Ref suggestion #37658

GuillaumeGomez · 2016-11-08T21:58:39Z

No description provided.

rust-highfive · 2016-11-08T21:58:49Z

r? @Aatch

(rust_highfive has picked a reviewer for you, use r? to override)

GuillaumeGomez · 2016-11-08T22:16:03Z

r? @nikomatsakis
cc @eddyb

Sub part of #37388.

Mark-Simulacrum · 2016-11-08T23:29:36Z

Will this suggest &mut foo where foo is not mutable? If yes, perhaps we should phrase the suggestion differently in that case.

eddyb · 2016-11-09T00:04:44Z

@Mark-Simulacrum To me getting the suggestion then, after making that change, an error about the variable not being mutable (with a suggestion to add mut to the let in that error) seems excellent.

oli-obk · 2016-11-09T08:33:29Z

src/test/compile-fail/coercion-slice.rs

@@ -14,6 +14,5 @@ fn main() {
    let _: &[i32] = [0];
    //~^ ERROR mismatched types
    //~| expected type `&[i32]`
-    //~| found type `[{integer}; 1]`


isn't this a regression in error reporting?

I have absolutely no clue from where this change comes since my code just added suggestion.

It's probably a stray side-effect? Try wrapping in probe if you haven't already. In fact, you probably want a version of try_coerce with no side-effects (call it probe_coerce).

oli-obk · 2016-11-09T08:34:32Z

src/librustc_typeck/check/demand.rs

+            let suggestions = self.check_ref(expr, checked_ty, expected);
+            let mut err = self.report_mismatched_types(origin, expected, expr_ty, e);
+            if let Some(suggestions) = suggestions {
+                err.help(&suggestions);


why not report a suggestion instead of a help?

Apparently, such a thing doesn't exist. (Or at least I didn't find it...).

It's called span_suggestion: http://manishearth.github.io/rust-internals-docs/rustc/?search=fn%3A%20suggestion

I was looking for .suggest... You bet I didn't find it haha.

There is a span_suggestion method on DiagnosticBuilder:

/// Prints out a message with a suggested edit of the code. /// /// See `diagnostic::RenderSpan::Suggestion` for more information. pub fn span_suggestion<S: Into<MultiSpan>>(&mut self, sp: S, msg: &str, suggestion: String) -> &mut Self {

I think the idea is that you replace the given span with the text suggestion, and also display the help message msg. Something like:

err.span_suggestion(span, "consider using the `&` operator", format!("&{}", span_to_string(span))

GuillaumeGomez · 2016-11-10T00:07:28Z

@nikomatsakis: So, with the current code, it panics it certain cases. Here is the stack trace: https://gist.github.com/GuillaumeGomez/097fec3b6ac4257419fdf100115535f8

@eddyb: proposed a few things:

so instead of this https://github.com/rust-lang/rust/blob/master/src/librustc_typeck/check/coercion.rs#L372

or rather, in addition to

you'd have to do something like https://github.com/rust-lang/rust/blob/master/src/librustc_typeck/check/coercion.rs#L493 and rename unsizing_obligations to obligations

However he said to consult you first. Any opinion/direction to give me? :-/

nikomatsakis

@GuillaumeGomez sorry for the delayed review, thanks for breaking this out into a separate PR! I'd say it's close, but for a few concerns:

Can you change to use span_suggestion?
Also, can you create ui tests instead of compile-fail tests for those cases where you are showing off what your error message does?
I'd like to get to the bottom of the unexpected changes as well; I think that coercion may be registering trait obligations or something like that? (But this is supposed to result in assertion failures, I thought.)

Do you want me to dig into that last point, @GuillaumeGomez?

nikomatsakis · 2016-11-10T17:43:20Z

src/librustc_typeck/check/demand.rs

+            let suggestions = self.check_ref(expr, checked_ty, expected);
+            let mut err = self.report_mismatched_types(origin, expected, expr_ty, e);
+            if let Some(suggestions) = suggestions {
+                err.help(&suggestions);


There is a span_suggestion method on DiagnosticBuilder:

/// Prints out a message with a suggested edit of the code. /// /// See `diagnostic::RenderSpan::Suggestion` for more information. pub fn span_suggestion<S: Into<MultiSpan>>(&mut self, sp: S, msg: &str, suggestion: String) -> &mut Self {

I think the idea is that you replace the given span with the text suggestion, and also display the help message msg. Something like:

err.span_suggestion(span, "consider using the `&` operator", format!("&{}", span_to_string(span))

nikomatsakis · 2016-11-10T17:45:43Z

Oh, @GuillaumeGomez, just saw your comment, let me process it.

nikomatsakis · 2016-11-10T17:50:52Z

Yes, ok, those panics are sort of what I expected. =) I'm not sure what is the most elegant fix here. We've been dancing around this problem for a bit.

