Conversation
|
@emilio I think you were interested in this. If you have feedback it would be great, particularly if you have evidence that sharing load-type tests (e.g. gecko crashtests) between engines is effective at finding issues. |
zcorpan
left a comment
There was a problem hiding this comment.
wpt already has a bunch of crash tests, that might use testharness.js or reftest just to be able to run. e.g.
css/css-overflow/shrink-to-fit-auto-overflow-relayout-crash.html
css/css-pseudo/first-letter-of-html-root-crash.html
This was my concern while reading the proposal, so I'm glad you included it! Personally, I'm pessimistic that tests like these do provide much value between implementations. I'm no implementer, though, so I'm very curious about what @emilio has to say on the subject. |
|
And this was a very simple test runner, there's tons of test-cases that I haven't ran or what not. |
|
I'd rather call them crash tests, given that's what they're called everywhere else. |
|
The reason for not calingthem crashtests was that crashing is only one possible failure mode; they could also fail with an assert or sanitizer failure or similar in vendor infrastructure. But I'm sympathetic to the idea that "load" can be confusing and there is precedent here. |
|
I think that's an overly pedantic viewpoint to take when there's prior art in all vendors here already, even if the name is slightly confusing. (Just make all assertions and saniziter failures crashes, pff!) |
|
web-platform-tests/wpt#20017 is a draft PR for this RFC. It's not done yet (hence draft), but what's there seems to work. |
jgraham
left a comment
There was a problem hiding this comment.
I think the only substantive decision left here is whether to use -crash or -crashtest in the filename. The former requires some additional work to rename existing files, but is also shorter and closer to existing practice for WebKit. I'd like to make a decision there and then close out this issue so we can land the changes.
|
My preference is for |
|
OK changed to |
|
Doing some archeology here, I wonder when we changed the wait class from |
Add a test type that just ensures that a page can be loaded without error.