Stream: t-compiler/wg-polonius

Topic: rustc tests


lqd (Aug 05 2020 at 07:12, on Zulip):

today (it's already the last day of the sprint :scream:) I'll start with looking at the test suite failures and updating the hackmd report

lqd (Aug 05 2020 at 07:14, on Zulip):

I'll also try to point out the weird cases I mentioned before, just to get a few more eyes on them, and be a bit more certain if they're because of work to be done in the future (specifically, requiring chalk integration) or if they're just bugs

lqd (Aug 05 2020 at 09:38, on Zulip):

(I got distracted with instrumenting our computation steps and datalog rules so I'll actually start on this after lunch)

lqd (Aug 05 2020 at 14:16, on Zulip):

this one is a cool "failure" https://github.com/rust-lang/rust/blob/master/src/test/ui/closures/closure-expected-type/expect-region-supply-region-2.rs

lqd (Aug 05 2020 at 14:17, on Zulip):

cool because of // Borrowck doesn't get a chance to run, but if it did it should error here. as polonius sees the error

Albin Stjerna (Aug 05 2020 at 14:34, on Zulip):

Cool!

lqd (Aug 06 2020 at 17:30, on Zulip):

I've completed my first pass at the new failures and have updated the document. Most are benign changes to the test output needing and update, or a new test added since the last time we put up a PR for the test expectations

lqd (Aug 06 2020 at 17:32, on Zulip):

apart from that, https://github.com/rust-lang/rust/blob/master/src/test/ui/issues/issue-26217.rs now fails, which is expected, but I don't understand why it fails now, the test is old, has NLL expectations, I would have expected it to fail earlier. Maybe some recent changes (maybe the leak check ones) had an impact here but I don't know. It's weird but probably not that important ?

lqd (Aug 06 2020 at 17:34, on Zulip):

and this one https://github.com/rust-lang/rust/blob/master/src/test/ui/issues/issue-74564-if-expr-stack-overflow.rs (with a very big MIR if expression) doesn't terminate or is extremely slow, somewhere in polonius (not during fact generation, which takes a long while but completes). I haven't checked yet where exactly in the process this happens but I'll look at it next.

lqd (Aug 06 2020 at 17:44, on Zulip):

fact generation is slow because the CFG has 183K edges, and 2 origins equal at all points of the cfg

lqd (Aug 06 2020 at 18:28, on Zulip):

ah it's somewhere in move errors, I killed the test after 20mins

lqd (Aug 06 2020 at 18:39, on Zulip):

and this time, unlike clap, we can't just disable move errors, as var_maybe_partly_initialized_on_exit also seems to grow or loop

Last update: Jun 20 2021 at 00:30UTC