In this particular case, I wonder if we should just add a way to clone the fulfillment context in its entirety, and then just restore it. After all, this is the error path, so performance isn't especially important. This would be like a poor man's persistent data structure.

GuillaumeGomez · 2016-11-10T18:05:35Z

Well, performance is important even in case of failure when there are a lot of them (just imagine a code with more than 100 errors in it, it'd be a shame if it took a few seconds to run).

Also, considering my very poor background on the rust's internals, I'll need more explanation.

eddyb · 2016-11-10T18:57:13Z

@nikomatsakis But most of the laziness is already there, shouldn't we do a bit a plumbing instead?
Coercions already have to deal with obligations from unsizing.
Or am I misunderstanding how far lazy normalization goes?

nikomatsakis · 2016-11-12T01:31:33Z

@GuillaumeGomez

sorry, didn't get a chance to try out my sample patch today, will do monday

@eddyb

But most of the laziness is already there, shouldn't we do a bit a plumbing instead?

I'm not quite sure what you are thinking here...

eddyb · 2016-11-12T08:11:22Z

I mean that autoderef already has an obligations list registered late and so does coercion (but only for unsizing). By combining the two we might not even need the ability to register obligations inside a snapshot!

nikomatsakis · 2016-11-15T00:09:27Z

@eddyb perhaps; I'll have to look into the code in a bit more detail to see what you mean.

brson · 2016-11-16T01:36:54Z

src/test/compile-fail/dst-bad-coercions.rs


    // Test that we cannot convert from *-ptr to &S and &T (mut version)
    let x: *mut S = &mut S;
    let y: &S = x; //~ ERROR mismatched types
    let y: &T = x; //~ ERROR mismatched types
+                   //~^ ERROR the trait bound `*mut S: T` is not satisfied


These new cases also don't seem to be reflected by the content of the patch, which doesn't add these errors.

They're fallout from bounds that leak out of a successful coercion attempt.
I've explained in my previous comments a path towards avoiding this side-effect.

brson · 2016-11-16T01:38:51Z

It's not obvious what the purpose of this patch is. When adding features, consider explaining in the PR what the feature is and why it is important.

GuillaumeGomez · 2016-11-16T10:39:36Z

@brson: I'll, but first I need to remove side-effects and need @nikomatsakis' help. :)

bors · 2016-11-17T12:40:18Z

☔ The latest upstream changes (presumably #37717) made this pull request unmergeable. Please resolve the merge conflicts.

nikomatsakis · 2016-11-22T15:15:07Z

OK, so, I just pushed a (hacky) commit that removes the side-effects. I'm not sure yet if I'm comfortable landing this or not =) @eddyb, did suggest he thought a cleaner path might be easy to get, and naturally I'd prefer that -- but this seems...not too terrible, for a focused hack.

Though it raises the question of how much value these suggestions have in the first place.

eddyb · 2016-11-22T15:16:57Z

@nikomatsakis r- on the hack, it does too much for me to be comfortable with it.

nikomatsakis · 2016-11-22T22:31:54Z

ok, yeah, so on IRC @eddyb elaborated a touch more on what he was thinking (propagating obligations up in a nicer way). Maybe one of us can find time to tinker with it soon.

GuillaumeGomez · 2016-11-23T10:32:27Z

Let me know if I can help in any way.

