Bug 4874 – Table with absolute bottom positioning is positioned incorrectly

• Status: RESOLVED
• Resolution: FIXED
• Component: Layout and Rendering

[Arthur Langereis]

See the URL. Giving a table the following styles:
position: absolute; bottom: 0;

will result in its _top_ coordinate being aligned with its containing element's bottom, causing it to appear 
below the P in this example.

Other browsers (IE, FF, Opera) place the P and the table at the same height, Safari doesn't. Tested in 2.0 
and ToT from around Sep. 1.

Created attachment 3802 [details]
testcase

Extended testcase

Created attachment 3803 [details]
testcase

previously uploaded the wrong file

Created attachment 3804 [details]
suggested patch

[Mark Rowe (bdash)]

Confirmed that the bug occurs with WebKit 412.7 and ToT. Haven't tried with Mitz's patch.

[Mark Rowe (bdash)]

I've tested with the patch and it resolves the problem and leaves all layout tests passing.

[Maciej Stachowiak]

Looks like a good fix to me.

[Dave Hyatt]

This does not seem right to me.  I need to mull this over.

[Dave Hyatt]

This piece of the patch confused me in particular:

-    if (isTable() && height.isVariable())
+    if (isTable() && (height.isVariable() || isPositioned()))
         // Height is never unsolved for tables. "auto" means shrink to fit.  Use our
         // height instead.

Because of the change in RenderTable, calcAbsoluteVerticalValues() gets called for a position table *after* 
height calculation, so it should not force a smaller height. If it does, the "Short table" in the testcase gets 
squeezed.

meant to say "a positioned table" of course

Comment on attachment 3804 [details]
suggested patch

I can't see what's wrong with the patch as it is. See my previous comment.

[Dave Hyatt]

Comment on attachment 3804 [details]
suggested patch

calcAbsoluteVerticalValue is only called for positioned elements.  Therefore as
far as I can tell, this line:

+    if (isTable() && (height.isVariable() || isPositioned()))

just reduces to

+ if (isTable())

I am not yet convinced that positioned tables under all circumstances are
supposed to shrink-to-fit.

[Sam Weinig]

Created attachment 8957 [details]
updated patch

This is an updated version of mitz's patch, applying hyatt's suggestions and updating it to the current code base.  It still causes no regressions in the test suite. 

Hyatt, have you come to a conclusion yet?

[Dave Hyatt]

Comment on attachment 8957 [details]
updated patch

I'll say r=me.  This patch still makes me nervous, since I'm not convinced we have enough tests yet to understand what the correct behavior is in all cases.

[Dave Hyatt]

Whoever commits this be sure to run the layout tests, since the mozilla table test suite was added to our regression suite since this patch was last worked on.

[David Kilzer (:ddkilzer)]

Committed revision 15105.

I ran all of the layout tests, and there were no failures.

*** Bug 7699 has been marked as a duplicate of this bug. ***