-nomove-pass.rs) to ensure that while there is a move of the binding in the successful case, the `else` case can still access the non-matching value. This mirrors `if let` behaviour. Added by 102b912 (rust-lang#89841): * [`ui/let-else/let-else-ref-bindings.rs`](https://github.com/rust-lang/rust/blob/1.61.0/src/test/ui/let-else/let-else-ref-bindings.rs) and [`ui/let-else/let-else-ref-bindings-pass.rs `](https://github.com/rust-lang/rust/blob/1.61.0/src/test/ui/let-else/let-else-ref-bindings-pass.rs) to check `ref` and `ref mut` keywords in the pattern work correctly and error when needed. Added by 2715c5f (rust-lang#89841): * Match ergonomic tests adapted from the `rfc2005` test suite. Added by fec8a50 (rust-lang#89841): * [`ui/let-else/let-else-deref-coercion-annotated.rs`](https://github.com/rust-lang/rust/blob/1.61.0/src/test/ui/let-else/let-else-deref-coercion-annotated.rs) and [`ui/let-else/let-else-deref-coercion.rs`](https://github.com/rust-lang/rust/blob/1.61.0/src/test/ui/let-else/let-else-deref-coercion.rs) to check deref coercions. #### Added since this stabilization report was originally written (2022-02-09) Added by 76ea566 (rust-lang#94211): * [`ui/let-else/let-else-destructuring.rs`](https://github.com/rust-lang/rust/blob/1.63.0/src/test/ui/let-else/let-else-destructuring.rs) to give a nice error message if an user tries to do an assignment with a (possibly refutable) pattern and an `else` block, like asked for in rust-lang#93995. Added by e7730dc (rust-lang#94208): * [`ui/let-else/let-else-allow-in-expr.rs`](https://github.com/rust-lang/rust/blob/1.61.0/src/test/ui/let-else/let-else-allow-in-expr.rs) to test whether `#[allow(unused_variables)]` works in the expr, as well as its non presence, as well as putting it on the entire `let else` *affects* the expr, too. This was adding a missing test as pointed out by the stabilization report. * Expansion of `ui/let-else/let-else-allow-unused.rs` and `ui/let-else/let-else-check.rs` to ensure that non-presence of `#[allow(unused)]` does issue the unused lint. This was adding a missing test case as pointed out by the stabilization report. Added by 5bd7106 (rust-lang#94208): * [`ui/let-else/let-else-slicing-error.rs`](https://github.com/rust-lang/rust/blob/1.61.0/src/test/ui/let-else/let-else-slicing-error.rs), a regression test for rust-lang#92069, which got fixed without addition of a regression test. This resolves a missing test as pointed out by the stabilization report. Added by 5374688 (rust-lang#98574): * [`src/test/ui/async-await/async-await-let-else.rs`](https://github.com/rust-lang/rust/blob/master/src/test/ui/async-await/async-await-let-else.rs) to test the interaction of async/await with `let else` Added by 6c529de (rust-lang#98574): * [`src/test/ui/let-else/let-else-temporary-lifetime.rs`](https://github.com/rust-lang/rust/blob/master/src/test/ui/let-else/let-else-temporary-lifetime.rs) as a (partial) regression test for rust-lang#98672 Added by 9b56640 (rust-lang#99518): * [`src/test/ui/let-else/let-else-temp-borrowck.rs`](https://github.com/rust-lang/rust/blob/master/src/test/ui/let-else/let-else-temporary-lifetime.rs) as a regression test for rust-lang#93951 * Extension of `src/test/ui/let-else/let-else-temporary-lifetime.rs` to include a partial regression test for rust-lang#98672 (especially regarding `else` drop order) Added by baf9a7c (rust-lang#99518): * Extension of `src/test/ui/let-else/let-else-temporary-lifetime.rs` to include a partial regression test for rust-lang#93951, similar to `let-else-temp-borrowck.rs` Added by 60be2de (rust-lang#99518): * Extension of `src/test/ui/let-else/let-else-temporary-lifetime.rs` to include a program that can now be compiled thanks to borrow checker implications of rust-lang#99518 Added by 47a7a91 (rust-lang#100132): * [`src/test/ui/let-else/issue-100103.rs`](https://github.com/rust-lang/rust/blob/master/src/test/ui/let-else/issue-100103.rs), as a regression test for rust-lang#100103, to ensure that there is no ICE when doing `Err(...)?