Bug 8848 – TFOOT borders are copied to THEAD and TBODY

• Component: CSS
• Status: RESOLVED FIXED

[Chris Bentley]

In Safari Version 2.0.3 (417.9.2), when the CSS rule "border-collapse:collapse" is applied to a TABLE and also CSS borders applied to the TFOOT element or its descendants, then the TFOOT borders are unexpectedly copied to the THEAD and TBODY elements. 

In the examples at http://code.clientside.net.au/safari/tablefoot.html only the TFOOT element should have red borders. Firefox 1.5 Mac renders as I expected - Opera 8.5 Mac renders in a similar way to Safari. IE 6 Windows cannot apply the borders directly to the TFOOT element but renders as expected when borders are applied to descendent cells

A work-around for this bug is to apply a border with the same width to the affected collateral elements.  This is not always desirable.

Possibly related to http://bugzilla.opendarwin.org/show_bug.cgi?id=3699

A fix is included in the patch for bug 6838.

(In reply to comment #1)
> A fix is included in the patch for bug 6838.
> 

I was wrong.

...that patch included half the fix in cellAbove/cellBelow.

Created attachment 8320 [details]
Patch w/change log and layout test

I'm not sure the empty-section-skipping behavior for cellAbove/Below is correct (the patch maintains the current behavior).

[Dave Hyatt]

Comment on attachment 8320 [details]
Patch w/change log and layout test

r=me

Created attachment 8552 [details]
Updated test results

A bunch of tests were affected by this patch.

[Maciej Stachowiak]

Comment on attachment 8552 [details]
Updated test results

r=me

[David Kilzer (:ddkilzer)]

Reopening for manual review of test results (Attachment 8552 [details]).

I'm guessing that the updated test results for this patch are okay, but in reviewing individual tests, I'm not comfortable committing the changes.  (Perhaps I should open a separate bug?)

It looks like there are numerous tests giving a false-positive results!  For example, look at this test in Firefox 1.5.0.3 and then WebKit ToT (I'm using a locally-built r14643):

LayoutTests/css2.1/t0801-c412-hz-box-00-b-a.html

This test has scrollbars drawn when it should not (again, compare with Firefox 1.5.0.3):

LayoutTests/fast/flexbox/016.html

There is also inconsistency between the rendering of this test in Firefox 1.5.0.3 and WebKit ToT:

LayoutTests/tables/mozilla/marvin/backgr_layers-opacity.html

Am I insane, or do we need to manually review all of the test results to restore confidence in them?

(In reply to comment #8)
> I'm guessing that the updated test results for this patch are okay, but in
> reviewing individual tests, I'm not comfortable committing the changes. 

Of the tests in the "Updated test results" patch, am I correct that the following is the only one you're uncomfortable with?

> LayoutTests/tables/mozilla/marvin/backgr_layers-opacity.html

The difference seems to be that WebKit doesn't paint the cell background in the border box. This probably warrants a separate bug (fixing bug 6838 only made the problem more apparent, but didn't create it).

[David Kilzer (:ddkilzer)]

(In reply to comment #9)
> (In reply to comment #8)
> > I'm guessing that the updated test results for this patch are okay, but in
> > reviewing individual tests, I'm not comfortable committing the changes. 
> 
> Of the tests in the "Updated test results" patch, am I correct that the
> following is the only one you're uncomfortable with?
> 
> > LayoutTests/tables/mozilla/marvin/backgr_layers-opacity.html

No, all of the tests I listed seem to be broken.  LayoutTests/css2.1/t0801-c412-hz-box-00-b-a.html clearly doesn't work.  Again, I doubt it was the changes to this bug that caused them, though.

Have these tests been broken all along (and just aren't labeled that way)?  I was assuming that every test that wasn't clearly labeled as "broken" (such as some of the Mozilla table tests) was working, but I probably haven't looked at the test results this closely before.

Is this not a valid assumption?

(In reply to comment #10)
> (In reply to comment #9)
> > (In reply to comment #8)
> > Of the tests in the "Updated test results" patch, am I correct that the
> > following is the only one you're uncomfortable with?
> > 
> > > LayoutTests/tables/mozilla/marvin/backgr_layers-opacity.html
> 
> No, all of the tests I listed seem to be broken. 

But only that one is in the patch attached to this bug. Never mind...

> Have these tests been broken all along (and just aren't labeled that way)?

One way to find out is to examine the svn history of the expected result, or try the test in a really old build.

> I was assuming that every test that wasn't clearly labeled as "broken" (such as
> some of the Mozilla table tests) was working
[...]
> Is this not a valid assumption?

I'm afraid it isn't (I know about several tests in tables/mozilla/, I don't know about css1 and css2.1). It might be valid for the tests in fast/.

[David Kilzer (:ddkilzer)]

(In reply to comment #11)
> (In reply to comment #10)
> > I was assuming that every test that wasn't clearly labeled as "broken" (such as
> > some of the Mozilla table tests) was working
> [...]
> > Is this not a valid assumption?
> 
> I'm afraid it isn't (I know about several tests in tables/mozilla/, I don't
> know about css1 and css2.1). It might be valid for the tests in fast/.

Ahh, sorry.  In that case, I'll land the "Updated test results" when I get home from work unless someone beats me to it.