Bug 107690 – Missing regression test for bug 92376

• Status: NEW
• Component: Tools / Tests

[Dominik Röttsches (drott)]

I was revisiting the discussion in bug 92376, trying to create a reliable regression test for the harfbuzz selection issue.
Here's my attempt so far, with two potential issues.

1) Can we import Mukti Narrow? (License is GPL v2), see http://savannah.nongnu.org/projects/freebangfont
2) This test is potentially flaky or timing out due to its probabilistc approach.

In any event, it's reproducible not only with Telugu, but also with Bengali - which had been Tony's original question.

[Dominik Röttsches (drott)]

Created attachment 184232 [details]
Patch

[Dominik Röttsches (drott)]

At this point, this is mostly fyi, Tony, since I promised to take a look. I don't see how I can land this, since it has too much potential for being flaky.

[Tony Chang]

We can't add GPL v2 to the repository, but for Chromium Linux, we install our font dependencies along with our build dependencies in install-build-deps.sh: http://src.chromium.org/viewvc/chrome/trunk/src/build/install-build-deps.sh?revision=176982&view=markup

It looks like we install ttf-indic-fonts and load MuktiNarrow.ttf: http://trac.webkit.org/browser/trunk/Tools/DumpRenderTree/chromium/TestShellX11.cpp#L144

Being flaky is not so great, although I guess the test passes if it doesn't crash so if we didn't trigger the bug, we would still "pass" the test, right?

I think it would be OK to land this, but we should move the test into platform/chromium-linux/editing/selection (since only Chromium Linux will have the font) and load the font from the system rather than checking in a copy of it.

[Dominik Röttsches (drott)]

(In reply to comment #3)
> We can't add GPL v2 to the repository, but for Chromium Linux, we install our font dependencies along with our build dependencies in install-build-deps.sh: http://src.chromium.org/viewvc/chrome/trunk/src/build/install-build-deps.sh?revision=176982&view=markup

Okay, yes, that's not a problem - we could have that font in our jhbuild package which is downloading dependencies similar to the script you're referring to.

> Being flaky is not so great, although I guess the test passes if it doesn't crash so if we didn't trigger the bug, we would still "pass" the test, right?

> I think it would be OK to land this, but we should move the test into platform/chromium-linux/editing/selection (since only Chromium Linux will have the font) and load the font from the system rather than checking in a copy of it.

I am gardening myself, so I think flakiness is in fact not acceptable for landing. The thing is unfortunately: It's about tuning how many iterations of random mouse selections you do on that page - if you do a large enough number of them and the patch is not applied, it crashes. If you tune that number down, the test completes before timing out, but your chance of reproducing the crash are reduced, even if the code still has the issue.

So, I think it's not a perfect test since it's not deterministic enough - and I currently don't know how to turn it into one. :-(