784691 - Upgrade clang to r162360

Status: RESOLVED FIXED

(Release Engineering :: General, defect)

Description

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

This upgrade will include
* A fix to a visibility bug noticed when building firefox with libc++ (llvm.org/pr13662)
* asan improvements for os x.
* a bit of performance work.

Comment 1

[(not currently active) Ted Mielczarek]

I suspect you mean for this to be in Release Engineering.

Comment 2

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

Yes, thanks.

Comment 3

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

I have just copied the files with:

ssh [e-mail] -L 1500:relengweb1.dmz.scl3.mozilla.com:22
scp -P 1500 * respindola@localhost:/var/www/html/runtime-binaries/tooltool/sha512/
ssh -p 1500 respindola@localhost
cd /var/www/html/runtime-binaries/tooltool/sha512/
chmod 444 ...

Comment 4

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

Attachment: Use a local copy of tooltool and setup.sh

In the next update I will try to merge build-clang.py and create-manifest.py, this is a step in that direction.

Comment 5

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

Attachment: build the new version

Comment 6

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

Attachment: update the clang manifest. To be committed after try is done.

I have removed

browser/config/tooltool-manifests/macosx32/clang.manifest
and
browser/config/tooltool-manifests/macosx64/clang.manifest

since they are redundant with the releng manifest now that OS X has switched.

Comment 7

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

The try run is at

https://tbpl.mozilla.org/?tree=Try&rev=54927cd7de92

and a dummy run for comparison is at

https://tbpl.mozilla.org/?tree=Try&rev=ea57805e4262

Comment 8

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/5853a841672e

Comment 9

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/16089a551f63

Comment 10

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

https://hg.mozilla.org/integration/mozilla-inbound/rev/f2f738bd793c

Comment 11

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

https://hg.mozilla.org/integration/mozilla-inbound/rev/9ce23b2a9146

Comment 12

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

The results I got with diff-talos were:

dromaeo_css
MacOSX 10.6 (rev4)                | (2854.75909091, 16.9361015385) -> (2760.39916667, 15.0210672414) 1.0342x worse
MacOSX 10.7                       | (2802.51916667, 22.2119552504) -> (2731.13333333, 18.5358431148) 1.0261x worse

kraken
MacOSX 10.8                       | (2150.70909091, 2.0410165778)  -> (2168.07272727, 4.95538710857) 1.0081x worse

tdhtmlr_nochrome_paint
MacOSX 10.6 (rev4)                | (412.080090909, 0.684712591228) -> (410.255769231, 0.966515525145) 1.0044x better

I went ahead with the upgrade because:
* These are compiler changes. Since we run the same binary on 10.6, 10.7 and 10.8, any "real" regression is expected to show up on all of them.
* I could not reproduce it locally.
* This upgrade is blocking more important ones: 797564, 797525 and specially 793109.
* On the last upgrade the regressions that I was seeing comparing two try revisions did not materialize on m-i (no email).
* This is extremely easy to revert if the regressions turn out to be real.

Comment 13

[Ed Morley [:emorley]]

https://hg.mozilla.org/mozilla-central/rev/f2f738bd793c
https://hg.mozilla.org/mozilla-central/rev/9ce23b2a9146