bug#72824: Eglot highlights matches incorrectly wrt case sensitivity

bug#72824: Eglot highlights matches incorrectly wrt case sensitivity

[Dmitry Gutov]

X-Debbugs-Cc: joaotavora@gmail.com

1. Server gopls, for example.
2. completion-ignore-case=nil, its default value.
3. Input "clCe" and completions for it like

   clockCenterX
   clockCenterY

Completions are case-sensitive, but completions' highlighting works as 
if it's insensitive - highlights are on "cl", second "c" and "e", 
whereas the capital "C" is not highlighted.

This fixes it for me:

diff --git a/lisp/progmodes/eglot.el b/lisp/progmodes/eglot.el
index 844fc634be9..acc197754db 100644
--- a/lisp/progmodes/eglot.el
+++ b/lisp/progmodes/eglot.el
@@ -3260,7 +3260,7 @@ eglot-completion-at-point
             (try-completion pattern (funcall proxies)))
            ((eq action t)                                 ; all-completions
             (let ((comps (funcall proxies)))
-             (dolist (c comps) (eglot--dumb-flex pattern c t))
+             (dolist (c comps) (eglot--dumb-flex pattern c 
completion-ignore-case))
               (all-completions
                ""
                comps

bug#72824: Eglot highlights matches incorrectly wrt case sensitivity

[João Távora]

I think the fix makes sense.  Please push it to Emacs 30, and thanks
in advance.

If you can please use this format for the commit message subject,
which is my more or less usual practice.

Eglot: fix the frobnication (bug#12345)

* lisp/progmodes/eglot.el (eglot-foo): pass foo to eglot-bar.

On Tue, Aug 27, 2024 at 2:41 AM Dmitry Gutov <dmitry@gutov.dev> wrote:
>
> X-Debbugs-Cc: joaotavora@gmail.com
>
> 1. Server gopls, for example.
> 2. completion-ignore-case=nil, its default value.
> 3. Input "clCe" and completions for it like
>
>    clockCenterX
>    clockCenterY
>
> Completions are case-sensitive, but completions' highlighting works as
> if it's insensitive - highlights are on "cl", second "c" and "e",
> whereas the capital "C" is not highlighted.
>
> This fixes it for me:
>
> diff --git a/lisp/progmodes/eglot.el b/lisp/progmodes/eglot.el
> index 844fc634be9..acc197754db 100644
> --- a/lisp/progmodes/eglot.el
> +++ b/lisp/progmodes/eglot.el
> @@ -3260,7 +3260,7 @@ eglot-completion-at-point
>              (try-completion pattern (funcall proxies)))
>             ((eq action t)                                 ; all-completions
>              (let ((comps (funcall proxies)))
> -             (dolist (c comps) (eglot--dumb-flex pattern c t))
> +             (dolist (c comps) (eglot--dumb-flex pattern c
> completion-ignore-case))
>                (all-completions
>                 ""
>                 comps
>
>
>


-- 
João Távora

bug#72824: Eglot highlights matches incorrectly wrt case sensitivity

[João Távora]

Also, btw, completion highlighting is naturally borked too.  There is
no guarantee that that naive flex algorithm is what the server used
to select those completions.

On Tue, Aug 27, 2024 at 3:36 PM João Távora <joaotavora@gmail.com> wrote:
>
> I think the fix makes sense.  Please push it to Emacs 30, and thanks
> in advance.
>
> If you can please use this format for the commit message subject,
> which is my more or less usual practice.
>
> Eglot: fix the frobnication (bug#12345)
>
> * lisp/progmodes/eglot.el (eglot-foo): pass foo to eglot-bar.
>
> On Tue, Aug 27, 2024 at 2:41 AM Dmitry Gutov <dmitry@gutov.dev> wrote:
> >
> > X-Debbugs-Cc: joaotavora@gmail.com
> >
> > 1. Server gopls, for example.
> > 2. completion-ignore-case=nil, its default value.
> > 3. Input "clCe" and completions for it like
> >
> >    clockCenterX
> >    clockCenterY
> >
> > Completions are case-sensitive, but completions' highlighting works as
> > if it's insensitive - highlights are on "cl", second "c" and "e",
> > whereas the capital "C" is not highlighted.
> >
> > This fixes it for me:
> >
> > diff --git a/lisp/progmodes/eglot.el b/lisp/progmodes/eglot.el
> > index 844fc634be9..acc197754db 100644
> > --- a/lisp/progmodes/eglot.el
> > +++ b/lisp/progmodes/eglot.el
> > @@ -3260,7 +3260,7 @@ eglot-completion-at-point
> >              (try-completion pattern (funcall proxies)))
> >             ((eq action t)                                 ; all-completions
> >              (let ((comps (funcall proxies)))
> > -             (dolist (c comps) (eglot--dumb-flex pattern c t))
> > +             (dolist (c comps) (eglot--dumb-flex pattern c
> > completion-ignore-case))
> >                (all-completions
> >                 ""
> >                 comps
> >
> >
> >
>
>
> --
> João Távora



