[WIP] make LUB of higher-ranked things actually do EQ

[nikomatsakis]

This is an experimental PR that I'm posting for the purposes of observing its effects via cargobomb (a crater run showed no regressions). The existing LUB/GLB code does some rather complex things around higher-ranked types (notably, grubbing about with region constraints) that I would like to move away from. I am not sure how often it is important in practice. This branch converts LUB/GLB of higher-ranked things into EQ.

r? @eddyb (this is just an experiment, don't really review =)

[nikomatsakis]

@bors try

[bors]

⌛ Trying commit 7226cda with merge ff1db21...

[bors]

☀️ Test successful - status-travis
State: approved= try=True

[eddyb]

cc @rust-lang/infra Cargobomb run requested.

[Mark-Simulacrum]

Marking as waiting on author since Travis failed. If these are intentional failures, please let us (infra team) know!

[01:07:43] running 23 tests
[01:07:43] ....FFFFFF.F.FFF.......
[01:07:43] failures:
[01:07:43]
[01:07:43] ---- test::glb_bound_free_infer stdout ----
[01:07:43] thread 'test::glb_bound_free_infer' panicked at 'unexpected error computing LUB: RegionsInsufficientlyPolymorphic(BrAnon(1), '_#1r)', /checkout/src/librustc_driver/test.rs:418:22
[01:07:43] note: Run with `RUST_BACKTRACE=1` for a backtrace.
[01:07:43]
[01:07:43] ---- test::glb_bound_free stdout ----
[01:07:43] thread 'test::glb_bound_free' panicked at 'unexpected error computing LUB: RegionsInsufficientlyPolymorphic(BrAnon(1), ReFree(DefId { krate: CrateNum(0), node: DefIndex(0) => test_crate/8cd878b }, BrAnon(1)))', /checkout/src/l
ibrustc_driver/test.rs:418:22
[01:07:43]
[01:07:43] ---- test::glb_bound_static stdout ----
[01:07:43] thread 'test::glb_bound_static' panicked at 'unexpected error computing LUB: RegionsInsufficientlyPolymorphic(BrAnon(1), ReStatic)', /checkout/src/librustc_driver/test.rs:418:22
[01:07:43]
[01:07:43] ---- test::glb_free_free_with_common_scope stdout ----
[01:07:43] thread 'test::glb_free_free_with_common_scope' panicked at 'Unexpected error: mismatched types Expected: []', /checkout/src/librustc_driver/test.rs:78:12
[01:07:43]
[01:07:43] ---- test::lub_bound_bound_inverse_order stdout ----
[01:07:43] thread 'test::lub_bound_bound_inverse_order' panicked at 'unexpected error in LUB: expected bound lifetime parameterBrAnon(2), found concrete lifetime', /checkout/src/librustc_driver/test.rs:410:26
[01:07:43]
[01:07:43] ---- test::lub_bound_free stdout ----
[01:07:43] thread 'test::lub_bound_free' panicked at 'unexpected error in LUB: expected bound lifetime parameterBrAnon(1), found concrete lifetime', /checkout/src/librustc_driver/test.rs:410:26
[01:07:43]
[01:07:43] ---- test::lub_bound_static stdout ----
[01:07:43] thread 'test::lub_bound_static' panicked at 'unexpected error in LUB: expected bound lifetime parameterBrAnon(1), found concrete lifetime', /checkout/src/librustc_driver/test.rs:410:26
[01:07:43]
[01:07:43] ---- test::lub_free_bound_infer stdout ----
[01:07:43] thread 'test::lub_free_bound_infer' panicked at 'unexpected error in LUB: expected bound lifetime parameterBrAnon(1), found concrete lifetime', /checkout/src/librustc_driver/test.rs:410:26
[01:07:43]
[01:07:43] ---- test::lub_returning_scope stdout ----
[01:07:43] thread 'test::lub_returning_scope' panicked at 'Unexpected error: mismatched types Expected: []', /checkout/src/librustc_driver/test.rs:78:12
[01:07:43]
[01:07:43] ---- test::lub_free_free stdout ----
[01:07:43] thread 'test::lub_free_free' panicked at 'Unexpected error: mismatched types Expected: []', /checkout/src/librustc_driver/test.rs:78:12
[01:07:43]
[01:07:43]
[01:07:43] failures:
[01:07:43] test::glb_bound_free
[01:07:43] test::glb_bound_free_infer
[01:07:43] test::glb_bound_static
[01:07:43] test::glb_free_free_with_common_scope
[01:07:43] test::lub_bound_bound_inverse_order
[01:07:43] test::lub_bound_free
[01:07:43] test::lub_bound_static
[01:07:43] test::lub_free_bound_infer
[01:07:43] test::lub_free_free
[01:07:43] test::lub_returning_scope

[eddyb]

@Mark-Simulacrum This is not supposed to be merged and those test failures probably make sense given the changes - we only need a cargobomb run result from it.

[aidanhs]

Cargobomb run started. Sorry for the delay, we didn't have spare boxes and then I was away for a few days.

[Mark-Simulacrum]

Run is complete: http://cargobomb-reports.s3.amazonaws.com/pr-44211/index.html. 15 regressions straight off, but haven't triaged.

[eddyb]

@nikomatsakis Looks like no actual regressions to me, only some flaky tests and timeouts.

[arielb1]

Given that there were no regressions of crater, I think we can merge this.

However, I would like it if on an LUB failure, we could at least have a special error message noting the new compiler update.

[eddyb]

@arielb1 from what @nikomatsakis was telling me, this branch is only an experiment to confirm that a separate reimplementation strategy could work without breaking existing code.

[nikomatsakis]

While @eddyb is correct, now that the experiment was successful, I've been contemplating landing (some variant of) this branch, just to pave the way for the future work. It should be possible to make a variant of this that offers a specialized error message, at least.

[aidanhs]

@nikomatsakis just pinging you for triage purposes to see how your contemplation of this PR mentioned in your last comment is going :)

[aidanhs]

@nikomatsakis ping again to see where this PR is up to! If this was just an experimental PR, maybe worth closing and reopening a fresh one with changes to be reviewed/merged?

[nikomatsakis]

@aidanhs I think we've decided to turn into something a bit more than experimental. I'm still hoping to refresh the approach in the next day or two, so I'm inclined not to close just yet.

[carols10cents]

I'm still hoping to refresh the approach in the next day or two, so I'm inclined not to close just yet.

7 days ago

Triage ping! How's it going?

[nikomatsakis]

Some progress, got blocked by other things.

[shepmaster]

This has been inactive for 18 days, so I'm going to go ahead and close this to keep things tidy. As always, please reopen once some updates happen! Thanks!