929374 - Enable TI and IonMonkey for chrome script by default

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Till Schneidereit [:till]]

What it says on the tin. Testing of bug 907201 has not shown any regressions, so we should be good here. (Hopefully) final try push coming up.

Comment 1

[Till Schneidereit [:till]]

Attachment: Enable TI and IonMonkey for chrome script by default

Comment 2

[Till Schneidereit [:till]]

https://tbpl.mozilla.org/?tree=Try&rev=fa4a90d2dd57

Comment 3

[Till Schneidereit [:till]]

Let's do this with less bustage on the mercurial queue: https://tbpl.mozilla.org/?tree=Try&rev=5eeebc5c7c1b

Comment 4

[Jan de Mooij [:jandem]]

Comment on attachment 820210 [details] [diff] [review]
Enable TI and IonMonkey for chrome script by default

Review of attachment 820210 [details] [diff] [review]:
-----------------------------------------------------------------

Nice :)

Comment 5

[Till Schneidereit [:till]]

Sadly, the xpcshell-test failure from the try run is real. The test_unique.js test crashes. And because things would be too simple otherwise, it doesn't crash when I run it under gdb. Jandem, can you take a look at the crash[1] and tell me how best to investigate this?


[1]: https://tbpl.mozilla.org/php/getParsedLog.php?id=29602109&tree=Try#error1

Comment 6

[Jan de Mooij [:jandem]]

It's an xpcshell signal handler issue, filed bug 931861.

Comment 8

[Jan de Mooij [:jandem]]

I pushed this patch to Try again and I'm getting browser-chrome nsGlobalWindow leaks. Will see if I can reproduce this locally.

https://tbpl.mozilla.org/?tree=Try&showall=1&rev=c47ae63c9c06

Comment 9

[Jan de Mooij [:jandem]]

(In reply to Jan de Mooij [:jandem] from comment #8)
> I pushed this patch to Try again and I'm getting browser-chrome
> nsGlobalWindow leaks. Will see if I can reproduce this locally.

I can reproduce it locally when I run a single test, even without the patch, so it's likely a pre-existing issue. With help from :smaug I'm still trying to track it down.

I wrote a patch to fix xpcshell (bug 931861) and it's green now [0] so hopefully once we fix the BC leaks this will be good to go...

[0] https://tbpl.mozilla.org/?tree=Try&rev=a6a2b1fbf40d

Comment 10

[Jan de Mooij [:jandem]]

I'm trying to track down the Linux-only, opt-only Mochitest-BC nsGlobalWindow leaks when I turn on TI for chrome, but I can't reproduce them locally so I have to use Try.

Unfortunately the BC suite takes more than an hour to run so this takes a lot of time :( But at least it's not a Win32 PGO build or an intermittent failure...

Comment 11

[Jan de Mooij [:jandem]]

More info: with Baseline disabled, it's green. Baseline enabled without attaching IC stubs is also green. So there's likely an IC stub that's keeping stuff alive. Will try to narrow it down with more Try pushes...

Comment 12

[Jan de Mooij [:jandem]]

Bug 970645 fixes the Linux Opt nsGlobalWindow leaks I saw on Try. I also posted a patch for the xpcshell issue, bug 931861. Will do another Try push once these bugs are fixed.

Comment 13

[Jan de Mooij [:jandem]]

Yet another problem: a Linux-only profiler assert with Mochitest-BC in debug builds. With the patches in bug 832437 and bug 931861 everything is finally green on Linux x64:

https://tbpl.mozilla.org/?tree=Try&rev=c641fd52bec4

Comment 14

[Jan de Mooij [:jandem]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/3916979b9c1c

Comment 15

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/3916979b9c1c