803239 - Huge regression on eideticker of cnn/reddit.com etc.

Status: RESOLVED DUPLICATE of bug 770144

(Core :: Graphics, defect)

Description

[Jeff Muizelaar [:jrmuizel]]

Here's an example:
http://eideticker.wrla.ch/#/samsung-gn/cnn/checkerboard

This happened around the 15th but there's a bunch of missing eideticker data so it's hard to know.

Comment 1

[Jeff Muizelaar [:jrmuizel]]

Looking at the regression it seems like this is related to a bunch of stuff being transparent that wasn't before.

Comment 2

[Jeff Muizelaar [:jrmuizel]]

After the regression a bunch of time is being spent in 8888_8888_etc that wasn't there before

Comment 3

[William Lachance (:wlach)]

Yeah, this is a really big issue. I'm going to work on bug 772948 so we can start putting backdated information into the dashboard. I was stuck on how to do this properly before but I think I've figured it out. Because of the unreliability of the dashboard, this keeps on coming up again and again, so it's probably best to get this fixed now for the future.

Comment 4

[Jeff Muizelaar [:jrmuizel]]

I believe this might be a dup of 795674

Comment 5

[Benoit Girard (:BenWa)]

The regression reduced today but it's still not at base levels.

Comment 6

[William Lachance (:wlach)]

Now that bug 772948, I'll try generating some backdated entries for CNN and get them uploaded to the dashboard.

Comment 7

[William Lachance (:wlach)]

Backdated datapoints are in:

http://eideticker.wrla.ch/#/samsung-gn/cnn/checkerboard

Looks like the regression first appeared on the Nightly of the 13th.

Comment 8

[Chris Lord [:cwiiis]]

A regression window (even a rough one) would be good.

Comment 9

[u279076]

kbrosnan, have you had a chance to look at this yet? Is this Mobile only?

Comment 10

[Alex Keybl [:akeybl]]

From the latest eideticker data, this isn't a critical release issue any longer. Found on 20, fixed on 20 (see https://wiki.mozilla.org/RapidRelease/Calendar and comment 7). Please re-nominate if there's any reason to believe this is unresolved on Firefox 19.

Comment 11

[Brad Lassey [:blassey] (use needinfo?)]

William, how can we get a regression window from eideticker?

Comment 12

[William Lachance (:wlach)]

(In reply to Brad Lassey [:blassey] from comment #11)
> William, how can we get a regression window from eideticker?

I'd have to do one manually. Unfortunately it's kind of a pain to do this with eideticker on anything greater than a daily granularity.

I believe this regression had to do with some DLBI work which landed around that time for B2G. IIRC :mattwoodrow fixed the fallout of that to the best of his ability. I'm not sure if a finer-grained regression range would tell us anything that we don't already know.

Comment 13

[William Lachance (:wlach)]

I think most of the work I was talking about was done here (scroll down):

https://bugzilla.mozilla.org/show_bug.cgi?id=770144

Comment 14

[Brad Lassey [:blassey] (use needinfo?)]

William, so is this a real regression? should it be tracking 19?

Comment 15

[William Lachance (:wlach)]

(In reply to Brad Lassey [:blassey] from comment #14)
> William, so is this a real regression? should it be tracking 19?

So my understanding was that Matt Woodrow addressed the issues in the regression as best he could, and that the remaining performance regression (note that the performance immediately after the 15th) was a deliberate trade off we decided on for B2G (I remember cjones telling me that the new behaviour improves the performance in B2G-style apps at the expense of things like the CNN benchmark). 

So, no, I don't think this is a regression. And moreover, it's essentially a duplicate of bug 770144.

Comment 16

[Erin Lancaster [:elan]]

Is this still useful to track?