Implement arbitrary_self_types #45870

mikeyhew · 2017-11-08T14:26:13Z

Partial implementation of #44874. Supports trait and struct methods with arbitrary self types, as long as the type derefs (transitively) to Self. Doesn't support raw-pointer self yet.

Methods with non-standard self types (i.e. anything other than &self, &mut self, and Box<Self>) are not object safe, because dynamic dispatch hasn't been implemented for them yet.

I believe this is also a (partial) fix for #27941.

If the feature is enabled, allow method `self` types to be any type that auto-derefs to `self`. - Currently, this supports inherent methods as well as trait methods. The plan AFAIK is to only allow this for trait methods, so I guess it won’t stay this way - Dynamic dispatch isn’t implemented yet, so the compiler will ICE if you define a trait method that takes `self: Rc<Self>` and try to call it on an `Rc<Trait>`. I will probably just make those methods non-object-safe initially.

…error Rewrote ExplicitSelf, adding a new `Other` variant for arbitrary self types. It’s a bit more sophisticated now, and checks for type equality, so you have to pass the type context and param env as arguments. There’s a borrow-checker error here that I have to fix Rewrote check_method_receiver, so it acts as if arbitrary self types are allowed, and then checks for ExplicitSelf::Other at the end and disallows it unless the feature is present.

Had to take the infer context as a parameter instead of the type context, so that the function can be called during inference

…of looping forever

- removed the inherent impls compile-fail test, because we’ll be supporting them - remove E0308-2 because it’s gonna be supported now (behind a feature gate) - replaced the mismatched method receiver error message with something better, so fixed the tests that that broke

- added some old code that used ExplicitSelf::determine to check for eqtype with the expected self type in the simple cases - this fixes problems with error messages being worse in those cases, which caused some compile-fail tests to fail

…osure, instead of infcx

…safety checks

I was only doing it for self_arg_ty, and ended up causing run-pass/associated-types-projection-from-known-type-in-impl.rs to fail.

…origin I doubt this changes anything, I was just trying to fix an issue with error messages and ended up doing this along with other things. Committing it separately so I can undo it easily.

… error messages that I don't understand why they changed, so the tests pass

arielb1 · 2017-11-08T17:16:29Z

src/librustc_typeck/check/wfcheck.rs

-                })
+
+        let cause = fcx.cause(span, ObligationCauseCode::MethodReceiver);
+        let eq = |expected, actual| fcx.at(&cause, fcx.param_env).eq(expected, actual);


I think you want to wrap this inside a commit_if_ok, because IIRC just an eq isn't transactional.

I don't know what you mean by transactional, but I'll try it; maybe it will help with the error messages problem

@arielb1 on second thought, I had another approach here. If I did that instead, would it address your concern? (i.e. is demand_eqtype_with_origin "transactional"?)

@mikeyhew

Sorry forgot to detail - the problem is that eq can have side-effects even if it fails (e.g. if you equate (_, u32) with (i32, i32), it will make the _ be i32 before erroring out).

(i.e. is demand_eqtype_with_origin "transactional"?)

Nope. You want fcx.commit_if_ok(|_| fcx.at(&cause, fcx.param_env).eq(expected, actual)), which creates a transaction and rolls it back if the equality fails.

arielb1 · 2017-11-08T17:18:03Z

src/test/compile-fail/explicit-self-lifetime-mismatch.rs

@@ -1,3 +1,5 @@
+//~  ERROR mismatched types


This doesn't make sense ~~how are you getting mismatched type errors on this line~~ you passed DUMMY_SP to a type error function, please use a correct span so that the error will appear in the correect line.

Ya so this is the thing I was having trouble with on Gitter. In e06cd31#diff-15d40489387405cc3e23e2f9f66a74c2L519 I removed some error checks that I (still think) are redundant, and it caused the error messages to go wonky. I can't figure out why

put the error message on one line so the test suite does not think it is two errors use a substring of the error message so it fits in 100 chars for tidy

arielb1 · 2017-11-08T22:54:53Z

The error wonkyness is caused by wonky constraint handling in region_inference - that function allows the cause of constraints from a snapshot to overwrite a cause from out of it:

rust/src/librustc/infer/region_inference/mod.rs

Lines 630 to 642 in 02004ef

    
           fn add_constraint(&self, constraint: Constraint<'tcx>, origin: SubregionOrigin<'tcx>) { 
        
               // cannot add constraints once regions are resolved 
        
               assert!(self.values_are_none()); 
        
               debug!("RegionVarBindings: add_constraint({:?})", constraint); 
        
               if self.constraints.borrow_mut().insert(constraint, origin).is_none() { 
        
                   if self.in_snapshot() { 
        
                       self.undo_log.borrow_mut().push(AddConstraint(constraint)); 
        
                   } 
        
               } 
        
           }

I think that converting this HashMap to keep the first cause instead of the last one should avoid problems like this.

