As seen on try here: https://treeherder.mozilla.org/#/jobs?repo=try&revision=bba3f8950ee48d3ef9a9cd021b4b9fa7ea5ec345&selectedJob=168992909

Julian, can you point us in the right direction for determining whether this is a false positive? Thanks!

Try this patch: https://hg.mozilla.org/try/rev/88e57127fbe2499f2ba76045878c5e1c9ada9ded (you can also change the git date and sha1 to try current valgrind master, although I don't know if it has more changes related to these things)

(In reply to Mike Hommey [:glandium] from comment #2) > Try this patch: > https://hg.mozilla.org/try/rev/88e57127fbe2499f2ba76045878c5e1c9ada9ded > (you can also change the git date and sha1 to try current valgrind master, > although I don't know if it has more changes related to these things) This yields a similar set of failures: https://treeherder.mozilla.org/#/jobs?repo=try&revision=738101ecd14efe19d132fb631cb57d19d18dad0a&selectedJob=169318348

I chased these around for a while, but I cannot see why they are happening nor a way to avoid them. I assume they are false errors. Without a smaller test case, it's difficult to say more.

WIP: a set of suppressions for Stylo as compiled by rustc 1.25. These seem to make it run mostly clean, and would be a good starting point for a fix for this bug. It's not my preferred solution -- that being, V reporting no errors at all -- but per the previous comment I am not confident that that's actually achievable.

Here's a run that uses the suppressions in comment 5 plus Chris's rustc-1.25 patch, and runs clean on V: https://treeherder.mozilla.org/#/jobs?repo=try&revision=a90062b2df22f87649aa6d1755f3116811ed86a4 By looking at the raw log https://taskcluster-artifacts.net/HMA-Ou0USVOx_KcbV_saBA/0/public/logs/live_backing.log its clear that the suppressions are getting used -- 15 different supps getting used about 2000 times in total.

Proposed "fix".

Comment on attachment 8962275 [details] [diff] [review] bug1447137-2.diff Review of attachment 8962275 [details] [diff] [review]: ----------------------------------------------------------------- Bleargh.

(In reply to Nicholas Nethercote [:njn] from comment #8) > Bleargh. I completely agree. At some point this approach is going to become infeasible. But for now I don't see any other option.

Something in the last two days landed some new instances of this involving Servo_StyleSheet_FromUTF8Bytes / mozilla::ServoStyleSheet::ParseSheetSync: https://treeherder.mozilla.org/#/jobs?repo=try&revision=c5104c4748dacd9c8e73ada0b24c8c4c65db5f41&selectedJob=171111429

It looks like Julian is on PTO, I will post the updated suppressions for review.

Pushed by cmanchester@mozilla.com: https://hg.mozilla.org/integration/mozilla-inbound/rev/35ffa0aa963b Valgrind errors building with rust 1.25: Conditional jump or move depends on uninitialised value in various parsing functions. r=njn