Kqueue refactoring regression about fsnotify

[nathany]

Hi Otto. Sorry for not digging into this sooner. I've never used Revel before, but I got the booking sample running with some additional logging (on the develop branch).

Your previous bisect had pointed to 4a0d1ae being the culprit. Most of those changes are to addWatch, so I added some logging there and compared against ca50e73. With the result you're seeing, it seems plausible that it was watching more than it was previously.

So far I haven't figured it out. Still digging. It might be worth confirming that bisect, since you said it wasn't 100% reproducible. If it was a commit before, that could change things quite a lot.

If there was a medium sized open source Revel app that we could reproduce this on, that would make it much easier. It might even be reproducible with a similar file structure to your app, even if those files don't contain the same code. /cc @robfig

[nathany]

@ottob I hope you have Spotlight disabled for your folder, which can randomly decide to create events. https://github.com/go-fsnotify/fsnotify/wiki/FAQ.

Though if not disabled, I'd expect you'd see the same problems on v1.0.4 as well.

[nathany]

[ottob]

Feel free to close this. I will re-open if I ever manage to write a test that triggers the issue.

[nathany]

I've ran into a very similar issue here. nathany/looper#18 It is reliably reproducible. Looking into a fix now.

[nathany]

@ottob I believe this to be fixed on master. Stupid mistake on my part. Sorry for all the trouble.

[nathany]

Feel free to reopen this issue if you still see the issue.

[ottob]

👏 thanks for tracking it down.