errors: only eagerly translate subdiagnostics #121085

davidtwco · 2024-02-14T14:31:39Z

Subdiagnostics don't need to be lazily translated, they can always be eagerly translated. Eager translation is slightly more complex as we need to have a DiagCtxt available to perform the translation, which involves slightly more threading of that context.

This slight increase in complexity should enable later simplifications - like passing DiagCtxt into AddToDiagnostic and moving Fluent messages into the diagnostic structs rather than having them in separate files (working on that was what led to this change).

r? @nnethercote

rustbot · 2024-02-14T14:31:49Z

Some changes might have occurred in exhaustiveness checking

cc @Nadrieril

Some changes occurred in compiler/rustc_codegen_gcc

cc @antoyo, @GuillaumeGomez

rustc_macros::diagnostics was changed

cc @davidtwco, @compiler-errors, @TaKO8Ki

The Miri subtree was changed

cc @rust-lang/miri

Some changes occurred to MIR optimizations

cc @rust-lang/wg-mir-opt

rustbot · 2024-02-14T16:32:25Z

Some changes occurred in src/tools/rustfmt

cc @rust-lang/rustfmt

rustc_errors::translation was changed

cc @davidtwco, @compiler-errors, @TaKO8Ki

nnethercote · 2024-02-14T23:19:41Z

compiler/rustc_errors/src/lib.rs

+    fn decorate(mut self, dcx: &DiagCtxtInner) -> Diagnostic {
+        // FIXME: Cannot use `Diagnostic::subdiagnostic` which takes `DiagCtxt`, but it
+        // just uses `DiagCtxtInner` functions.
+        let subdiag_with = |diag: &mut Diagnostic, msg| {


Can you pull this closure out further and use it for the FIXME cases above?

I'm not sure I understand what that would look like.

If you pulled this closure out into a method on DiagCtxtInner, then you could use it in flush_delayed, where the FIXME comment is.

It's a little tricky to do this. I can't make the closure a method because the signature wouldn't match (add_to_diagnostic_with doesn't expect a &Self), it can't be a function which doesn't take self because it needs the reference to DiagCtxtInner (which we capture currently), and returning a closure that does match from this function is tricky because the closure captures the lifetime. I'm hoping to be able to remove the closure here in a follow-up anyway.

compiler/rustc_expand/src/mbe/diagnostics.rs

compiler/rustc_macros/src/diagnostics/diagnostic_builder.rs

src/tools/rustfmt/src/parse/session.rs

nnethercote

This seems mostly good, but it feels like it's missing a couple of things.

Should DiagnosticMessage::Eager and SubdiagnosticMessage::Eager be renamed? Also, the comments on those variants talk about eager vs. non-eager translation, and should be updated.
Can AddToDiagnostic be simplified? Is add_to_diagnostic useful any more? Maybe add_to_diagnostic_with can be removed and add_to_diagnostic always do what that subdiag_with closure does (though that requires a dcx).

rustbot · 2024-02-15T09:52:12Z

rustc_error_messages was changed

cc @davidtwco, @compiler-errors, @TaKO8Ki

davidtwco · 2024-02-15T09:53:19Z

Should DiagnosticMessage::Eager and SubdiagnosticMessage::Eager be renamed? Also, the comments on those variants talk about eager vs. non-eager translation, and should be updated.

Fixed!

Can AddToDiagnostic be simplified? Is add_to_diagnostic useful any more? Maybe add_to_diagnostic_with can be removed and add_to_diagnostic always do what that subdiag_with closure does (though that requires a dcx).

I intend to do this as a follow-up, this is just the first part of larger changes I'm making to move Fluent messages into the structs and out of the ftl files, to merge IntoDiagnostic, AddToDiagnostic and LintDiagnostic together, and things like that.

compiler/rustc_error_messages/src/lib.rs

nnethercote · 2024-02-15T10:30:51Z

AFAICT you've updated DiagnosticMessage, but not SubdiagnosticMessage.

Subdiagnostics don't need to be lazily translated, they can always be eagerly translated. Eager translation is slightly more complex as we need to have a `DiagCtxt` available to perform the translation, which involves slightly more threading of that context. This slight increase in complexity should enable later simplifications - like passing `DiagCtxt` into `AddToDiagnostic` and moving Fluent messages into the diagnostic structs rather than having them in separate files (working on that was what led to this change). Signed-off-by: David Wood <david@davidtw.co>

davidtwco · 2024-02-15T10:35:26Z

AFAICT you've updated DiagnosticMessage, but not SubdiagnosticMessage.

