bug#61726: [PATCH] Eglot: Support positionEncoding capability

[João Távora]

On Sun, Feb 26, 2023 at 6:03 AM Eli Zaretskii <eliz@gnu.org> wrote:
>
> > From: Augusto Stoffel <arstoffel@gmail.com>
> > Cc: Eli Zaretskii <eliz@gnu.org>,  61726@debbugs.gnu.org
> > Date: Sun, 26 Feb 2023 00:57:42 +0100
> >
> > The linear cousin of "lsp-abiding" is very similar to the
> > function that Eli wrote:
> >
> > (defun eglot-move-to-lsp-abiding-column-linearly (column)
> >   "Move to COLUMN as computed using the LSP `utf-16' criterion."
> >   (let* ((bol (line-beginning-position))
> >        (goal-char (+ bol column))
> >        (eol (line-end-position)))
> >     (goto-char bol)
> >     (while (and (< (point) goal-char)
> >               (< (point) eol))
> >       (if (<= #x010000 (char-after) #x10ffff)
> >         (setq goal-char (1- goal-char)))
> >       (forward-char 1))))
> >
> > It would be very hard to believe they have different performance
> > characteristics.  In fact, in some hastily done tests, I get the
> > following relative running times:
> >
> > eglot-move-to-colum: 1.0
> > eglot-move-to-lsp-abiding-column-linearly: 8.4
> > eglot-move-to-bytewise-column: 8.0
> > eglot-move-to-lsp-abiding-column: 14.4
>
> The current version of eglot-move-to-lsp-abiding-column calls
> encode-coding-region in the loop, which I guess is the main reason for
> its being slower.

Yup, here's a surprise.  We don't need encode-coding-region
after all and the binary search I implemented back in
https://github.com/joaotavora/eglot/pull/125
is mostly useless.  In that issue, noone thought of counting
characters by examining the code point values and counting
code units.

So I pushed this much faster, simpler version to emacs-29.

I credited Eli in the commit, as we wrote the code. Hope
that's OK.

João