Dangling dev server process

[probablykasper]

Describe the bug
Sometimes when tauri exits, I'm left with the beforeDevCommand still running in the background (as a node process), which I notice because the port is still in use the next time I run tauri dev. I'm guessing this happens on some crashes or some other unusual situations.

To Reproduce
Haven't found a way to reproduce it reliably yet

Platform and Versions (please complete the following information):

OS: macOS 10.15.7
Node: 14.16.0
NPM: 6.14.11
Rustc: 1.51.0

[probablykasper]

Found two ways to reproduce:

• Quit the app itself by right-clicking it in the Dock.
• Call app.exit() in the webview

[probablykasper]

I think the issue could be that kill_before_dev_process() is not necessarily called depending on the if statements here:

tauri/tooling/cli.rs/src/dev.rs

Lines 171 to 190 in 0f86bc4

	let status = child_clone.wait().expect("failed to wait on child");
	if exit_on_panic {
	// we exit if the status is a success code (app closed) or code is 101 (compilation error)
	// if the process wasn't killed by the file watcher
	if (status.success() || status.code() == Some(101))
	// `child_wait_rx` indicates that the process was killed by the file watcher
	&& child_wait_rx
	.lock()
	.expect("failed to get child_wait_rx lock")
	.try_recv()
	.is_err()
	{
	kill_before_dev_process();
	exit(0);
	}
	} else if status.success() {
	// if we're no exiting on panic, we only exit if the status is a success code (app closed)
	kill_before_dev_process();
	exit(0);
	}

[lucasfernog]

What framework are you using? What's your beforeDevCommand?

[probablykasper]

I'm using vite/Svelte

[lucasfernog]

I can confirm this issue, it should affect all users. It happens because we're using sh -c $beforeDevCommand and it's not killing the child process.

[lucasfernog]

I've fixed it on the Tauri side, but Svelte still has an issue, the server kill isn't working for some reason.

[probablykasper]

@lucasfernog Hmm, a bit confused by that. Sending SIGINT to the tauri dev process works just fine, so wouldn't it be possible to handle this the same way?

[lucasfernog]

I think SIGINT is handled by cargo, not sure what kind of magic they're doing. But a simple pkill -TERM -P pid doesn't kill the Svelte subprocess :/ even though it receives the signal successfully. It works just fine on Vue.js.

[probablykasper]

Reproduced it with just the svelte template, so opened an issue there. Thanks for helping with this!