🤦 fixed!

nnethercote · 2024-02-15T18:50:51Z

@bors r+

bors · 2024-02-15T18:50:54Z

📌 Commit b80fc5d has been approved by nnethercote

It is now in the queue for this repository.

…, r=nnethercote errors: only eagerly translate subdiagnostics Subdiagnostics don't need to be lazily translated, they can always be eagerly translated. Eager translation is slightly more complex as we need to have a `DiagCtxt` available to perform the translation, which involves slightly more threading of that context. This slight increase in complexity should enable later simplifications - like passing `DiagCtxt` into `AddToDiagnostic` and moving Fluent messages into the diagnostic structs rather than having them in separate files (working on that was what led to this change). r? `@nnethercote`

…, r=nnethercote errors: only eagerly translate subdiagnostics Subdiagnostics don't need to be lazily translated, they can always be eagerly translated. Eager translation is slightly more complex as we need to have a `DiagCtxt` available to perform the translation, which involves slightly more threading of that context. This slight increase in complexity should enable later simplifications - like passing `DiagCtxt` into `AddToDiagnostic` and moving Fluent messages into the diagnostic structs rather than having them in separate files (working on that was what led to this change). r? ``@nnethercote``

…iaskrgr Rollup of 8 pull requests Successful merges: - rust-lang#120952 (Don't use mem::zeroed in vec::IntoIter) - rust-lang#121079 (distribute tool documentations and avoid file conflicts on `x install`) - rust-lang#121085 (errors: only eagerly translate subdiagnostics) - rust-lang#121091 (use build.rustc config and skip-stage0-validation flag) - rust-lang#121193 (Use fulfillment in next trait solver coherence) - rust-lang#121209 (Make `CodegenBackend::join_codegen` infallible.) - rust-lang#121210 (Fix `cfg(target_abi = "sim")` on `i386-apple-ios`) - rust-lang#121228 (create stamp file for clippy) r? `@ghost` `@rustbot` modify labels: rollup

…iaskrgr Rollup of 9 pull requests Successful merges: - rust-lang#120952 (Don't use mem::zeroed in vec::IntoIter) - rust-lang#121085 (errors: only eagerly translate subdiagnostics) - rust-lang#121091 (use build.rustc config and skip-stage0-validation flag) - rust-lang#121149 (Fix typo in VecDeque::handle_capacity_increase() doc comment.) - rust-lang#121193 (Use fulfillment in next trait solver coherence) - rust-lang#121209 (Make `CodegenBackend::join_codegen` infallible.) - rust-lang#121210 (Fix `cfg(target_abi = "sim")` on `i386-apple-ios`) - rust-lang#121228 (create stamp file for clippy) - rust-lang#121231 (remove a couple of redundant clones) r? `@ghost` `@rustbot` modify labels: rollup

Rollup merge of rust-lang#121085 - davidtwco:always-eager-diagnostics, r=nnethercote errors: only eagerly translate subdiagnostics Subdiagnostics don't need to be lazily translated, they can always be eagerly translated. Eager translation is slightly more complex as we need to have a `DiagCtxt` available to perform the translation, which involves slightly more threading of that context. This slight increase in complexity should enable later simplifications - like passing `DiagCtxt` into `AddToDiagnostic` and moving Fluent messages into the diagnostic structs rather than having them in separate files (working on that was what led to this change). r? ```@nnethercote```

ytmimi · 2024-02-19T02:01:50Z

src/tools/rustfmt/src/parse/session.rs

+
+    // Override `translate_message` for the silent emitter because eager translation of
+    // subdiagnostics result in a call to this.
+    fn translate_message<'a>(
+        &'a self,
+        message: &'a rustc_errors::DiagnosticMessage,
+        _: &'a rustc_errors::translation::FluentArgs<'_>,
+    ) -> Result<Cow<'_, str>, rustc_errors::error::TranslateError<'_>> {
+        rustc_errors::emitter::silent_translate(message)
+    }


Haven't got confirmation from the user yet if they're using a recent nightly version of rustfmt, but we just got a report about the SilentEmitter in rust-lang/rustfmt#6082 and this PR is the only one I can think of that made changes recently. @davidtwco would you mind looking into this.

@davidtwco @nnethercote I got confirmation from the user who reported the issue that this started happening after they installed the latest nightly. Also, they helped to provide this minimal input snippet, which I've also tested and can confirm that it causes a panic when running rustfmt +nightly-2024-02-18

macro_rules! test { ($T:ident, $b:lifetime) => { Box<$T<$b>> }; }

Yes, this PR is almost certainly the cause. I assume @davidtwco will look into it soon.

I'm working on this in #121301.

