Using watchexec with Wisp

Wisp is a web framework for Gleam.

There is no hot-reloading out of the box for Gleam or for Wisp. But, it’s not a big deal, as you could use watchexec or fswatch (or I’m sure other tools too) to set up some file watching. I’ve used both and they’re both fine.

If you set a basic watchexec command like watchexec --restart --watch src -- gleam run, you will likely see some errors that look something like this, Error! Failed to eval: abc@@main:run(abc) and this, {init_failed,<<"Failed to start socket listener: Eaddrinuse">>}.

That error is EADDRINUSE. If you check out the linux manpage, you will see something like, “The specified local address is already in use or the filesystem socket object already exists.”

Yes, so the watchexec isn’t really giving the server process enough time to clean up properly. (At least, I expect that’s what’s going on after doing some experimenting.)

So, add a sleep 2 in front of the command that watchexec is to run:

watchexec \
 --debounce 500ms \
 --restart \
 --watch src \
 --watch test \
 --watch gleam.toml \
 -- 'sleep 2 && gleam run'

You might look at that and think: “Oh! That sleep X && blah pattern is no good, watchexec even has an option for it!” I did try that first, and for whatever reason, it wasn’t working for me. Here are a couple things that I tried that didn’t work:

--delay-run 2s: According to the watchexec manpage, ‘This is like using “sleep 5 && command” in a shell, but portable and slightly more efficient.’. However, it didn’t work for me.
Setting --stop-timeout: Didn’t work, but I also didn’t try to figure out why.

I think it may be possible to figure out what is going wrong with those two options by looking through the watchexec logs if you set the verbosity high enough.

Anyway, the above command works fine. You may need to play around with different combinations of debounce time and sleep seconds to really dial it in.

Troubleshooting

Sometimes, things may get weird. You may get some sort of crash or other event that causes things to break. In these cases you may need to manually kill the beam.smp process and restart the watcher.

You can find the correct process using ps:

ps aux | grep beam

In that process list, look for some command called beam.smp and look for your app in the arguments. E.g., if your app is called abc, then you would look for something like this: -eval token_tally@@main:run(token_tally) in the arguments.

For example, you may see something like this:

ryan 12345 ... /opt/homebrew/Cellar/erlang/28.1/lib/erlang/erts-16.1/bin/beam.smp -- ... -eval abc@@main:run(abc) -noshell -- --

(The ... placeholders will be filled with a ton of stuff, so be aware of that.)

Go ahead an kill it:

kill 12345

Wait a few seconds, then restart the watcher and you should be good to go.