-- 
João Távora

bug#72824: Eglot highlights matches incorrectly wrt case sensitivity

[Dmitry Gutov]

On 27/08/2024 17:38, João Távora wrote:
> Also, btw, completion highlighting is naturally borked too.  There is
> no guarantee that that naive flex algorithm is what the server used
> to select those completions.

True. For example, tn this scenario the server uses case-insensitive 
matching, whereas we (by default) match case-sensitively in the 
completion style.

This change should help eliminate the inconsistency between parts of the 
current completion style, but for the next release we might migrate to a 
more "passthrough" one.

Or not, you know. I've seen users request prefix-only matching instead 
of the fuzzy one provided by the language servers, and I'm sure there 
will be ones who prefer to match case-sensitively only too.

bug#72824: Eglot highlights matches incorrectly wrt case sensitivity

[Dmitry Gutov]

Version: 30.1

On 27/08/2024 17:36, João Távora wrote:
> I think the fix makes sense.  Please push it to Emacs 30, and thanks
> in advance.
> 
> If you can please use this format for the commit message subject,
> which is my more or less usual practice.
> 
> Eglot: fix the frobnication (bug#12345)
> 
> * lisp/progmodes/eglot.el (eglot-foo): pass foo to eglot-bar.

No problem. Pushed as

https://git.savannah.gnu.org/cgit/emacs.git/commit/?h=emacs-30&id=9ffa38d3925d70dea704988f68e9e3eca6c7d19c

Thanks, and closing.

bug#72824: Eglot highlights matches incorrectly wrt case sensitivity

[João Távora]

On Tue, Aug 27, 2024 at 11:57 PM Dmitry Gutov <dmitry@gutov.dev> wrote:

> This change should help eliminate the inconsistency between parts of the
> current completion style, but for the next release we might migrate to a
> more "passthrough" one.

What's that?  Modulo bugs (one of which you've fixed recently) the "dumb"
is the most passthrough one, except for the completion-ignore-case.
Do you want to remove that?  I added it at some user request... seemed
reasonable.  Maybe it's not.

> Or not, you know. I've seen users request prefix-only matching instead
> of the fuzzy one provided by the language servers, and I'm sure there
> will be ones who prefer to match case-sensitively only too.

There is no way to do that.  We don't have a say in what the server
gives us.  If the user types foo and the server ignores "foobar" and
prefers "farfromsober" or std::foobar for whatever scoring reason, then
there's nothing we can do about it.  These users are continuously
misinformed  about LSP inability to follow Emacs's completion styles,
despite my frequent admonishing.

João Távora

bug#72824: Eglot highlights matches incorrectly wrt case sensitivity

[Dmitry Gutov]

On 28/08/2024 12:03, João Távora wrote:
> On Tue, Aug 27, 2024 at 11:57 PM Dmitry Gutov <dmitry@gutov.dev> wrote:
> 
>> This change should help eliminate the inconsistency between parts of the
>> current completion style, but for the next release we might migrate to a
>> more "passthrough" one.
> 
> What's that?  Modulo bugs (one of which you've fixed recently) the "dumb"
> is the most passthrough one, except for the completion-ignore-case.
> Do you want to remove that?  I added it at some user request... seemed
> reasonable.  Maybe it's not.

"dumb" does some filtering in its all-completion call (the predicate arg 
is used for its return value, AFAICT).

A "passthrough" style would only add highlighting but not filter.

>> Or not, you know. I've seen users request prefix-only matching instead
>> of the fuzzy one provided by the language servers, and I'm sure there
>> will be ones who prefer to match case-sensitively only too.
> 
> There is no way to do that.  We don't have a say in what the server
> gives us.  If the user types foo and the server ignores "foobar" and
> prefers "farfromsober" or std::foobar for whatever scoring reason, then
> there's nothing we can do about it.

I suppose that can happen if total completions exceed the cutoff number.

> These users are continuously
> misinformed  about LSP inability to follow Emacs's completion styles,
> despite my frequent admonishing.
My observation from the sidelines is that people are still happier to 
have such half-measures available, rather than not: 
https://github.com/emacs-lsp/lsp-mode/issues/1653#issuecomment-1125055989

It might be a matter of expectations.