Bug 7080 – Provide some way to stop a JavaScript infinite loop

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

[Joost de Valk (AlthA)]

At current, infinite javascript loops are really infinite, we need a way to detect and stop them.
This currently occurs in:

- bug 3552
- bug 6693
- bug 6958

This bug is here to track the development of a way to get out of this loop. All three bugs above have testcases, which could be used to test this behavior until their underlying bugs are fixed. All three of them have a root cause getting them into this infinite loop, so if these root causes are fixed they won't be useful for testing anymore.

[Geoffrey Garen]

In Radar as <rdar://problem/3895579>

[Alexey Proskuryakov]

*** Bug 6089 has been marked as a duplicate of this bug. ***

[Krishna]

*** Bug 8949 has been marked as a duplicate of this bug. ***

[Anders Carlsson]

Created attachment 8863 [details]
Hang detection

This implements the JSC part of a hang-detection implementation that works pretty much like the one in Mozilla.

Unfortunately, when turned on this causes a performance regression in ibench. The other approach I tried (using UNIX signals) proved to be complicated and error-prone and it too gave a performance regression (albeit a smaller one)

[Darin Adler]

Off. It's going to be tricky to come up with a way to do this without slowing things down!

[Geoffrey Garen]

Comment on attachment 8863 [details]
Hang detection

Clearing the review bit, since this was a perf. regression.

[Anders Carlsson]

Created attachment 8872 [details]
Much better patch

Here's a new patch that doesn't cause any measurable slowdown at all (Less than 0.2% on ibench)

[Maciej Stachowiak]

Comment on attachment 8872 [details]
Much better patch

r=me

ggaren likes it too

[David Kilzer (:ddkilzer)]

Committed as r14893.

[Anders Carlsson]

Created attachment 8886 [details]
Add WebKit delegate

[Darin Adler]

Comment on attachment 8886 [details]
Add WebKit delegate

SCRIPT_TIMEOUT_TIME_MS should be a C++ const, not a macro.

The shouldInterruptScript functions could be const.

The WebDefaultUIDelegate is a pretty heavyweight way to make shouldInterruptJavaScript default to no. WebFrameBridge should just say return NO -- I don't think we need this method int the WebDefaultUIDelegate.

[Anders Carlsson]

Created attachment 8902 [details]
Address comments

[Anders Carlsson]

Created attachment 8903 [details]
Address Geoff's comments

I talked to Geoff on IRC and he had a couple of comments which this patch addresses.

[Geoffrey Garen]

Comment on attachment 8903 [details]
Address Geoff's comments

r=me

and the crowd goes wild