Bug 117373 – REGRESSION: ASSERT failures in stack walking code on FTL branch

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

[Mark Lam]

The first assertion, in CallFrame::setLocationAsBytecodeOffset(), wrongly assumes that the incoming offset argument is in units of bytes. This is not true. It is in units of Instruction*. Hence, the assertion which checks for the low 2 bits to be clear can fail.  The fix is to remove the assertion.

[Mark Lam]

Created attachment 204095 [details]
the patch.

[Oliver Hunt]

Comment on attachment 204095 [details]
the patch.

Can we have an enum type so distinguish whether an offset should be bytes or instructions?

e.g. enum InstructionOffset {};  enum ByteOffset {}; ByteOffset byteOffsetFromInstruction(unsigned i) { return i * sizeof Instruction } , etc

[Mark Lam]

<rdar://problem/14100876>

[Mark Lam]

(In reply to comment #2)
> (From update of attachment 204095 [details])
> Can we have an enum type so distinguish whether an offset should be bytes or instructions?
> 
> e.g. enum InstructionOffset {};  enum ByteOffset {}; ByteOffset byteOffsetFromInstruction(unsigned i) { return i * sizeof Instruction } , etc

I'm not sure that will work.  My gut feeling is that it will result in a lot of casting, which would defeat the type safety checks (which is why I think you want the enums, right?).

Thanks for the review.  Patch landed in r151347: <http://trac.webkit.org>.

[Mark Lam]

(In reply to comment #4)
> Patch landed in r151347: <http://trac.webkit.org>.

Landed in r151347: <http://trac.webkit.org/changeset/151347>.