[Webkit-unassigned] [Bug 72254] [GTK] Rounding errors on 32-bit machines causes tests to fail

Mon Jan 9 09:29:48 PST 2012

https://bugs.webkit.org/show_bug.cgi?id=72254

--- Comment #19 from Martin Robinson <mrobinson at webkit.org>  2012-01-09 09:29:48 PST ---
(In reply to comment #18) 

> Does this patch affect only buildbot?

If you are building a development release, you should use configure and make manually to compile. This change only affects build-webkit, which is used by developers working on trunk, the build bots and the EWS.

> And developers will still have these failures?

Anyone not compiling with build-webkit will see these failures. These are only "failures" in the sense that they cause the layout tests to fail, as they are rounding errors that typically result in one pixel differences. This is not something that matters or would be noticed when actually using WebKit.

On the other hand, these tests probably fail on ARM anyway, if rounding affects the results.

> I thought development builds and buildbot builds should be exactly the same?

They are not. The build bot often tests features that are unfinished or experimental, but we want test coverage for.

> If so, what's the point of buildbot then if developers will get a different results?

The build bots typically test a superset of the features in development and stable releases. In this case, it was appropriate to add these flags only for build-webkit, because they improve result consistency at the expense of some unknown performance impact. We didn't want this work-around to affect the performance of releases.

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