After applying the patch for bug 4547, the following layout tests fail: dom/html/level2/html/HTMLTableElement35 dom/html/level2/html/HTMLTableElement38 dom/html/level2/html/HTMLTableRowElement16 dom/html/level2/html/HTMLTableRowElement19 dom/html/level2/html/HTMLTableSectionElement26 dom/html/level2/html/HTMLTableSectionElement29 fast/js/date-big-setmonth fast/js/date-parse-date fast/dom/quadraticCurveTo (draws a sine with ~1px amplitude along the axis) I'm putting them all together in one bug because it's likely that they have a common root cause (something broken in the way JS does math), and because I'm lazy.
Created attachment 3942 [details] Fix by rolling out simple_number.h changes Because we have JavaScriptCore test failures on TOT, I can't be sure that this patch is the right solution, but it fixes all the layout tests without introducing additional regressions to the JavaScriptCore tests.
Comment on attachment 3942 [details] Fix by rolling out simple_number.h changes I think we should land this patch, land a fix for the other JSC regressions, and then go back and redo the int/long modifications to simple_number.h, once we have a working baseline of tests.
Comment on attachment 3942 [details] Fix by rolling out simple_number.h changes Given that this fixes the bug, r=me This change should be revisited however.
I'm marking this verified even though date-big-setmonth is still failing for me. I seem to remember it failing even before the long->int move.
date-big-setmonth was failing because it's not written to be TZ independent. I'll fix that.
Opened bug 5061 about date-big-setmonth