Closed Bug 990163 Opened 11 years ago Closed 11 years ago

6% tart regression on linux 32/64 seen on fx-team due to bug 477948

Categories

(Firefox :: Theme, defect)

Product:
Firefox
Component:
Theme
Platform:
x86_64
Linux
Type:
defect
Priority:
Not set
Severity:
normal

Tracking

()

Status:
RESOLVED WONTFIX

People

(Reporter: jmaher, Unassigned)

References

Details

(Keywords: perf, regression, Whiteboard: [talos_regression])

in bug 990084 we have a tresize regression from the same change. This is a bug for tart specifically: http://graphs.mozilla.org/graph.html#tests=[[293,132,33],[293,132,31],[293,132,37],[293,132,35],[293,132,25],[293,64,24]]&sel=1395679908087,1396284708087&displayrange=7&datatype=running you can see linux 32 and 64 regression, but not windows or osx (on the 28th, please don't focus on the future dates) On datazilla we can see what specific tests regressed here: https://datazilla.mozilla.org/?start=1395679254&stop=1396284054&product=Firefox&repository=Fx-Team-Non-PGO&os=linux&os_version=Ubuntu%2012.04&test=tart&graph_search=7fe4e2d84d76&x86_64=false&project=talos icon-close-DPI1.all.TART icon-close-DPI1.half.TART icon-close-DPI2.all.TART icon-close-DPI2.half.TART icon-open-DPI1.all.TART icon-open-DPI2.all.TART icon-open-DPI2.error.TART icon-open-DPI2.half.TART iconFade-close-DPI2.all.TART iconFade-close-DPI2.half.TART newtab-open-preload-no.error.TART simple-close-DPI1.all.TART simple-close-DPI1.half.TART simple-open-DPI1.all.TART simple-open-DPI1.error.TART
Blocks: 477948
Who's going to do the analysis from bug 477948 comment 38? Mike?
Flags: needinfo?(mdeboer)
Oops, looks like that happened in bug 477948 comment 41 and bug 477948 comment 43. (A 6% regression isn't something I'd expect to see experimentally anyways, so not sure I see the value in it really, but it sounds like there's not much of an option here anyhow.)
Status: NEW → RESOLVED
Closed: 11 years ago
Flags: needinfo?(mdeboer)
Resolution: --- → WONTFIX
I am fine with a wontfix, but I don't really agree with "it doesn't appear slower when looking at it in person". We have performance tests for a reason, and tart specifically is designed to ensure we have a benchmark for the tab animations. 6% seems like a decent amount to be worried about with no justification of 'this new feature improves user experience and security'. We have ignored 2% regressions in the past, likewise 10% regressions, but for many alerts of <6% regression we have investigated time and fixed them. Avi, do you have any concerns?
Flags: needinfo?(avihpit)
My concerns were that I'd hope we won't have this regression, but in practice we apparently do. I'm OK to let this go (noted on bug 477948 comment 42) after asking few people with linux to play with it and report back if they notice a regression between the two builds (from bug bug 477948 comment 39). This is an important feature to have, and preventing it from being released together with the other platform's Australis debut over an invisible 6% regression seems like an inappropriate measure.
Flags: needinfo?(avihpit)
thanks Avi and Gavin for looking into this bug and agreeing on a wontfix.
Note that bug 477948 is blocked also by bug 990084 (10% tresize regression from the same patch), and it's quite likely that the same change would fix/improve both tresize and tart. I haven't yet given up on bug 990084, so hopefully this tart regression will improve once (hopefully) we're able to improve 990084.
You need to log in before you can comment on or make changes to this bug.