@nikomatsakis

Ban registering obligations during InferCtxt snapshots. Back in #33852, a flag was added to `InferCtxt` to prevent rolling back a snapshot if obligations were added to some `FulfillmentContext` during the snapshot, to prevent leaking fresh inference variables (created during that snapshot, so their indices would get reused) in obligations, which could ICE or worse. But that isn't enough in the long run, as type-checking ends up relying on success implying that eager side-effects are fine, and while stray obligations *do* get caught nowadays, those errors prevent, e.g. the speculative coercions from #37658, which *have to* be rolled back *even* if they succeed. We can't just allow those obligations to stay around though, because we end up, again, in ICEs or worse. Instead, this PR modifies `lookup_method_in_trait_adjusted` to return `InferOk` containing the obligations that `Autoderef::finalize_as_infer_ok` can propagate to deref coercions. As there shouldn't be *anything* left that registers obligations during snapshots, it's completely banned. r? @nikomatsakis @arielb1

GuillaumeGomez · 2017-04-20T22:45:53Z

Ok, now I need to fix ui tests. But otherwise seems all good.

eddyb · 2017-04-20T22:47:34Z

src/librustc_typeck/check/coercion.rs

-use rustc::infer::type_variable::TypeVariableOrigin;
-use rustc::traits::{self, ObligationCause, ObligationCauseCode};
+use rustc::infer::type_variable::{TypeVariableOrigin};
+use rustc::traits::{self, /*FulfillmentContext,*/ ObligationCause, ObligationCauseCode};


Comment still here.

eddyb · 2017-04-20T22:47:49Z

src/librustc_typeck/check/coercion.rs

+        /*let saved_fulfillment_cx =
+            mem::replace(
+                &mut *self.inh.fulfillment_cx.borrow_mut(),
+                FulfillmentContext::new());*/


Same here. Also not a hack right now.

GuillaumeGomez · 2017-04-21T13:07:38Z

Tests passed! \o/

cc @eddyb @nikomatsakis

eddyb · 2017-04-21T13:12:07Z

@GuillaumeGomez You have comments to remove and a rebase fixup commit to squash.

GuillaumeGomez · 2017-04-21T20:52:01Z

Updated.

eddyb · 2017-04-21T20:53:02Z

src/librustc_typeck/check/coercion.rs

@@ -65,7 +65,7 @@ use check::{Diverges, FnCtxt};
 use rustc::hir;
 use rustc::hir::def_id::DefId;
 use rustc::infer::{Coercion, InferResult, InferOk, TypeTrace};
-use rustc::infer::type_variable::TypeVariableOrigin;
+use rustc::infer::type_variable::{TypeVariableOrigin};


Heh, leftover edit.

eddyb · 2017-04-21T20:53:16Z

src/librustc_typeck/check/coercion.rs

+    pub fn can_coerce(&self,
+                      expr_ty: Ty<'tcx>,
+                      target: Ty<'tcx>)
+                      -> bool {


Doesn't this signature fit on one line?

eddyb · 2017-04-21T20:54:20Z

src/librustc_typeck/check/coercion.rs

+        let coerce = Coerce::new(self, cause);
+        let result = self.probe(|_| coerce.coerce::<hir::Expr>(&[], source, target)).is_ok();
+
+        result


Can just have self.probe(...) as the returned expression.

GuillaumeGomez · 2017-04-21T20:59:16Z

@bors: r=nikomatsakis,eddyb

bors · 2017-04-21T20:59:16Z

📌 Commit 7ce1eb7 has been approved by nikomatsakis,eddyb

…komatsakis,eddyb Ref suggestion

Rollup of 4 pull requests - Successful merges: #37658, #41405, #41432, #41435 - Failed merges:

rust-highfive assigned Aatch Nov 8, 2016

rust-highfive assigned nikomatsakis and unassigned Aatch Nov 8, 2016

GuillaumeGomez force-pushed the ref_suggestion branch from 5131722 to 8a2215c Compare November 8, 2016 22:17

oli-obk reviewed Nov 9, 2016

View reviewed changes

GuillaumeGomez force-pushed the ref_suggestion branch from 5e858bf to af793af Compare November 9, 2016 23:51

nikomatsakis requested changes Nov 10, 2016

View reviewed changes

alexcrichton added the T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. label Nov 10, 2016

brson reviewed Nov 16, 2016

View reviewed changes

nikomatsakis force-pushed the ref_suggestion branch from cdea804 to ab4bdf8 Compare November 22, 2016 15:15

nikomatsakis mentioned this pull request Nov 29, 2016

Start of implementation of proposal for E0308 #37388

Closed

GuillaumeGomez force-pushed the ref_suggestion branch from 1852221 to f9a2ac0 Compare April 19, 2017 11:19

eddyb reviewed Apr 20, 2017

View reviewed changes

GuillaumeGomez added 2 commits April 21, 2017 12:28

Add suggestion for & coercions

d360091

Update tests

8fe3a9a

GuillaumeGomez force-pushed the ref_suggestion branch from f9a2ac0 to 1dbd0b4 Compare April 21, 2017 11:29

