Bloop: Add restarting mode to `bloop run --watch`

[guersam]

Currently, bloop run --watch only works for self-terminating apps so that it's not applicable for non-terminating apps such as servers.

What about adding an option to kill the forked vm and restart it when a source file change is detected? This feature would be drop-in replacement of https://github.com/spray/sbt-revolver.

[jvican]

Thanks for opening this ticket Jisco. I agree this is valuable and would love to see it available in a release soon. Feel free to give it a try, I don't think the changes will be big because we already support cancellation of forked processes via Monix.

[tindzk]

This is a duplicate of https://github.com/scalacenter/bloop/issues/475.

[jvican]

Is it? You didn't specify that your application was still running and you were expecting bloop to kill the JVM and restart it. If it is, let's close #475 in favor of this ticket that is clearer about the goal and more actionable.

[svalaskevicius]

btw this can be worked around with some tools :) http://eradman.com/entrproject/

fd scala | entr -r bloop run ...

[jvican]

Nice :smile:

I'd like to implement this for the CLI at some point too, but if I can avoid developing/maintaining it the better. Thanks for sharing that incantation!

On the IDE side (Metals/IntelliJ), a way to emulate the behavior expected in the ticket is to have the build clients send a BSP buildTarget/run request to a file whenever there are are changes in a file in the editor.

[svalaskevicius]

Bloop starts up quickly (thanks for that!), so using entr for this is actually quite ok usability-wise.

Also, the Unix philosophy.. :)

[davojan]

The only downside of the entr workaround is that whenever you add new a new source file, it won't be watched. So native support of killing daemons will work a little better.

@svalaskevicius big thanks for this solution.

[dkovalenko]

+1 for this issue, without it there's no way to iterate fast with server development.

[kubukoz]

@dkovalenko if you need to iterate that fast, maybe you should write a test instead of starting an entire server ;)

On a more serious note (or is it?), I'd also appreciate this feature.