Re: [GNU ELPA] eglot-x.el: Protocol extensions for Eglot

[Eli Zaretskii <eliz@gnu.org>]

> From: João Távora <joaotavora@gmail.com>
> Cc: Felician Nemeth <felician.nemeth@gmail.com>,  emacs-devel@gnu.org
> Date: Fri, 05 May 2023 15:20:21 +0100
> 
> > At the very least, eglot.el could benefit from offering a
> > supported mechanism for adding such extensions; there should be no
> > need for using advice for that.
> 
> I agree.  Eglot has an API.  I haven't looked at eglot-x in depth but I
> suspect it's using some of this API in part.  If it's not sufficient,
> then Eglot could enhance it.  What part of eglot-x is using advice, if
> any?
> 
> > João, why would you not consider supporting these extensions as part
> > of eglot.el?
> 
> They're non-standard, each server has its own set of extensions, which
> in some cases could be volatile and become obsolete.  Furthermore, I
> would suspect that if eglot-x.el modifies eglot.el's standard behaviour,
> it is not as well tested across different servers and could provide a
> surprise for users of other servers.

I understand and agree in general, but surely each change should be
judged by itself, and some, maybe many, might not have these
problematic aspects?

> A guiding principle of eglot.el (and indede LSP) is to not have any
> server-specific code.

If "server-specific" behavior is, for example, just some non-standard
capabilities, that doesn't count as "server-specific code" in my book.
After all, a standard-compliant server could very well implement only
a subset of the capabilities documented by the standard.

> So in general, I think having eglot-x.el as a kind of laboratory where
> authors can experiment freely with new LSP-powered UIs is an excellent
> approach.  As these extensions make their way into the official LSP
> protocol Eglot can support them.  For example, I believe "inlay hints"
> started as some kind of rust-specific protocol extension and is now
> well-supported in Eglot for many servers (I use it daily).
> 
> We could consider bundling eglot-x.el with the Eglot ELPA package (I do
> hope Felicián is aware of the ins and outs of :core packages though).
> And I think a separate package isn't really bad either.

But it should not use advice, IMNSHO.