Bug 21280 – AXTables on page are probably not tables

• Component: Accessibility
• Status: RESOLVED FIXED

[chris fleizach]

Request is to improve table detection. There are two non-table tables on

http://www.msnbc.msn.com/id/26760673/#storyContinued

[chris fleizach]

Created attachment 23986 [details]
Patch to make AXTable detection more robust

[Darin Adler]

Comment on attachment 23986 [details]
Patch to make AXTable detection more robust

I think I would use the word "heuristic" somewhere here to describe what we're doing.

To me it seems overkill to look at all the cells in a table. Once you've found borders or colors on 10 different cells, I'm not sure what the value is of examining 100 others to determine that it's "more than half of the cells" that have borders or colors. I think "two rows" was probably too limited, but "every single cell" is probably too open ended. In particular we want good performance on absurdly huge tables, and querying every cell of such tables is obviously not helpful.

+            // considered a bordered cell. Try the cell borders and the style's borders

I don't see code corresponding to the style's borders.

The check for background differences isn't quite right -- it doesn't account for transparent colors.

+    unsigned neededCellCount = (validCellCount >> 1);

It's bad style to do ">> 1" when you really mean "/ 2" and doesn't make the code any faster. We also don't usually put parentheses around this operation.

I'm going to say r=me, but I think it's worth pondering further whether it's sensible to iterate the entire table even for pathologically huge tables.

[chris fleizach]

that's reasonable. i can put in a limit to how many cells are checked at max. Let's say 10 cells

[chris fleizach]

http://trac.webkit.org/changeset/37216