[CSS21] Overconstrained fixed table layout

Gérard pointed out to me that the spec for fixed table layout
seems to have some problems in the area of overconstrained
fixed table layout. The relevant test cases are:

http://test.csswg.org/suites/css2.1/nightly-unstable/html4/fixed-table-layout-025.htm
http://test.csswg.org/suites/css2.1/nightly-unstable/html4/fixed-table-layout-026.htm
http://test.csswg.org/suites/css2.1/nightly-unstable/html4/fixed-table-layout-027.htm
http://test.csswg.org/suites/css2.1/nightly-unstable/html4/fixed-table-layout-028.htm
http://test.csswg.org/suites/css2.1/nightly-unstable/html4/fixed-table-layout-029.htm
http://test.csswg.org/suites/css2.1/nightly-unstable/html4/fixed-table-layout-030.htm
http://test.csswg.org/suites/css2.1/nightly-unstable/html4/fixed-table-layout-031.htm

Specifically, the issue is how padding and borders factor into
the initial column width calculations. The algorithm says that
you only account for the width of cells with specified widths.
   http://www.w3.org/TR/CSS21/tables.html#fixed-table-layout

We don't have interop on this.

The problems I see so far are:
   * A column element's 'width' property sets the column's width.
     That's fine. But if it's width is 'auto', you take the 'width'
     of the first cell. But the 'width' typically does not include
     the cell's borders/padding -- shouldn't this be 'width' plus
     horizontal borders/padding?
   * When calculating a column width from the cell, the cell's
     parameters are ignored when its width is 'auto'. Shouldn't
     borders/padding factor into the table's width; and shouldn't
     they be subtracted from the remaining width before distributing
     to 'auto' columns?
   * Fixed table layout can result in table cells with negative
     widths. It's not specified what that means in terms of layout,
     since layout generally assumes that containing blocks are
     non-negative.

~fantasai

Received on Thursday, 10 January 2013 22:12:40 UTC