Dmitry Gutov <dmitry@gutov.dev> writes:

> On 25/11/2023 10:33, Andrey Listopadov wrote:
>
>>> And here's another aspect: the default built-in theme doesn't
>>> distinguish many of the faces (and the same is true for many 
>>> other
>>> built-in themes). E.g. it doesn't distinguish 
>>> variable-name-face from
>>> variable-use-face or function-name-face from 
>>> function-call-face.
>> I'm wondering if font-lock.el needs a bit more generic faces, 
>> as
>> packages often define their own faces, that aren't supported by 
>> themes
>> in any way.  Again, the example with elixir-mode isn't to bash 
>> the
>> developers, but before 2019 elixir-mode (not elixir-ts-mode) 
>> defined a
>> few faces with explicit colors.  Here's a commit that fixed 
>> that
>> https://github.com/elixir-editors/emacs-elixir/commit/f101c676cc9485aa22ec088a71d8afc72cda3d58
>> but before it, `elixir-atom-face' and `elixir-attribute-face' 
>> were
>> `RoyalBlue4' and `MediumPurple4' no matter what theme you were 
>> using.
>> IIRC the CIDER package also defines some faces like that, so 
>> it's
>> somewhat common.
>
> As long as the faces are for unusual contexts and have some 
> fallbacks
> (or preferably inherit from some of the core ones), that's fair
> practice.
>
>> I can't come up with missing faces, and most modes I use define 
>> extra
>> faces in terms of inheritance to the inbuilt faces,
>
> Right.
>
>> but maybe
>> font-lock-symbol-face is worth including, as some languages may 
>> want to
>> distinguish these like elixir does right now with 
>> `elixir-ts-atom-face'.
>
> I agree we could add more. E.g. a face like that could 
> automatically
> be used for "keywords" in Elisp (and Clojure, and other Lisps) 
> and
> "symbols" in Elixir in Ruby.
>
> What makes me pause is naming: the terminology is a mess here 
> across
> languages. "symbols" usually mean something else in Emacs (and 
> in Lisp
> languages in general), whereas "keywords" mean something else 
> across
> most other languages. Using the name font-lock-symbol-face is 
> bound to
> cause confusion at least across Lisp programmers. Luckily,
> 'font-lock-keyword-face' is already taken, so we don't have to
> consider this alternative (which would puzzle the rest of the
> programming world).
>
> The docstring of 'font-lock-constant-face' says "Face name to 
> use for
> constant and label names", but a name 'font-lock-label-name' 
> sounds
> pretty bland... OTOH, there are labels in C, but nothing with 
> that
> particular name in Elixir, Ruby or Lisp (aside from one macro, I
> suppose).


Here is a patch to address numerous issues flagged on Elixir 
slack,
Github and in this thread.  It will not be perfect, but since the
changes are pretty large I want to get this in and then we can 
pick on
specific issues afterwards if that makes sense?

I am making the assumption that it is okay to rename custom faces 
as
elixir-ts-mode is only for 30.

One thing I tried to get right is to ensure that each level works
relatively well, which means a bit more brute forcing queries.  I 
have
not seen a major performance issue on massic Elixir files, so 
think its
fine.