wf: handle "livelock" checking before reaching `WfPredicates::compute`. #70170

eddyb · 2020-03-19T23:56:54Z

For wf::obligations's "livelock" handling, this PR shouldn't cause any behavioral changes, as the check moved to it should be equivalent to the old one in WfPredicates::compute.

However, it fixes #70168 by making other users of WfPredicates::compute (that is, wf::predicate_obligations and compute's own upvar handling) correct for ty::Infer, in that they now get a WellFormed(ty::Infer(_)) obligation instead of silently ignoring the type.

r? @nikomatsakis

eddyb · 2020-03-19T23:58:20Z

@bors try @rust-timer queue (there may be new noop shallow_resolve calls introduced)

rust-timer · 2020-03-19T23:58:21Z

Awaiting bors try build completion

bors · 2020-03-19T23:58:30Z

⌛ Trying commit 1b15eb9a35ef930ba373fc945ee84dcabededdff with merge 05f512090ead730e044e52571dba2da5ff8f9dcf...

eddyb · 2020-03-20T00:00:54Z

(I forgot to push some comment tweaks, hopefully it doesn't kill the try build)

eddyb · 2020-03-20T04:24:38Z

@rust-timer build 05f512090ead730e044e52571dba2da5ff8f9dcf

rust-timer · 2020-03-20T04:24:40Z

Queued 05f512090ead730e044e52571dba2da5ff8f9dcf with parent f4c675c, future comparison URL.

rust-timer · 2020-03-20T06:44:04Z

Finished benchmarking try commit 05f512090ead730e044e52571dba2da5ff8f9dcf, comparison URL.

eddyb · 2020-03-20T15:38:14Z

The worst regression, clap-rs-check clean has 143 less specializes calls? What's up with that?
But typeck_tables_of getting slower makes sense, and clap-rs-debug clean also has that.

I guess I'll try to come up with a synthetic stress test that hits the worst case of this PR (wf::obligations with ty::Infer that resolves to a different ty::Infer), and see what I can do to alleviate it.

But still, 7.5% regression of typeck_tables_of, from this change? That means a lot of clap-rs' type-checking time is spent on solving WF obligations.

bors · 2020-03-21T08:07:39Z

☔ The latest upstream changes (presumably #70205) made this pull request unmergeable. Please resolve the merge conflicts.

eddyb · 2020-04-03T09:35:23Z

src/librustc_trait_selection/traits/wf.rs

                    for upvar_ty in substs.as_closure().upvar_tys() {
+                        // FIXME(eddyb) add the type to `walker` instead of recursing.
                        self.compute(upvar_ty);


Wait I forgot about this, this is technically a bugfix (before, if compute returned false here, those inference variables would be forgotten about).

So it might make sense that it's slower. ~~Let me, uh, make a PR that only changes this, and check that for perf regressions.~~

EDIT: Actually, there's the entirety of #70168, predicate_obligations being the other user of .compute w/o checking the return value. I might be able to pinpoint where the extra obligations are coming from.

This is it.

typeck_tables_of on clap-rs gets slightly faster (instead of being a significant regression), if I ignore upvar types that are ty::Infer(ty::TyVar(_)) and don't resolve.

But that's a WF hole, right? Do we just accept the regression?

Sorry, I lost track of this PR. I agree that this is a bug-fix.

eddyb · 2020-04-03T10:28:45Z

I added some fast paths, let's see if they improve anything (but also see #70170 (comment)):

@bors try @rust-timer queue

rust-timer · 2020-04-03T10:28:46Z

Awaiting bors try build completion

bors · 2020-04-03T10:28:56Z

⌛ Trying commit ef83aa34b11d92b10f53b514d04a5c2b475707d2 with merge bb0cbcbd0a0e408606210fa280cf86c90448764c...

bors · 2020-04-03T12:54:00Z

☀️ Try build successful - checks-azure
Build commit: bb0cbcbd0a0e408606210fa280cf86c90448764c (bb0cbcbd0a0e408606210fa280cf86c90448764c)

rust-timer · 2020-04-03T12:54:02Z

Queued bb0cbcbd0a0e408606210fa280cf86c90448764c with parent 424c793, future comparison URL.

rust-timer · 2020-04-03T15:50:55Z

Finished benchmarking try commit bb0cbcbd0a0e408606210fa280cf86c90448764c, comparison URL.

eddyb · 2020-04-03T16:49:22Z

Hehe, I think my second commit, wf: {Int,Float}Var can only infer to always-WF ints/floats., made the old macro-expanded version of inflate that perf.rust-lang.org tests, significantly faster to compile, it seems, so I guess it was spending a lot of time on WF-ing ty::IntVars?

I'm not surprised, there were a lot of macro-replicated integer literals in that code.

nikomatsakis · 2020-04-21T09:41:14Z

I'm inclined to merge this PR @eddyb -- r=me on the code itself -- the clap-rs-debug perf regression seems small and I agree it's a bug fix, and there are some significant (albeit somewhat surprising?) improvements, too. I guess your explanation makes sense, though.

eddyb · 2020-05-01T06:43:19Z

Let's get some new numbers, post-rust-lang/rustc-perf#645:

@bors try @rust-timer queue

rust-timer · 2020-05-01T06:43:21Z

Awaiting bors try build completion

bors · 2020-05-01T06:43:31Z

⌛ Trying commit 05a872d with merge ad57f0a2528d29aca128300e3fa1b174a9eb497e...

bors · 2020-05-01T09:21:49Z

☀️ Try build successful - checks-azure
Build commit: ad57f0a2528d29aca128300e3fa1b174a9eb497e (ad57f0a2528d29aca128300e3fa1b174a9eb497e)

rust-timer · 2020-05-01T09:21:50Z

Queued ad57f0a2528d29aca128300e3fa1b174a9eb497e with parent e94eaa6, future comparison URL.

