994240 - ~8% tresize regression on windows 8, april 4th on fx-team

Status: RESOLVED WONTFIX

(Testing :: Talos, defect)

Description

[Joel Maher ( :jmaher ) (UTC -8)]

graphs show some change:
http://graphs.mozilla.org/graph.html#tests=[[254,64,31],[254,132,31]]&sel=none&displayrange=7&datatype=running

did some retriggers to see the difference here:
https://tbpl.mozilla.org/?tree=Fx-Team&fromchange=a433a41e7a11&tochange=5da838032667&jobname=talos%20chromez

It ends up being related to this push:
https://hg.mozilla.org/integration/fx-team/pushloghtml?changeset=dcc436df9d67

I am not sure if this is expected or not, but we should have this change documented.

Comment 1

[Panos Astithas (he/him) [:past] (please ni?)]

How is that even possible? The files modified in those changesets aren't loaded during normal browsing.

Comment 2

[Joel Maher ( :jmaher ) (UTC -8)]

past, apologies, I was mixing up data on pgo vs non-pgo- thanks for questioning this, we had a large pile of regressions around that time and a lot of logs are missing as well- that only makes for a fun time.

Comment 3

[Panos Astithas (he/him) [:past] (please ni?)]

Ah, OK, because I began to doubt my own sanity :-)

Comment 4

[Victor Porof [:vporof][:vp]]

Me too....

Comment 5

[Joel Maher ( :jmaher ) (UTC -8)]

I did some retriggers- i see in win8 PGO mode, we have a noticeable regression:
https://tbpl.mozilla.org/?tree=Fx-Team&fromchange=a433a41e7a11&tochange=5da838032667&jobname=WINNT%206.2%20fx-team%20pgo%20talos%20chromez

it falls on:
https://hg.mozilla.org/integration/fx-team/rev/6b695883d4ff

from bug 986324.  That same bug had a TART regression (bug 993932) which was fixed yesterday.  Quite possible this might be fixed when we get more win8 PGO data.

Comment 6

[Joel Maher ( :jmaher ) (UTC -8)]

this has shown up on beta as the patch was landed:
https://hg.mozilla.org/releases/mozilla-beta/pushloghtml?changeset=3f2d6f68c415

Dao, can you look at this- the more we understand this the better we can make a decision to fix, backout, keep.

Comment 7

[Dão Gottwald [:dao]]

Mike, any idea what part of your patch could have caused this?

I don't think a straight backout is an option, given bug 986324's impact on accessibility.

Comment 8

[Mike de Boer [:mikedeboer]]

I'd say the transition properties would be the main culprit, but these were fixed by bug 993932.

Joel, does the tresize regression persist after that patch landed?

Comment 9

[Joel Maher ( :jmaher ) (UTC -8)]

The patch in bug 993032 landed on beta/aurora on the 14th, I have a few data points since then and we don't seem to have dropped our tresize numbers at all:
http://graphs.mozilla.org/graph.html#tests=[[254,64,31],[254,132,31],[254,52,31],[254,53,31]]&sel=1395146324401,1397738324401&displayrange=30&datatype=running

the patch landed on fx-team on the 10th there is no change in the numbers either (see above graph ^)

Quite possible we need to tweak something in the patch?

Comment 10

[Joel Maher ( :jmaher ) (UTC -8)]

this is still a problem on fx-team, do we have a plan to fix this?  If we don't do something soon we will need to fix this after we uplift 31 to Aurora.

Comment 11

[Mike de Boer [:mikedeboer]]

(In reply to Joel Maher (:jmaher) from comment #10)
> this is still a problem on fx-team, do we have a plan to fix this?  If we
> don't do something soon we will need to fix this after we uplift 31 to
> Aurora.

Joel, I don't have a plan to fix this. Bug 993932 was my last straw, other than that I can't think of something; the patch from bug 986324 is so small and insignificant that I really can't move anything around anymore, without making things look bad.

Comment 12

[Joel Maher ( :jmaher ) (UTC -8)]

Mike, thanks for the reply here.  Avi, any thoughts here?  Should we accept this and move onto other fires?  It looks like we will gain accessibility coverage for a perf hit on one platform, and one test.

Looking at the graph over a larger time period:
http://graphs.mozilla.org/graph.html#tests=[[254,64,31],[254,132,31]]&sel=none&displayrange=90&datatype=running

it looks like we have made significant wins overall.

Comment 13

[Avi Halachmi (:avih)]

(In reply to Joel Maher (:jmaher) from comment #12)
> it looks like we have made significant wins overall.

We did over the past few months, but not since this regression appeared.

However:

(In reply to Mike de Boer [:mikedeboer] from comment #11)
> Joel, I don't have a plan to fix this. Bug 993932 was my last straw, other
> than that I can't think of something; the patch from bug 986324 is so small
> and insignificant that I really can't move anything around anymore, without
> making things look bad.

IMO our mandate ends here (by our own definition).

We asked the owner's attention and opinion, Mike says he can't do more about it, significant related changes landed on trunk after this regression appeared, which makes it impractical to specifically address this regression.

Comment 14

[Sylvestre Ledru [:Sylvestre]]

Wontfix, untracking.