DumpRenderTree: Windows Cairo 'Mostly' Good

I recently found and corrected the cause of the hated scrolling bug. A question by Jonghoon Kim about the inconsistent use of SetGraphicsMode(bitmapDC, GM_ADVANCED); prompted me to review the handling of graphics context between Windows and Cairo. This led to the discovery that the Windows HDC and the Cairo cairo_t contexts were not being kept in sync in all cases, and so various transformations were being inconsistently applied, resulting in the strange behavior that prompted me to file the bug. Now that the only major rendering discrepancy was resolved, it's time to start solidifying the Cairo port.

I finally got around to updating DumpRenderTree so that it can run the enormous suite of layout tests included with the WebKit sources. I was pretty sure that things would be fairly broken, but I am pretty happy that DumpRenderTree decided that the Windows Cairo build desired a grade of "D".

Here are the test results:

7907.74s total testing time

5598 test cases (54%) succeeded

4597 test cases (44%) had incorrect layout

19 test cases (<1%)>

4 test cases (<1%)>

409 test cases (4%) had stderr output

Clearly there is a lot of room for improvement. However, as you can see in the screen grab above, many of the differences are a matter of one or two pixels.

I'll probably target the crashes first.