787568 - Upgrade clang to r161152

Status: RESOLVED FIXED

(Release Engineering :: General, defect)

Description

[Rafael Ávila de Espíndola (:espindola) (not reading bugmail)]

Attachment: build new version

Testing bug 784691 found a performance regression and lots of hard to reproduce issues. As part of trying to figure it out I tested r161152 on try with the new diff-talos.py and found that r161152 was not the one responsible for the regression.

Since the build is tested, deployed and doesn't cause any regressions, it is probably a good idea to enable it as a stepping stone.

Comment 1

[Rafael Ávila de Espíndola (:espindola) (not reading bugmail)]

Attachment: update the manifests

Comment 2

[Rafael Ávila de Espíndola (:espindola) (not reading bugmail)]

https://hg.mozilla.org/integration/mozilla-inbound/rev/98a37bb0fbdc
https://hg.mozilla.org/integration/mozilla-inbound/rev/a412d39f5526

Comment 3

[Rafael Ávila de Espíndola (:espindola) (not reading bugmail)]

I forgot to clobber the build, so to test the results at the time of the switch I did a dummy try push and one reverting the change. The results found by

diff-talos.py d1e8ba348798 19df46b223c2

are

--------------------------------------------------------------
dromaeo_css
MacOSX 10.8                       | (5387.485,     20.6175526751)  -> (5428.5875,    13.8504870083)  1.0076x better

dromaeo_dom
MacOSX 10.7                       | (281.467833333, 1.47639194691) -> (278.4475,     1.04198602984)  1.0108x worse

kraken
MacOSX 10.8                       | (3334.36666667, 6.37096002557) -> (3315.59166667, 4.94021473626) 1.0057x better

-------------------------------------------------------

Which is not amusing given that the try runs before the switch had no significant regressions :-(
Given that the regression (and the speedup) only show up in one version of OS X and we run the same binary on 3 of them, it is very likely this is just some form of bias on our testing, but I don't have time to check it right now as I am going on vacations.

CCing ehsan to decide if this should be reverted for now.

Comment 4

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/98a37bb0fbdc
https://hg.mozilla.org/mozilla-central/rev/a412d39f5526

Comment 5

[(no longer active)]

(In reply to comment #3)
> CCing ehsan to decide if this should be reverted for now.

Hmm, this is a tough call.  Do these clang versions provide us with other benefits?

Comment 6

[Rafael Ávila de Espíndola (:espindola) (not reading bugmail)]

> Hmm, this is a tough call.  Do these clang versions provide us with other
> benefits?

Mostly being a stepping stone for upgrading to r162360 in bug 784691 and that revision has a fix for http://llvm.org/pr13662.