On Thu, Jan 26, 2023 at 6:28 AM Eli Zaretskii <eliz@gnu.org> wrote:

> > From: Dimitri Belopopsky <dimitri@belopopsky.com>
> > Date: Wed, 25 Jan 2023 23:34:02 +0100
> >
> > I've been working on adding support for inlay hints inside eglot using
> overlays.
> > Here is a working patch, but I'm still missing a few things:
> >
> > - I can't figure out a way to show the hints on a document without
> causing lags or timeouts from the lsp
> > server
> > - I'm currently updating the hints by sending the whole file each time
> (to make sure all hints get updated
> > correctly). I'm not sure on how to make this more efficient (or if it
> even necessary).
> >
> > On the implementation side:
> > - implemented with overlays as a minor model, enabled by default
> > - shows all hints supported by the protocol
> > - there is a customisation to disable the minor mode if the user doesn't
> want the feature
> >
> > I'd love to get a few points to finish this patch, and of course any
> ideas and feedbacks are welcome!
>
> Thank you for working on this important feature.
>
> AFAIU, inlay hints provide information of the same kind as ElDoc and
> in similar manner from the display and UX POV.  So I think this
> feature should work via ElDoc, not as a separate from-the-scratch
> implementation.  ElDoc is already capable of using Eglot-supplied
> information, so perhaps the only feature we need to add is the
> capability of ElDoc to (optionally) display the information in
> overlays near point.  (I thought we already had such a capability in
> eldoc.el, but it looks like I was dreaming, because I cannot find it
> there.)
>
> The advantage of basing this on ElDoc is that then we will be able to
> provide similar features from information sources other than Eglot.
>
> João, WDYT?
>

I think you're mostly right in your analysis.  I also think this should be
provided
via some LSP-agnostic infrastructure and not as a separate from-the-scratch
implementation, as you put it. This is (and was) also my stance in other
similar
matters.

The devil is in the details, of course.  Is eldoc.el the right
infrastructure
for it. What augmentation does it need, if any?  Are these augmentations
practical given the relative size, complexity and history of eldoc.el?
Isn't
an "inlay.el" or a "little-hint.el" support library a better choice? Or
maybe
flymake.el with it's use of overlay-based annotations is also acceptable
here?  I don't (yet) have answers to these questions.

Of course the existence of this prototype by Dimitri is certainly
No Bad Thing and a good starting point.

João