REGRESSION (r61619): Memory corruption in open-source ICO decoder
BMPImageReader.cpp erroneously accesses |m_parent| when setSize() fails. setSize() has already called setFailed(), which has deleted |this|, thus we shouldn't access |m_parent| (and don't need to).
Created attachment 59572 [details] patch v1 Fixes the corruption and one other technically-wrong place I noticed. This adds a regression .ico to an existing LayoutTest, unfortunately I can't actually update expected results at the moment, so I'm going to need those from somewhere.
Comment on attachment 59572 [details] patch v1 ok
Fixed in r61788. I'll land the updated test expectations once the bots have them.
http://trac.webkit.org/changeset/61800 might have broken SnowLeopard Intel Release (Tests)
Gtk is still broken from this: http://build.webkit.org/results/GTK%20Linux%2032-bit%20Debug/r61820%20(7197)/fast/images/icon-decoding-pretty-diff.html
Hopefully fixed in http://trac.webkit.org/changeset/61821
The reason I didn't update GTK is because it matches Chromium rather than Safari and it wasn't clear to me that those two text dumps would be the same. (And I couldn't get a Chromium dump at the time.) I'm not lazy! I was just gone. I got two other people to help take care of this before I had to go.