170579 - Entire elements disappear from display when using alternate stylesheets and doing some combinations of load/reload

Status: RESOLVED WORKSFORME

(Core :: Layout, defect)

Description

[Tomas Ögren]

User-Agent:       Mozilla/5.0 (X11; U; SunOS sun4u; en-US; rv:1.1) Gecko/20020910
Build Identifier: Mozilla/5.0 (X11; U; SunOS sun4u; en-US; rv:1.1) Gecko/20020910

This is a very weird bug that requires a bunch of things to be fulfilled to
trigger the bug. Different versions display the bug in different ways.

The page http://www.acc.umu.se/~stric/tmp/moz/altss/test.html consists of some
text, a big table with a bunch of tables and inside them a large chunk of <div>s
with for instance title attributes. After the large table there is some text.
On some platforms (Solaris/sparc, using both 1.0 and 1.1) for instance, when
loading the page for the first time (or hitting enter in the url bar) only "Some
very interesting data here below table" is shown.

Here's a table of what is missing on different versions using different methods:
           Visit url   Paste url    Ctrl-L<ret>   Ctrl-R     Ctrl-Shift-R
Lnx/x86 1: nothing     above+table  above+table   nothing    above+table[1]
Lnx/x86 2: above+table above+table  above+table   nothing    above+table[2]
Lnx/x86 3: above+table above+table  table         nothing    above+table
Lnx/x86 4: above+table above+table  above+table   nothing    above+table
Sol/sprc1: above       above+table  above+table   nothing    above+table
Sol/sprc2: above+table above+table  above+table   nothing    above+table

[1] Only reproduced once..
[2] all the time...
Visit url = Visit the url by clicking on a link, cold cache
Paste url = Select url in clipboard, middleclick in browser window to go there
The key combos should be obvious.

Lnx/x86 1 is mozilla 1.0 on a P4/2.5GHz displaying on a Sun SS5
Lnx/x86 2 is mozilla 1.2a on a P4/2.5GHz displaying on a Sun SS5
Lnx/x86 3 is mozilla latest nightly (2002092404) on a dual Athlon MP 1800+
displaying on a IBM RS/6000 42T.
Lnx/x86 4 is mozilla 1.1b on some PC running Linux, displaying locally
Sol/sprc1 is mozilla 1.0 on a Sun V280 (2x750MHz) displaying on a Sun SS5
Sol/sprc2 is mozilla 1.1 on a Sun V280 (2x750MHz) displaying on a Sun SS5

Some elements are not gone on all tries, but most of the time.
There are some key issues here required to make this web page buggy:
* Size of the document, if I remove too much (doesn't matter what) it stops buggying
* The fact that I have an (almost empty) CSS as main CSS and the same as an
alternate CSS with a different (almost empty) CSS between.

The CSS files contain "a { }", having other elements makes it buggy as well. But
having an empty file causes the bug to stop appearing.
If I change place so the CSS are "maincss=foo1, alternate1=foo1,
alternate2=foo2" it stops bugging, but with "maincss=foo1, alternate1=foo2,
alternate2=foo1".

At first when I tried this, only the stuff above the table got lost. When making
the testcase smaller I got to the point where the entire table got lost. Trying
to make it smaller from there stopped the bug from appearing.



Reproducible: Always

Steps to Reproduce:
1. Visit the URL
2. Either visit the url, paste it, hit ctrl-L + ret, ctrl-r or ctrl-shift-r
3. Watch parts of the page disappear

Actual Results:  
Elements disappeared from the page

Expected Results:  
Nothing should have disappeared.

Given that it exists on different platforms and appear differently on different
machines I'd guess it's timing related somehow.. But also other stuff..
I don't like bugs like this, they're hard to debug :/

Comment 1

[Josh Birnbaum]

Seeing the problem with 1.2a on XP.  For me, the page displays without the text
above the table.  Pressing Back and then Forward then shows the page without the
table.

The HTML and CSS passes validation.

Comment 2

[Boris Zbarsky [:bzbarsky]]

Attachment: css_dark.css

Comment 3

[Boris Zbarsky [:bzbarsky]]

Attachment: testcase

Comment 4

[Boris Zbarsky [:bzbarsky]]

So... the DOM is correct.  The huge chunk of whitespace in the middle is needed
to reproduce the layout problem -- it seems to depend on how things get chunked
up in necko or the parser.  All three sheets are needed; first and third must
have the same url.  Middle one must have _different_ url (it can be the same
sheet, however).  I can't look at the frame model here because I do not have
viewer...

David, any ideas?

Comment 5

[Boris Zbarsky [:bzbarsky]]

Argh.  and now 30 seconds later the testcase no longer reproduces the problem
for me... :(  original page still does, however.

Comment 6

[Boris Zbarsky [:bzbarsky]]

I can't seem to reproduce this on the url anymore... is this a problem in
current builds?

Comment 7

[Josh Birnbaum]

This is WFM now, tested with 1.4a on WinXP (I had seen the problem before with
1.2a on XP)

Comment 8

[Christopher Hoess (gone)]

Per report, resolving WFM.