[Webkit-unassigned] [Bug 126432] REGRESSION (r161195): Acid2 regression tests frequently fail

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=126432





--- Comment #10 from Alexey Proskuryakov <ap at webkit.org>  2014-01-07 08:44:38 PST ---
> > There needs to be a specific reason to keep brokenness in the tree.

> - The problem does not reproduce locally so I'll have to try landing fixes for bots to test. 

This is one specific reason that could validate keeping the tree in bad state while fixing it. It also almost certainly doesn't apply here - the problem is well localized, and should be easy to attack locally. Some ideas:

- run the tests many times in fully parallel mode, forcing heavy CPU use;

- change the delay from 0 to some positive value.

> Repeatedly remanding the original creates churn and confusing history. It may break other peoples work-in-progress patches.

These general reasons are valid, but quite weak. Any and all changes landed can break others' work in progress patches, and that's not a reason to halt all work. Also, the faster you roll out, the less likely it is to affect anyone.

> - One ~5-10% flaky test doesn't really affect anyone else's work much.

This is both incorrect, and not applicable in this case.

1. In this case, there are 4 flaky tests, each with a much higher rate of failure.

2. There are dozens engineers on the project, and if everyone felt entitled to cause 5-10% flakiness on tests, it would not be a project worth working on.

3. Flaky tests slow down EWS a lot, and there really isn't a way around that. When we only have one test with 5% failure rate, that means that 5% of patches take at least 3x time as long to be processed.

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