Bug 84513

Created attachment 138202 [details] Patch

Committed r114803: <http://trac.webkit.org/changeset/114803>

Reopen, because it broke zillion jsc tests and layout tests in debug mode on the GTK and on the Qt x86_64 bots. Maybe they are broken on Mac too, but unfortunately Mac debug bots aren't working properly now. - http://build.webkit.org/builders/GTK%20Linux%2064-bit%20Debug/builds/32645 - http://build.webkit.sed.hu/builders/x86-64%20Linux%20Qt%20Debug/builds/22391 Could you check and fix this regression please?

Hi guys, looks like all debug build (EFL, GTK, Win, etc) bots are sick after this change. Release bots are passing on these tests. http://build.webkit.org/builders/EFL%20Linux%20Debug/builds/379 ** Danger, Will Robinson! Danger! The following failures have been introduced: ecma/Expressions/11.2.2-1-n.js ecma/Expressions/11.2.2-10-n.js ecma/Expressions/11.2.2-7-n.js ecma/Expressions/11.2.2-8-n.js ecma/Expressions/11.2.2-9-n.js ecma/GlobalObject/15.1-1-n.js ecma/GlobalObject/15.1-2-n.js ecma/Math/15.8-2-n.js ecma/Math/15.8-3-n.js ecma_2/Exceptions/exception-001.js ecma_2/Exceptions/exception-002.js ecma_2/Exceptions/expression-005.js ecma_2/Exceptions/expression-006.js ecma_2/Exceptions/expression-012.js ecma_2/Exceptions/expression-013.js ecma_2/Exceptions/expression-014.js ecma_2/Exceptions/expression-019.js ecma_2/Exceptions/global-001.js ecma_2/Exceptions/global-002.js js1_2/function/regexparg-2-n.js js1_4/Regress/function-004-n.js

Hi Oliver, any idea about this issue?

I'm looking at this today -- the assertion that's firing is *very* bad. It's actually showing up on a number of other sites in mac (just not on the test suite :-/ ). This is currently my #1 priority.

Thanks for getting back on this - if it's taking longer than a day - do you think we could roll this out until the sites are fixed that hit this assertion? Reason is, this hits the "20 crashes to bailout" threshold on our (EFL Linux Debug) buildbot. Similar for GTK Debug Buildbot. So we can't use the results.

Ah ha, if LLINT isn't enabled, i get crashes running webkit tests on mac. I assume effected builds have llint disabled. Investigating some more.

It looks like there are a few places in the baseline jit that mix-and-match the current callframe and the caller's callframe. Fixing.

Running tests on my patch, then some tidying up and should be good. Then we'll just be stuck with the horrible bug that i originally added these assertions to catch.

(In reply to comment #10) > Running tests on my patch, then some tidying up and should be good. Then we'll just be stuck with the horrible bug that i originally added these assertions to catch. I'm still online for a couple of hours if you need a second opinion (i.e. run your patch/fix on other port).

Debug Buildbots are back to work after Oliver removed one assertion: http://trac.webkit.org/changeset/114953

(In reply to comment #10) > Running tests on my patch, then some tidying up and should be good. Then we'll just be stuck with the horrible bug that i originally added these assertions to catch. Thanks Oliver! Bots are bleeding less now :) This bug can be closed now, maybe?