Catch the finished promise AbortError and ignores it

[jaywhy]

Fixes issue #247 by putting duct tape over it.

[excid3]

Think there's any downsides to doing this?

[jaywhy]

Yes, I think there could be unforeseen downsides.

Thinking about it more, I wouldn't say I like this patch. It doesn't fix the problem, it just covers it up. If the AbortError is caught, animation glitches still occur. It only fixes the issue with Promise.all rejecting ALL animations when one promise is rejected. So, one animation won't cause others to stop.

I tested el-transition and Alpine.js yesterday. el-transition has the same problem. Alpine.js doesn't, but the Alpine folks do something completely different in their transition library.

I have some time later this week. I can look into it more.

[excid3]

Ours is a fork of el-transition trying to fix a few things, so it's very similar. I tried to understand how Alpine does it but couldn't quite wrap my head around it.

Appreciate your help on this! It's definitely out of my depths.

[jaywhy]

Ok. I tried to fix it without rewriting it, but I couldn't get that to work.

Are you okay with me rewriting the transition.js library like this?

The code fixes the issues with AbortError's. The transition code now works sort of like Alpine.js. It sets a _stimulus_transitions property on the element, and if another transition comes along and sees that, it will cancel the already running transition.

However, the new transition.js breaks some tests. But I think the broken tests are all timing-related or race condition-like things because the code no longer awaits on the animation Promises. I don't think this will affect production code unless someone relies on these weird timing inconsistencies. Famous last words, though.

I'll give a couple of examples of the test problems...

The dropdown_test.js needs to run await nextFrame() twice, not once, for it to pass. It requires two calls because the transition.js calls requestAnimationFrame twice and doesn't use await.

One of the alert_test.js tests breaks because the closeButton.click() event in that test now happens before the setTimeout() in AlertController#connect() gets to run. The click() calls leave(), and then the setTimeout() code runs, calling enter(), which cancels the leave().

If you're ok with this change, I'll clean things up, fix the broken tests, and squash my commits.

[excid3]

This is looking good @jaywhy! Will give it a test locally later today, but it seems good to me.

[excid3]

Looks like it's working great!

Anything else to do or is it ready to merge?

[jaywhy]

Yes. I think so.

[excid3]

Thanks so much for this @jaywhy 👍