…verwritten

now that we've fixed the bug where constraint origins were getting overwritten, the good error messages are back (with some tweaks)

mikeyhew · 2017-11-09T14:29:42Z

@arielb1 I fixed the bug, and the good error messages are back. There's a slight difference because instead of comparing &Self with &Self, we're comparing Self with Self.

There was also one ui test whose error message changed, presumably because of the bugfix (constraint origins are no longer being overwritten, even when we're outside of a snapshot).

arielb1 · 2017-11-09T14:36:45Z

src/test/ui/span/issue-27522.stderr


-error: aborting due to previous error
+error: arbitrary `self` types are unstable (see issue #44874)


Could you not have this error if you already emitted an "invalid self type" error?

Currently it emits both errors. I can change it though, wasn't sure whether to emit both or just the "invalid self type" error

You should change it to just emit the "invalid self type" error - you don't want to emit unnecessary feature gate errors because they are confusing.

…onstraint

arielb1 · 2017-11-09T15:43:37Z

This is LGTM, just:

change the code to not emit a feature gate error if the self type is truly invalid - spurious feature gate errors are confusing to end users.
add a run-pass test that tests that both trait and inherent methods with generalized self work.

r=me with these fixed

mikeyhew · 2017-11-09T17:49:30Z

@arielb1 should be good to go once the CI finishes running

arielb1 · 2017-11-09T17:52:39Z

@bors r+

bors · 2017-11-09T17:52:40Z

📌 Commit 77cd993 has been approved by arielb1

bors · 2017-11-12T07:31:14Z

⌛ Testing commit 77cd993 with merge b087ded...

@arielb1

Implement arbitrary_self_types r? @arielb1 cc @nikomatsakis Partial implementation of #44874. Supports trait and struct methods with arbitrary self types, as long as the type derefs (transitively) to `Self`. Doesn't support raw-pointer `self` yet. Methods with non-standard self types (i.e. anything other than `&self, &mut self, and Box<Self>`) are not object safe, because dynamic dispatch hasn't been implemented for them yet. I believe this is also a (partial) fix for #27941.

bors · 2017-11-12T09:45:55Z

☀️ Test successful - status-appveyor, status-travis
Approved by: arielb1
Pushing b087ded to master...

mikeyhew added 16 commits November 7, 2017 13:36

add tests for the arbitrary_self_types, which won't pass yet

edc8c76

Fix the lifetime error in ExplicitSelf

a3f5866

Had to take the infer context as a parameter instead of the type context, so that the function can be called during inference

Improve feature gate error, and return after emitting errors instead …

9a592e6

…of looping forever

get the old error messages back

aa0df3d

- added some old code that used ExplicitSelf::determine to check for eqtype with the expected self type in the simple cases - this fixes problems with error messages being worse in those cases, which caused some compile-fail tests to fail

Add arbitrary_self_types feature gate error to some tests

6fd2156

Update/improve documentation of ExpliciSelf

0d739b5

wrap error code in DiagnosticId::Error so it compiles

1d29966

modify ExplicitSelf::determine to take an is_self_type predicate cl…

0b27dcc

…osure, instead of infcx

move ExplicitSelf to rustc::ty::util, and use it to implement object …

3902643

…safety checks

normalize associated types in both self_ty and self_arg_ty

2369746

I was only doing it for self_arg_ty, and ended up causing run-pass/associated-types-projection-from-known-type-in-impl.rs to fail.

