Re: What is the most useful potential feature which Emacs lacks? A: Autocompletion

[Dmitry Gutov <dgutov@yandex.ru>]

On 03.06.2020 17:40, Konstantin Kharlamov wrote:

> I mean, I was typing outside Emacs, in the same terminal where
> `libinput debug-events` was running. I was just testing how fast I can
> type, to figure out how small the timeout should be set if I want a
> completion to pop up.

Not sure how to interpret its output. Here's what I was getting while 
typing 'prog':

  event15  KEYBOARD_KEY     +21.57s	*** (-1) pressed
  event15  KEYBOARD_KEY     +21.66s	*** (-1) pressed
  event15  KEYBOARD_KEY     +21.68s	*** (-1) released
  event15  KEYBOARD_KEY     +21.74s	*** (-1) pressed
  event15  KEYBOARD_KEY     +21.76s	*** (-1) released
  event15  KEYBOARD_KEY     +21.84s	*** (-1) pressed
  event15  KEYBOARD_KEY     +21.86s	*** (-1) released
  event15  KEYBOARD_KEY     +22.00s	*** (-1) released

That seems like I hit a key per ~100ms, but handling the releases adds 
its latency.

>>> I don't see any lags, so I assume using an async backend with the
>>> timeout set to 0 should work fine.
>>
>> Indeed, that's what asynchronous means. But the quality of the user
>> experience also depends on how quickly the backing server can handle
>> those requests.
> 
> Well, I paused the backing server in the steps above, so server
> couldn't answer. This was emulating a behaviour where a user works with
> a project too big for backend to return a completion immediately.

That tests the best case (which is, it assumes that when you turn the 
server back, it can handle the load). No typing latency is good (of 
course), but if you don't see completion suggestions because the backing 
process is overloaded, that's not so great either.

I'm not saying it's going to be a problem with Irony or LSP necessarily, 
but I've seen this reported in other circumstances. Just something to 
pay attention to.

>> The majority of backends are synchronous. And the "standard"
>> completion
>> API for Emacs (which we want to integrate with) still only supports
>> the
>> synchronous convention.
> 
> I am a bit confused by the last sentence. What's the relation between
> the Emacs API and already working company-mode? Did you mean, company-
> mode is trying to be compatible with backends for some standard Emacs
> API, and those can't be async?

Pretty much.

Eglot does include a hack to minimize typing latency as well, but it's 
an ad-hoc solution, and there's no way Company would be able to detect 
this kind of approach being used.