[webkit-reviews] review denied: [Bug 14771] Mostly reproducible crash in modified celtickane benchmark : [Attachment 15724] Possibly A Patch?

[bugzilla-daemon at webkit.org]

Mark Rowe (bdash) <mrowe at apple.com> has denied John Moe <john at johnmoe.com>'s
request for review:
Bug 14771: Mostly reproducible crash in modified celtickane benchmark
http://bugs.webkit.org/show_bug.cgi?id=14771

Attachment 15724: Possibly A Patch?
http://bugs.webkit.org/attachment.cgi?id=15724&action=edit

------- Additional Comments from Mark Rowe (bdash) <mrowe at apple.com>
I'm going to mark this r-.  I spoke with Maciej on IRC about the general
approach this patch takes, and he said:

07:04 <othermaciej> bdash: upon further thought - I think the right rule should
be "don't use shared substring append on a string that is only a small part of
its current buffer"
07:04 <oothermaciej> bdash: I think the patch might actually implement
something close to that (it looks at post-append length though)

It should be simple to adjust the patch to this slightly different strategy. 
It would also be a good idea to see if we can't come up with some decent
criteria for "small part of its current buffer" that's not just picking a magic
number out of thin air :-)

My other concern is that "10 * length" looks like we're asking for overflow to
happen when dealing with large strings.