Switch from using At::eq to InferCtxt::can_eq and demand_eqtype_with_…

bbe755c

…origin I doubt this changes anything, I was just trying to fix an issue with error messages and ended up doing this along with other things. Committing it separately so I can undo it easily.

add a NOTE comment to the object safety test so that it passes

02ce3ac

Remove the error check that I think is redundant, and change the test…

e06cd31

… error messages that I don't understand why they changed, so the tests pass

kennytm added S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. T-lang Relevant to the language team, which will review and decide on the PR/issue. labels Nov 8, 2017

kennytm assigned arielb1 Nov 8, 2017

arielb1 reviewed Nov 8, 2017

View reviewed changes

mikeyhew added 4 commits November 8, 2017 15:03

tidy things up a bit

0a3a46d

shorten line length for tidy

0954e54

fix error message in arbitrary-self-types-not-object-safe test

aa00f17

put the error message on one line so the test suite does not think it is two errors use a substring of the error message so it fits in 100 chars for tidy

update ui test to new error message

7f8b003

mikeyhew force-pushed the arbitrary_self_types branch from 03b3294 to 7f8b003 Compare November 8, 2017 21:46

mikeyhew added 3 commits November 9, 2017 08:42

fix the bug in region_inference where constraint origins were being o…

dcbb27a

…verwritten

fixed all the compile-fail error messages

31d3783

now that we've fixed the bug where constraint origins were getting overwritten, the good error messages are back (with some tweaks)

update test/ui/static-lifetime.stderr with new error message

c046fda

arielb1 reviewed Nov 9, 2017

View reviewed changes

add a comment explaining the bugfix to infer::region_inference::add_c…

5d170f0

…onstraint

mikeyhew added 2 commits November 9, 2017 11:03

Don't emit the feature error if it's an invalid self type

ddc21d5

add run-pass tests

77cd993

kennytm 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 Nov 10, 2017

bors merged commit 77cd993 into rust-lang:master Nov 12, 2017

This was referenced Nov 12, 2017

Support ::crate in paths #45771

Merged

Generic Associated Types Parsing & Name Resolution #45904

Merged

std: Add a new wasm32-unknown-unknown target #45905

Merged

GabrielMajeri mentioned this pull request Nov 16, 2017

Use arbitrary self types (self: Arc<Self>) vulkano-rs/vulkano#878

Closed

Centril mentioned this pull request Sep 14, 2019

Stabilize nested self receivers in 1.41.0 #64325

Merged

adetaylor mentioned this pull request Nov 7, 2022

Tracking issue for RFC 3519: arbitrary_self_types #44874

Open

9 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Implement arbitrary_self_types #45870

Implement arbitrary_self_types #45870

mikeyhew commented Nov 8, 2017

arielb1 Nov 8, 2017

mikeyhew Nov 8, 2017

mikeyhew Nov 8, 2017

arielb1 Nov 8, 2017

arielb1 Nov 8, 2017 •

edited

Loading

mikeyhew Nov 8, 2017

arielb1 commented Nov 8, 2017 •

edited

Loading

mikeyhew commented Nov 9, 2017

arielb1 Nov 9, 2017

mikeyhew Nov 9, 2017

arielb1 Nov 9, 2017 •

edited

Loading

arielb1 commented Nov 9, 2017 •

edited

Loading

mikeyhew commented Nov 9, 2017

arielb1 commented Nov 9, 2017

bors commented Nov 9, 2017

bors commented Nov 12, 2017

bors commented Nov 12, 2017


		error: aborting due to previous error
		error: arbitrary `self` types are unstable (see issue #44874)

Implement arbitrary_self_types #45870

Implement arbitrary_self_types #45870

Conversation

mikeyhew commented Nov 8, 2017

arielb1 Nov 8, 2017

Choose a reason for hiding this comment

mikeyhew Nov 8, 2017

Choose a reason for hiding this comment

mikeyhew Nov 8, 2017

Choose a reason for hiding this comment

arielb1 Nov 8, 2017

Choose a reason for hiding this comment

arielb1 Nov 8, 2017 • edited Loading

Choose a reason for hiding this comment

mikeyhew Nov 8, 2017

Choose a reason for hiding this comment

arielb1 commented Nov 8, 2017 • edited Loading

mikeyhew commented Nov 9, 2017

arielb1 Nov 9, 2017

Choose a reason for hiding this comment

mikeyhew Nov 9, 2017

Choose a reason for hiding this comment

arielb1 Nov 9, 2017 • edited Loading

Choose a reason for hiding this comment

arielb1 commented Nov 9, 2017 • edited Loading

mikeyhew commented Nov 9, 2017

arielb1 commented Nov 9, 2017

bors commented Nov 9, 2017

bors commented Nov 12, 2017

bors commented Nov 12, 2017

arielb1 Nov 8, 2017 •

edited

Loading

arielb1 commented Nov 8, 2017 •

edited

Loading

arielb1 Nov 9, 2017 •

edited

Loading

arielb1 commented Nov 9, 2017 •

edited

Loading