[Webkit-unassigned] [Bug 97036] A lot of ERROR messages in FractionalLayoutUnits when opening fast/overflow/overflow-height-float-not-removed-crash3.html regression test

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=97036





--- Comment #3 from Levi Weintraub <leviw at chromium.org>  2012-09-18 15:39:57 PST ---
(In reply to comment #2)
> I think that this bug needs to be fixed by whoever uses LayoutUnits incorrectly.

Here's a possible culprit: https://bugs.webkit.org/show_bug.cgi?id=68744

I'll have to dig in when I have some time to positively identify this particular issue.

> Not sure if there is complete consensus on what LOG_ERROR should be used for. I think that it's really usually about things largely outside WebKit control, like corrupt sqlite databases. Broken HTML input should never cause any sort of console logging.
> 
> Using LOG_ERROR as a temporary solution before turning it into an ASSERT seems very reasonable to me, as long as we agree that making it an assertion is the goal.

To be clear, the logging only happens on Debug builds.

I agree in theory with what you're saying. It's currently possible to overflow rendering code any number of ways, and usually we handle it in some sort of graceful way, such as maxing with zero when we expect positive numbers. That and outstanding issues such as the one I mentioned prevented us from going with ASSERTS from the beginning (it was our initial implementation). Casting negative numbers to unsigned is almost never the right thing to do, so this particular check would likely be a great candidate for an ASSERT.

Longterm, Emil is experimenting with saturated math that would enable us to ASSERT for any actual overflows.

-- 
Configure bugmail: https://bugs.webkit.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.