Diagnostic floating window disappears automatically

[suguruwataru]

When showing the diagnostic messages via the floating window, the floating window automatically disappears after ~5 seconds. This often does not given enough time to read the message. Currently there also isn't a way to configure the timeout or whehter the floating window should disappear.

[suguruwataru]

Still trying to figure out what this happens, but I already got interesting observation: the floating window disappears not after 5 seconds since it appears, but 5 seconds after the cursor moves onto the problematic piece of code. If g:lsp_diagnostics_float_delay is set to 3000, the floating window appears 3 seconds after the cursor moves onto the problematic code, stays there for 2 seconds, and disappears.

[suguruwataru]

At this point I'm pretty sure this is caused by the diagnostic window created by the CursorMoved event being overriden by the window created by the CursorHold event. CursorHold event is fired in normal mode after the cursor doesn't move for updatetime, which by default is 4000ms (OK I'm not too good at counting seconds).

So what would happen is, if g:lsp_diagnostics_float_delay is set to 3000, when the cursor moves onto the problematic code, CursorMoved event fires, and a diagnostic window gets created 3 seconds later. Then, one second later, since cursor hasn't moved for a total of 4000ms, CursorHold fires. It closes the window first (see line 28 of the referenced code below), and makes a setup that would show the window again after g:lsp_diagnostics_float_delay. However, the window won't show again, as buffer status hasn't changed since the CursorMoved event, and therefore the distinctUntilChanged call prevents a new diagnostic floating window from being created.

vim-lsp/autoload/lsp/internal/diagnostics/float.vim

Lines 18 to 36 in cb3406a

	let s:Dispose = lsp#callbag#pipe(
	\ lsp#callbag#merge(
	\ lsp#callbag#fromEvent(['CursorMoved', 'CursorHold']),
	\ lsp#callbag#pipe(
	\ lsp#callbag#fromEvent(['InsertEnter']),
	\ lsp#callbag#filter({_->!g:lsp_diagnostics_float_insert_mode_enabled}),
	\ lsp#callbag#tap({_->s:hide_float()}),
	\ )
	\ ),
	\ lsp#callbag#filter({_->g:lsp_diagnostics_float_cursor}),
	\ lsp#callbag#tap({_->s:hide_float()}),
	\ lsp#callbag#debounceTime(g:lsp_diagnostics_float_delay),
	\ lsp#callbag#map({_->{'bufnr': bufnr('%'), 'curpos': getcurpos()[0:2], 'changedtick': b:changedtick }}),
	\ lsp#callbag#distinctUntilChanged({a,b -> a['bufnr'] == b['bufnr'] && a['curpos'] == b['curpos'] && a['changedtick'] == b['changedtick']}),
	\ lsp#callbag#filter({_->mode() is# 'n'}),
	\ lsp#callbag#filter({_->getbufvar(bufnr('%'), '&buftype') !=# 'terminal' }),
	\ lsp#callbag#map({_->lsp#internal#diagnostics#under_cursor#get_diagnostic()}),
	\ lsp#callbag#subscribe({x->s:show_float(x)}),
	\ )

[suguruwataru]

To fix this, I think we should remove CursorHold from the events that can show the diagnostic floating window. This should be fine, as there once was a time that this event wasn't listened. @ryu-ichiroh added it to the list in this commit:

3244bed

@ryu-ichiroh what's the reason you added this event? Do you think there's a reason that we shouldn't remove it?

[suguruwataru]

After removing CursorHold from the list of events that trigger diagnostic floating window update, the window indeed stops disappearing. I've created PR #1624.