Create a new method to run coercion inside probe

7d3284e

GuillaumeGomez force-pushed the ref_suggestion branch from 1dbd0b4 to 8603060 Compare April 21, 2017 14:13

eddyb reviewed Apr 21, 2017

View reviewed changes

Update ui test

7ce1eb7

GuillaumeGomez force-pushed the ref_suggestion branch from 8603060 to 7ce1eb7 Compare April 21, 2017 20:58

frewsxcv added a commit to frewsxcv/rust that referenced this pull request Apr 22, 2017

Rollup merge of rust-lang#37658 - GuillaumeGomez:ref_suggestion, r=ni…

8284ef6

…komatsakis,eddyb Ref suggestion

frewsxcv mentioned this pull request Apr 22, 2017

Rollup of 5 pull requests #41455

Closed

frewsxcv added a commit to frewsxcv/rust that referenced this pull request Apr 22, 2017

Rollup merge of rust-lang#37658 - GuillaumeGomez:ref_suggestion, r=ni…

e7bf524

…komatsakis,eddyb Ref suggestion

frewsxcv mentioned this pull request Apr 22, 2017

Rollup of 5 pull requests #41458

Closed

frewsxcv added a commit to frewsxcv/rust that referenced this pull request Apr 22, 2017

Rollup merge of rust-lang#37658 - GuillaumeGomez:ref_suggestion, r=ni…

5da19f2

…komatsakis,eddyb Ref suggestion

frewsxcv mentioned this pull request Apr 22, 2017

Rollup of 4 pull requests #41459

Merged

bors added a commit that referenced this pull request Apr 22, 2017

Auto merge of #41459 - frewsxcv:rollup, r=frewsxcv

ff13b7c

Rollup of 4 pull requests - Successful merges: #37658, #41405, #41432, #41435 - Failed merges:

bors merged commit 7ce1eb7 into rust-lang:master Apr 22, 2017

GuillaumeGomez deleted the ref_suggestion branch April 22, 2017 22:04

Ref suggestion #37658

Ref suggestion #37658

Conversation

GuillaumeGomez commented Nov 8, 2016

rust-highfive commented Nov 8, 2016

GuillaumeGomez commented Nov 8, 2016

Mark-Simulacrum commented Nov 8, 2016

eddyb commented Nov 9, 2016

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

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

GuillaumeGomez commented Nov 10, 2016

nikomatsakis left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

nikomatsakis commented Nov 10, 2016

nikomatsakis commented Nov 10, 2016

GuillaumeGomez commented Nov 10, 2016 • edited Loading

eddyb commented Nov 10, 2016

nikomatsakis commented Nov 12, 2016

eddyb commented Nov 12, 2016 • edited Loading

nikomatsakis commented Nov 15, 2016

brson Nov 16, 2016 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

brson commented Nov 16, 2016 • edited Loading

GuillaumeGomez commented Nov 16, 2016

bors commented Nov 17, 2016

nikomatsakis commented Nov 22, 2016

eddyb commented Nov 22, 2016

nikomatsakis commented Nov 22, 2016

GuillaumeGomez commented Nov 23, 2016

GuillaumeGomez commented Apr 20, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

GuillaumeGomez commented Apr 21, 2017

eddyb commented Apr 21, 2017

GuillaumeGomez commented Apr 21, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

GuillaumeGomez commented Apr 21, 2017

bors commented Apr 21, 2017

GuillaumeGomez commented Nov 10, 2016 •

edited

Loading

eddyb commented Nov 12, 2016 •

edited

Loading

brson Nov 16, 2016 •

edited

Loading

brson commented Nov 16, 2016 •

edited

Loading