1037542 - [Settings] - 100ms launch time regression

Status: RESOLVED INVALID

(Firefox OS Graveyard :: Performance, defect, P1)

Description

[Jon Hylands [:jhylands]]

On July 9, there was another 100ms launch regression in settings (on top of the one reported in bug 1036394).

Comment 1

[Jon Hylands [:jhylands]]

I've bisected this issue, and the following commit was identified as the root cause:

e222b3f9c5ab734ddc7f2285834ae227f811b5cd is the first bad commit
commit e222b3f9c5ab734ddc7f2285834ae227f811b5cd
Author: Ryan VanderMeulen <[e-mail]>
Date:   Wed Jul 9 14:24:37 2014 -0400

    Backed out changeset eba5d7aa80b0 (bug 997101) for causing frequent B2G layout/reftests/svg/sizing reftest failures.

RyanVM - can you comment on this? It seems odd that a backout would cause a performance regression...

Comment 2

[Ryan VanderMeulen [:RyanVM]]

Did bug 997101 improve perf such that backing it out "regressed" it again? Either way, better to be asking Kan-Ru than me since there's not much I'm going to be able to do to help here.

Comment 3

[Ryan VanderMeulen [:RyanVM]]

Taking a quick look at the summary of bug 997101, that certainly looks plausible - "Mediaquery in CSS may extend load time of app".

Comment 4

[Jon Hylands [:jhylands]]

Thanks for the info, Ryan - I've added a comment to that bug, and will continue to watch how it is resolved there.

Comment 5

[Jon Hylands [:jhylands]]

Attachment: Datazilla graph

This snapshot of datazilla shows what I believe to be the 100ms gain caused by the initial application of the fix in bug 997101, and then the subsequent 100ms loss caused by the backout.

The actual graph can be found here (for another couple weeks):

https://datazilla.mozilla.org/b2g/?branch=master&device=flame&range=30&test=cold_load_time&app_list=settings&app=settings&gaia_rev=2580a49ddeb99f4b&gecko_rev=d232b9ce6b96&plot=median

Comment 6

[Jon Hylands [:jhylands]]

I'm going to close this, since it appears to be a non-issue, and bug 997101 is not currently being worked on.