HasIndexedProperty clobberize rule is wrong for Array::ForceOSRExit

[Saam Barati]

...

[Saam Barati]

<rdar://problem/27328836>

[Saam Barati]

Created attachment 290393 [details]
patch

[Filip Pizlo]

Comment on attachment 290393 [details]
patch

View in context: https://bugs.webkit.org/attachment.cgi?id=290393&action=review

Ouch.  R=me.

> Source/JavaScriptCore/dfg/DFGClobberize.h:267
> +        if (mode.type() == Array::ForceExit) {
> +            write(SideState);
> +            return;
> +        }
> +

Why didn't you put this in the switch statement, below?

[Saam Barati]

Created attachment 290397 [details]
patch for landing

[Saam Barati]

I'm going to try to find a repro test before landing, but it might be tricky

[Saam Barati]

(In reply to comment #5)
> I'm going to try to find a repro test before landing, but it might be tricky

This is difficult to do.

[Saam Barati]

(In reply to comment #6)
> (In reply to comment #5)
> > I'm going to try to find a repro test before landing, but it might be tricky
> 
> This is difficult to do.

Basically, we need to profile something as contiguousInbounds, and then somehow not have a prediction for either the base or the index.

[Saam Barati]

Ok, I'm going to give up on trying to get this to reproduce on ToT. It's pretty clear what the bug is, but it's probably quite difficult to write a program that exhibits the bug. Anyways, I'm going to commit this change for now and then if I find a way to write a test that exhibits the bug, I'll check in the test later.

[WebKit Commit Bot]

Comment on attachment 290397 [details]
patch for landing

Clearing flags on attachment: 290397

Committed r206955: <http://trac.webkit.org/changeset/206955>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.