206516 - change display property to none and block breaks table cell rendering

Status: VERIFIED FIXED

(Core :: Layout: Tables, defect)

Description

[Osvaldo Santana Neto]

User-Agent:       Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.3b) Gecko/20030317
Build Identifier: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.3b) Gecko/20030317

Try check/uncheck for many times the radioboxes in page:
http://r2d2.rantac.com.br/ispman/bug.html and you will see the problem.

Reproducible: Always

Steps to Reproduce:
1. enter in http://r2d2.rantac.com.br/ispman/bug.html
2. check/uncheck radioboxes in page (many times)
3. done

Actual Results:  
The page render crashes (table column)

Expected Results:  
Disable/Enable table columns visibility.

Comment 1

[Boris Zbarsky [:bzbarsky]]

The page is setting table elements (tr and td) to be display:block.  We have
some existing issues with this (not disposing of pseudo-frames properly), but
even if those were fixed the page layout would still be "broken" (because CSS
table display types are completely broken in IE and the page expects the IE
behavior).  Using table-row and table-cell for the display, as appropriate, will
work in Mozilla and other CSS2-compliant browsers (but not in CSS1 browsers like
IE/Windows).

Over to tables to dup to the bug on pseudo-frame issues.

Comment 2

[Madhur Bhatia]

checked the url on winXP and Linux on 2003-05-28trunk builds 
WFM!

Osvaldo, please check this again in the latest build and if u do get a crash,
provide the talkback crash id number or the stack signature.

Comment 3

[Bernd]

testcase has gone :-(

Comment 4

[Osvaldo Santana Neto]

Sorry for incovenience.

The testcase has moved to:
http://web.rantac.com.br/ab/bug.html

Comment 5

[Bernd]

Osvaldo could you please attach the testcase to the bug just hit the Create a
New Attachment link and follow the instruction. I suspect however that this bug
is a dupe of 77019 which hopefully will be fixed in a near future

Comment 6

[Osvaldo Santana Neto]

Attachment: Testcase

Comment 7

[Bernd]

Osvaldo, is your testcase just a html page? attach the html instead of the
archive please. I can't open the archive.

Comment 8

[Osvaldo Santana Neto]

Attachment: testcase html file

Comment 9

[Osvaldo Santana Neto]

Attachment: testcase css file

Comment 10

[Osvaldo Santana Neto]

Attachment: testcase js file

Comment 11

[Osvaldo Santana Neto]

the testcase is:
1 .html
1 .css
1 .js

I will attach the 3 items separated...

Comment 12

[Boris Zbarsky [:bzbarsky]]

Attachment: HTML file in usable form (pointing to the other 2)

Comment 13

[Boris Zbarsky [:bzbarsky]]

Resummarizing to make it clear what the bug is about.  Testcase never uses
"visibility" and this has nothing to do with bug 77019

Comment 14

[Daniel Santos]

Attachment: reproduces bug

Comment 15

[Daniel Santos]

Windows XP and 2000, current Mozilla and Foxfire. I could not get a crash, but I
get the described layout problems. Can we move this to confirmed? (attached
another reproduction case)

Comment 16

[Boris Zbarsky [:bzbarsky]]

We can't move it to confirmed, since it's a duplicate (at least the layout
problems in your testcase are duplicates).  Note the DUPEME keyword.

Also note that even if the layout problems were fixed your testcase would still
be "broken" because putting both of those cells in the first column when that
"show" function runs is correct per the CSS spec.

Comment 17

[Bernd]

Attachment: reduced testcase

Comment 18

[Bernd]

Attachment: patch

This fixes a problem in block layout, but leaves the frame construction issue.
The block problem is that the block caches the mAscent and if there are no
lines inside the block the cached value is reported to the containing table
cell.

Comment 19

[Boris Zbarsky [:bzbarsky]]

Bernd, you want to spin that off into a separate bug, or check it in under this one?

Comment 20

[Bernd]

Just the block reflow issue here and 162063 for the frame issue. The fix here
will not complete give the space back as the remaining anonymous table cells
will have their cellpadding cellspacing still applied.

Comment 21

[Bernd]

fix checked in the reminder is bug 162063

Comment 22

[Mathieu Parent]

Maybe this should be done on branch

the patch seems to be simple (any regression) ?

REOPEN ?

Comment 23

[Boris Zbarsky [:bzbarsky]]

> Maybe this should be done on branch

If you feel that way, request approval for the patch (see the flags on the patch
itself) and explain in the approval request why this is needed and why it's very
safe.

> the patch seems to be simple (any regression) ?

We probably won't know for at least a few weeks.  Unless you have a source of
knowledge I'm not aware of.

> REOPEN ?

Why?  It's fixed.  Whatever happens on branch doesn't affect the resolution,
which tracks things on trunk.

Not a branch blocker in any case, unless this is a very common problem out
there.  If it is, please point out the sites broken by it.

Comment 24

[Bernd]

The reduced testcase shows red again. As I did not checkin the test into the layout regression tests I can not pinpoint when this went bad again.

Comment 25

[Daniel Veditz [:dveditz]]

"DUPEME" seems obsolete if work's going on here.

Comment 26

[Bernd]

the regression in the reduced testcase has been fixed by the reflow branch

Comment 27

[Ryan VanderMeulen [:RyanVM]]

Attachment: Reftests

These are based on the last testcase posted to this bug. I'm a bit concerned that this test will have to be marked as random, though. I've noticed that when I load the test, if I refresh the page quickly, I will sometimes see a very brief flash or red under the green cell. Depending on when the reftest takes the screenshot, that might pose a problem.

Comment 28

[Ryan VanderMeulen [:RyanVM]]

Attachment: Reftests round 2

Whoops, forgot to remove the table background color from the reference

Comment 29

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

Comment on attachment 257572 [details] [diff] [review]
Reftests round 2

Checked in after testing in pre- and post-reflow-branch builds.