Bug 66845 – ErrorInstance::create sometimes has two heap object constructions in flight at once

• Status: RESOLVED
• Resolution: FIXED
• Component: JavaScriptCore

[Filip Pizlo]

Under the newly refactored heap object creation scheme, only one heap object can have its C++ constructor in flight at a time.  This ensures that a GC cannot be fired while some object has yet to have its initial state (i.e. vtable) set.  However, the ErrorInstance::create(ExecState*, Structure*, JSValue) constructor violates this constraint by calling toString() in the argument list to a placement new call.

[Filip Pizlo]

Created attachment 104974 [details]
the patch

[Filip Pizlo]

Comment on attachment 104974 [details]
the patch

Tests pass, ready to commit if others concur.  Interestingly, this failure is already covered by our run-javascriptcore-tests, but not in LayoutTests.

Should we port the relevant test into LayoutTests, or maybe we should have run-webkit-tests trigger run-javascriptcore-tests automatically?

[Darin Adler]

(In reply to comment #2)
> maybe we should have run-webkit-tests trigger run-javascriptcore-tests automatically?

This is the right long term direction. But ideally we want this to run tests and show failures the same way run-webkit-tests does, not an entirely different way. The internal machinery doesn’t have to be identical, but the format of results should be.

[WebKit Review Bot]

Comment on attachment 104974 [details]
the patch

Clearing flags on attachment: 104974

Committed r93710: <http://trac.webkit.org/changeset/93710>

[WebKit Review Bot]

All reviewed patches have been landed.  Closing bug.