@davidtwco thank you for the help on this one 🙏🏼

…, r=nnethercote errors: only eagerly translate subdiagnostics Subdiagnostics don't need to be lazily translated, they can always be eagerly translated. Eager translation is slightly more complex as we need to have a `DiagCtxt` available to perform the translation, which involves slightly more threading of that context. This slight increase in complexity should enable later simplifications - like passing `DiagCtxt` into `AddToDiagnostic` and moving Fluent messages into the diagnostic structs rather than having them in separate files (working on that was what led to this change). r? ```@nnethercote```

rustbot assigned nnethercote Feb 14, 2024

This comment has been minimized.

Sign in to view

davidtwco force-pushed the always-eager-diagnostics branch from 2bd6241 to f6cd442 Compare February 14, 2024 15:12

This comment has been minimized.

Sign in to view

davidtwco force-pushed the always-eager-diagnostics branch from f6cd442 to 7721242 Compare February 14, 2024 16:32

nnethercote reviewed Feb 14, 2024

View reviewed changes

compiler/rustc_expand/src/mbe/diagnostics.rs Show resolved Hide resolved

nnethercote reviewed Feb 14, 2024

View reviewed changes

compiler/rustc_macros/src/diagnostics/diagnostic_builder.rs Show resolved Hide resolved

nnethercote reviewed Feb 14, 2024

View reviewed changes

src/tools/rustfmt/src/parse/session.rs Show resolved Hide resolved

nnethercote reviewed Feb 14, 2024

View reviewed changes

davidtwco force-pushed the always-eager-diagnostics branch from 7721242 to 2097ec3 Compare February 15, 2024 09:52

nnethercote reviewed Feb 15, 2024

View reviewed changes

compiler/rustc_error_messages/src/lib.rs Outdated Show resolved Hide resolved

davidtwco force-pushed the always-eager-diagnostics branch from 2097ec3 to b80fc5d Compare February 15, 2024 10:35

bors added S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels Feb 15, 2024

oli-obk mentioned this pull request Feb 15, 2024

Rollup of 20 pull requests #121164

Closed

matthiaskrgr mentioned this pull request Feb 17, 2024

Rollup of 9 pull requests #121234

Closed

matthiaskrgr mentioned this pull request Feb 17, 2024

Rollup of 8 pull requests #121235

Closed

matthiaskrgr mentioned this pull request Feb 17, 2024

Rollup of 9 pull requests #121240

Merged

bors merged commit 45d5773 into rust-lang:master Feb 17, 2024
11 checks passed

rustbot added this to the 1.78.0 milestone Feb 17, 2024

ytmimi reviewed Feb 19, 2024

View reviewed changes

ytmimi mentioned this pull request Feb 19, 2024

thread 'main' panicked at src/tools/rustfmt/src/parse/session.rs:42:9: silent emitter attempted to translate a diagnostic rust-lang/rustfmt#6082

Closed

davidtwco deleted the always-eager-diagnostics branch February 19, 2024 13:10

nnethercote mentioned this pull request Feb 23, 2024

Explicitly call emit_stashed_diagnostics. #121487

Merged

matthiaskrgr mentioned this pull request Feb 26, 2024

ICE: rustfmt: silent emitter attempted to translate a diagnostic #121638

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

errors: only eagerly translate subdiagnostics #121085

errors: only eagerly translate subdiagnostics #121085

davidtwco commented Feb 14, 2024

rustbot commented Feb 14, 2024

This comment has been minimized.

This comment has been minimized.

rustbot commented Feb 14, 2024

nnethercote Feb 14, 2024

davidtwco Feb 15, 2024

nnethercote Feb 15, 2024

davidtwco Feb 15, 2024

nnethercote Feb 15, 2024

nnethercote left a comment

rustbot commented Feb 15, 2024

davidtwco commented Feb 15, 2024

nnethercote commented Feb 15, 2024

davidtwco commented Feb 15, 2024

nnethercote commented Feb 15, 2024

bors commented Feb 15, 2024

ytmimi Feb 19, 2024

ytmimi Feb 19, 2024

nnethercote Feb 19, 2024

davidtwco Feb 21, 2024

ytmimi Feb 21, 2024

errors: only eagerly translate subdiagnostics #121085

errors: only eagerly translate subdiagnostics #121085

Conversation

davidtwco commented Feb 14, 2024

rustbot commented Feb 14, 2024

This comment has been minimized.

This comment has been minimized.

rustbot commented Feb 14, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

nnethercote left a comment

Choose a reason for hiding this comment

rustbot commented Feb 15, 2024

davidtwco commented Feb 15, 2024

nnethercote commented Feb 15, 2024

davidtwco commented Feb 15, 2024

nnethercote commented Feb 15, 2024

bors commented Feb 15, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment