Sending repeated SIGINT (Ctrl-C) signals to watchmedo auto-restart exits without stopping the child process

[bengotow]

Hey folks! I've been using the watchmedo CLI to make a web server automatically restart when the contents of a directory change. Unfortunately we ran in to issues with this pretty quickly after launch because sending a SIGINT to watchmedo:

1. Cleans up the file observers
2. Sends a SIGINT to the child process

Step 1 takes about 500ms (at least on large directories like the one we're watching) and sending a second SIGINT in that time frame kills watchmedo without a SIGINT ever getting sent to the child.

If you're one of those keyboard bashing folks that hammers Ctrl-C, watchmedo basically never stops the child process when you kill it.

We've implemented a fix for this over here (dagster-io/dagster#930) that involves calling event_handler.stop() before cleaning up the file observers here: https://github.com/gorakhargosh/watchdog/blob/master/src/watchdog/watchmedo.py#L124. This might have undesirable side effects when observe_with is used in other modes (I'm not sure?) but it ensures that the child process is killed as soon as watchmedo receives a signal and it doesn't leave dangling processes.

Would a PR for this be mergeable?

[BoboTiG]

Hello @bengotow,

Thanks for opening the issue. I think a PR would be mergeable, yes :)
(If you manager to add a simple test inside, it wold be great too!)