Otherwise, we may be in a state where we don't have a valid exit state. So if we exitted at that point, we wouldn't know how to recover certain values. validateFTLOSRExitLiveness catches these issues. The solution is to not emit an exit due to an exception being thrown if the node we're compiling isn't allowed to throw. This should also be a tiny bit of a speedup/code size reduction.
<rdar://problem/45220139>
Created attachment 352746 [details] patch
Attachment 352746 [details] did not pass style-queue: ERROR: Source/JavaScriptCore/ChangeLog:13: Line contains tab character. [whitespace/tab] [5] Total errors found: 1 in 4 files If any of these errors are false positives, please file a bug against check-webkit-style.
Created attachment 352747 [details] patch
Comment on attachment 352747 [details] patch View in context: https://bugs.webkit.org/attachment.cgi?id=352747&action=review r=me > Source/JavaScriptCore/ChangeLog:21 > + emitting an exception check. A node node being able to exit implies I think you meant to say "A node not being able" > Source/JavaScriptCore/ChangeLog:22 > + that it can't exit for exceptions (hence, it can't throw an exception). I suggest saying "it also can't".
Created attachment 352814 [details] patch for landing
Comment on attachment 352814 [details] patch for landing Clearing flags on attachment: 352814 Committed r237297: <https://trac.webkit.org/changeset/237297>
All reviewed patches have been landed. Closing bug.