rust-timer · 2020-05-01T20:36:31Z

Finished benchmarking try commit ad57f0a2528d29aca128300e3fa1b174a9eb497e, comparison URL.

eddyb · 2020-05-01T20:47:31Z

@bors r=nikomatsakis rollup=never

bors · 2020-05-01T20:47:32Z

📌 Commit 05a872d has been approved by nikomatsakis

bors · 2020-05-02T06:51:13Z

⌛ Testing commit 05a872d with merge 08dfbfb...

bors · 2020-05-02T10:13:50Z

☀️ Test successful - checks-azure
Approved by: nikomatsakis
Pushing 08dfbfb to master...

rust-highfive assigned nikomatsakis Mar 19, 2020

rust-highfive added the S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. label Mar 19, 2020

eddyb mentioned this pull request Mar 19, 2020

Potential WF unsoundness: predicate_obligations doesn't check compute's return type. #70168

Closed

eddyb force-pushed the wf-early-exit branch from 1b15eb9 to 60093d3 Compare March 20, 2020 00:00

JohnCSimon added S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels Mar 28, 2020

eddyb force-pushed the wf-early-exit branch from 60093d3 to ef83aa3 Compare April 3, 2020 09:34

eddyb commented Apr 3, 2020

View reviewed changes

lcnr mentioned this pull request Apr 9, 2020

WF-check all ty::Const's, not just array lengths. #70107

Merged

joelpalmer added S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. and removed S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. labels Apr 14, 2020

nikomatsakis approved these changes Apr 21, 2020

View reviewed changes

joelpalmer added S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. and removed S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. labels Apr 30, 2020

eddyb added 2 commits April 30, 2020 23:01

wf: handle "livelock" checking before reaching WfPredicates::compute.

d1dc2af

wf: {Int,Float}Var can only infer to always-WF ints/floats.

05a872d

eddyb force-pushed the wf-early-exit branch from ef83aa3 to 05a872d Compare April 30, 2020 20:09

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-author Status: This is awaiting some action (such as code changes or more information) from the author. labels May 1, 2020

bors added the merged-by-bors This PR was explicitly merged by bors. label May 2, 2020

bors merged commit 08dfbfb into rust-lang:master May 2, 2020

eddyb deleted the wf-early-exit branch May 2, 2020 11:54

aDotInTheVoid mentioned this pull request Dec 27, 2020

ICE unwrap of none value in librustc_hir/definitions.rs #74318

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

wf: handle "livelock" checking before reaching `WfPredicates::compute`. #70170

wf: handle "livelock" checking before reaching `WfPredicates::compute`. #70170

eddyb commented Mar 19, 2020

eddyb commented Mar 19, 2020

rust-timer commented Mar 19, 2020

bors commented Mar 19, 2020

eddyb commented Mar 20, 2020

eddyb commented Mar 20, 2020

rust-timer commented Mar 20, 2020

rust-timer commented Mar 20, 2020

eddyb commented Mar 20, 2020

bors commented Mar 21, 2020

eddyb Apr 3, 2020 •

edited

Loading

eddyb Apr 3, 2020

nikomatsakis Apr 21, 2020

eddyb commented Apr 3, 2020

rust-timer commented Apr 3, 2020

bors commented Apr 3, 2020

bors commented Apr 3, 2020

rust-timer commented Apr 3, 2020

rust-timer commented Apr 3, 2020

eddyb commented Apr 3, 2020 •

edited

Loading

nikomatsakis commented Apr 21, 2020

eddyb commented May 1, 2020

rust-timer commented May 1, 2020

bors commented May 1, 2020

bors commented May 1, 2020

rust-timer commented May 1, 2020

rust-timer commented May 1, 2020

eddyb commented May 1, 2020

bors commented May 1, 2020

bors commented May 2, 2020

bors commented May 2, 2020

wf: handle "livelock" checking before reaching WfPredicates::compute. #70170

wf: handle "livelock" checking before reaching WfPredicates::compute. #70170

Conversation

eddyb commented Mar 19, 2020

eddyb commented Mar 19, 2020

rust-timer commented Mar 19, 2020

bors commented Mar 19, 2020

eddyb commented Mar 20, 2020

eddyb commented Mar 20, 2020

rust-timer commented Mar 20, 2020

rust-timer commented Mar 20, 2020

eddyb commented Mar 20, 2020

bors commented Mar 21, 2020

eddyb Apr 3, 2020 • edited Loading

Choose a reason for hiding this comment

eddyb Apr 3, 2020

Choose a reason for hiding this comment

nikomatsakis Apr 21, 2020

Choose a reason for hiding this comment

eddyb commented Apr 3, 2020

rust-timer commented Apr 3, 2020

bors commented Apr 3, 2020

bors commented Apr 3, 2020

rust-timer commented Apr 3, 2020

rust-timer commented Apr 3, 2020

eddyb commented Apr 3, 2020 • edited Loading

nikomatsakis commented Apr 21, 2020

eddyb commented May 1, 2020

rust-timer commented May 1, 2020

bors commented May 1, 2020

bors commented May 1, 2020

rust-timer commented May 1, 2020

rust-timer commented May 1, 2020

eddyb commented May 1, 2020

bors commented May 1, 2020

bors commented May 2, 2020

bors commented May 2, 2020

wf: handle "livelock" checking before reaching `WfPredicates::compute`. #70170

wf: handle "livelock" checking before reaching `WfPredicates::compute`. #70170

eddyb Apr 3, 2020 •

edited

Loading

eddyb commented Apr 3, 2020 •

edited

Loading