:wq silently exits without an error if target file is not writable

[Pistahh]

Reproduction steps

:wq should fail and indicate if it couldn't save the file but it doesn't, it just exits.

1. edit a file that you have no permissions to, e.g. /etc/hosts as non-root
2. Try to save it and exit with :wq
3. watch as hx exits without any indication that it couldn't save the changes
4. inspect /etc/hosts that it indeed wasn't changed

Environment

• Platform: Linux
• Terminal emulator: gnome terminal
• Helix version: v0.5.0-186-g371c84f

[dead10ck]

I started trying to look into this one. At first glance, it seems like this should exit with an error before quitting. However, if we look at write_impl, we can see that it looks like it just adds an async job and returns immediately, which would explain why it just proceeds to quit, and doesn't really care if the write succeeds or not:

helix/helix-term/src/commands.rs

Lines 2087 to 2094 in 333c294

	let future = doc.format_and_save(fmt);
	cx.jobs.add(Job::new(future).wait_before_exiting());
	if path.is_some() {
	let id = doc.id();
	let _ = cx.editor.refresh_language_server(id);
	}
	Ok(())

I thought it would make sense to wait on the job in between calling write and calling quit, so looking at the Jobs type, it looks like there is a function for waiting on all the pending jobs:

helix/helix-term/src/job.rs

Lines 95 to 99 in 333c294

	/// Blocks until all the jobs that need to be waited on are done.
	pub fn finish(&mut self) {
	let wait_futures = std::mem::take(&mut self.wait_futures);
	helix_lsp::block_on(wait_futures.for_each(|_| future::ready(())));
	}

However, as we can see, this doesn't care about the result of the jobs either. Would it make sense to have this function return a Result that's an error if any of the pending jobs error, and Ok otherwise?

[DannyJJK]

I'd love to try using Helix as my main editor, but this single issue makes me nervous to use it. There's various reasons a file could fail to write that are not just permissions based, I wouldn't want to think I'd saved a file successfully to find that no changes have been made. Has there been any further progress on this?

[dead10ck]

I agree, this seems rather important from a reliability perspective. I'll take a crack at the approach I mentioned above.