` inside else blocks. Added by e3c5bd6 (rust-lang#100443): * [`src/test/ui/let-else/let-else-then-diverge.rs`](https://github.com/rust-lang/rust/blob/master/src/test/ui/let-else/let-else-then-diverge.rs), to verify that there is no unreachable code error with the current desugaring. Added by 9818526 (rust-lang#100443): * [`src/test/ui/let-else/issue-94176.rs`](https://github.com/rust-lang/rust/blob/master/src/test/ui/let-else/issue-94176.rs), to make sure that a correct span is emitted for a missing trailing expression error. Regression test for rust-lang#94176. Added by e182d12 (rust-lang#100434): * [src/test/ui/unpretty/pretty-let-else.rs](https://github.com/rust-lang/rust/blob/master/src/test/ui/unpretty/pretty-let-else.rs), as a regression test to ensure pretty printing works for `let else` (this bug surfaced in many different ways) Added by e262856 (rust-lang#99954): * [`src/test/ui/let-else/let-else-temporary-lifetime.rs`](https://github.com/rust-lang/rust/blob/master/src/test/ui/let-else/let-else-temporary-lifetime.rs) extended to contain & borrows as well, as this was identified as an earlier issue with the desugaring: rust-lang#98672 (comment) Added by 2d8460e (rust-lang#99291): * [`src/test/ui/let-else/let-else-drop-order.rs`](https://github.com/rust-lang/rust/blob/master/src/test/ui/let-else/let-else-drop-order.rs) a matrix based test for various drop order behaviour of `let else`. Especially, it verifies equality of `let` and `let else` drop orders, [resolving](rust-lang#93628 (comment)) a [stabilization blocker](rust-lang#93628 (comment)). Added by 1b87ce0 (rust-lang#101410): * Edit to `src/test/ui/let-else/let-else-temporary-lifetime.rs` to add the `-Zvalidate-mir` flag, as a regression test for rust-lang#99228 Added by af591eb (rust-lang#101410): * [`src/test/ui/let-else/issue-99975.rs`](https://github.com/rust-lang/rust/blob/master/src/test/ui/let-else/issue-99975.rs) as a regression test for the ICE rust-lang#99975. Added by this PR: * `ui/let-else/let-else.rs`, a simple run-pass check, similar to `ui/let-else/let-else-run-pass.rs`. ### Things not currently tested * ~~The `#[allow(...)]` tests check whether allow works, but they don't check whether the non-presence of allow causes a lint to fire.~~ → *test added by e7730dc* * ~~There is no `#[allow(...)]` test for the expression, as there are tests for the pattern and the else block.~~ → *test added by e7730dc* * ~~`let-else-brace-before-else.rs` forbids the `let ... = {} else {}` pattern and there is a rustfix to obtain `let ... = ({}) else {}`. I'm not sure whether the `.fixed` files are checked by the tooling that they compile. But if there is no such check, it would be neat to make sure that `let ... = ({}) else {}` compiles.~~ → *test added by e7730dc* * ~~rust-lang#92069 got closed as fixed, but no regression test was added. Not sure it's worth to add one.~~ → *test added by 5bd7106* * ~~consistency between `let else` and `if let` regarding lifetimes and drop order: rust-lang#93628 (comment) → *test added by 2d8460e* Edit: they are all tested now. ### Possible future work / Refutable destructuring assignments [RFC 2909](https://rust-lang.github.io/rfcs/2909-destructuring-assignment.html) specifies destructuring assignment, allowing statements like `FooBar { a, b, c } = foo();`. As it was stabilized, destructuring assignment only allows *irrefutable* patterns, which before the advent of `let else` were the only patterns that `let` supported. So the combination of `let else` and destructuring assignments gives reason to think about extensions of the destructuring assignments feature that allow refutable patterns, discussed in rust-lang#93995. A naive mapping of `let else` to destructuring assignments in the form of `Some(v) = foo() else { ... };` might not be the ideal way. `let else` needs a diverging `else` clause as it introduces new bindings, while assignments have a default behaviour to fall back to if the pattern does not match, in the form of not performing the assignment. Thus, there is no good case to require divergence, or even an `else` clause at all, beyond the need for having *some* introducer syntax so that it is clear to readers that the assignment is not a given (enums and structs look similar). There are better candidates for introducer syntax however than an empty `else {}` clause, like `maybe` which could be added as a keyword on an edition boundary: ```Rust let mut v = 0; maybe Some(v) = foo(&v); maybe Some(v) = foo(&v) else { bar() }; ``` Further design discussion is left to an RFC, or the linked issue.

…low, r=oli-obk Restore control flow on error in EUV cc `@Nilstrieb` Fix rust-lang#104649 Since rust-lang#98574 refactored a piece of scrutinee memory categorization out as a subroutine, there is a subtle change in handling match arms especially when the categorization process faults and bails. In the correct case, it is not supposed to continue to process the arms any more. This PR restores the original control flow in EUV. I promise to add a compile-fail test to demonstrate that this indeed fixes the issue after coming back from a nap.

rustbot added the T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. label Jun 27, 2022

rust-highfive assigned wesleywiser Jun 27, 2022

rust-highfive added the S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. label Jun 27, 2022