Bug 98209 – fast/text/international/float-as-only-child-of-isolate-crash.html is flaky on Windows

• Status: NEW
• Component: Tools / Tests

[Roger Fong]

Skipping this test for now to paint bots green.

For some reason when the Windows specific results are there the output matches Mac results.
When I get rid of the Windows results (to use the Mac results) it goes back to matching the original Windows specific results.

Which makes no sense. 

Don't really have time to work on this right now but it's either a really unfortunate coincidence or something (else) is terribly wrong with DRT.

Anyways, right now I'm using Windows specific results and it's matching Mac results. I'm not going to remove the Windows specific results because it will probably just fail again anyways after I do that.

[Roger Fong]

skipped: http://trac.webkit.org/changeset/130227

[Levi Weintraub]

Did this just start happening? What change prompted this?

[Roger Fong]

The test itself has been failing since before https://bugs.webkit.org/show_bug.cgi?id=87998.
It only started "passing" about two weeks ago, the bots were in a bad state at that point so the progression range was somehow gigantic.
Thus I got rid of the failing expected results. And then it started failing again. 

Now it's passing again. (As of yesterday)
I didn't see any patches in the regression that looked like it would change the test. This seems like more of a Windows DRT thing than anything else.

I can try using the passing results again if you want...

I'm now noticing that fast/text/international/harfbuzz-buffer-overrun.html
I wonder if it's related to me skipping this test. The regression range for that one is somehow about 100 regressions large despite the fact that it seems to have just started failing recently.

[Roger Fong]

Hm...that failing test passes locally...