534981 - Regression: Pages take longer to show initially than they did with Fennec Beta 5

Status: NEW

(Core :: Widget: Gtk, defect)

Description

[Stuart Parmenter]

If you do a side-by-side comparison of beta 5 versus trunk, you'll see that beta 5 shows the page content much faster than we do now.  This may be a platform problem, but having trouble identifying it.

Comment 1

[Aakash Desai [:aakashd]]

The regression seems to be on 12/16's nightly build. Progressive rendering is still going on, but performance has gone down quite a bit. Here's a list of bugs set to resolved:fixed during that time period

https://bugzilla.mozilla.org/buglist.cgi?chfieldto=2009-12-16;query_format=advanced;chfield=bug_status;chfieldfrom=2009-12-15;chfieldvalue=resolved;resolution=FIXED;product=Fennec

Comment 2

[Aakash Desai [:aakashd]]

There's two bugs here:

1. The regression range found on 12/16 was for the degredation of performance of chrome during page load.


2. I found another regression range between 12/13-12/16 for the long wait time to have content shown on the screen initially during page load.

Comment 3

[Aakash Desai [:aakashd]]

Bug https://bugzilla.mozilla.org/show_bug.cgi?id=535648 was created for the issue associated with #1 in comment #2

Comment 4

[Benjamin Stover (:stechz)]

Attachment: First mozafterpaint log

These numbers correspond with the regression we've seen for loading.  I tried nytimes with the permutations {xulrunner,fennec} X {b5 version, tip version}.

I am rather convinced this problem is somewhere in platform.

Comment 5

[Stuart Parmenter]

This is a regression from 509278

Comment 6

[Stuart Parmenter]

Attachment: fix?

Comment 7

[Stuart Parmenter]

Comment on attachment 418952 [details] [diff] [review]
fix?

I wonder if we should just remove this code entirely for all platforms?

Comment 8

[Karl Tomlinson (:karlt)]

<stuart> basically the problem is that compared with a build from before,
         there is a 5-15 second delay in initial paint

Initial paint of content IIUC.
What is shown before the initial paint?
Previous content?
Or fill content provided by the front end?
Or the background of the GTK theme?
Or garbage?

Comment 9

[Karl Tomlinson (:karlt)]

Comment on attachment 418296 [details]
First mozafterpaint log

>latest nytimes.com
>    (begin) : on location change 
>    655 : mozafterpaint 

Time from js window location change to first MozAfterPaint event?
Is the unit ms?
Is there only one MozAfterPaint event?
Is this for invalidation of the frontend document or of the content document?

>    (begin) : on location change 
>    124 : mozafterpaint

Is this another sample (of the same test)?

Comment 10

[Karl Tomlinson (:karlt)]

Comment on attachment 418952 [details] [diff] [review]
fix?

I'm OK if you'd like to land this as a workaround for now, given that we don't really understand what is happening and we won't be worse off than before the change from bug 509278.

However, this is not ideal so I don't want to do this on all platforms, and
I'd still like to know the answers to the above questions as they may help
understand how to fix this properly.

AFAIK, the only effect from Show()ing and already Show()n window is to
invalidate the whole window.

* This will cause unnecessary painting.

* It suggests that a well-placed Invalidate() (or perhaps Update()) would be
  enough and possibly cause content to appear earlier than happens with the
  extra Show(), whose caller was not intending to invalidate (so may not have
  called this at the appropriate time).

Comment 11

[:Gavin Sharp [email: [e-mail]]]

I've pushed this to 1.9.2. I think we can just leave this bug open for a better trunk fix.

https://hg.mozilla.org/releases/mozilla-1.9.2/rev/e1eeb7719442

Comment 12

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Perhaps it's better to push it to trunk too, just so it doesn't get forgotten and then produce an unexplained regression when switching to a later branch?

Comment 13

[Stuart Parmenter]

handing this over to karl to find best solution for the regression

Comment 14

[Aakash Desai [:aakashd]]

Is this blocking 1.0 now, stuart?

Comment 15

[Mark Finkle (:mfinkle) (use needinfo?)]

(In reply to comment #14)
> Is this blocking 1.0 now, stuart?

No, and it's not blocking 1.1 either