Re: Use (eval-when-compile 'treesit) to save us from writing declare-function forms

[Yuan Fu]

> On Dec 11, 2024, at 10:43 PM, Eli Zaretskii <eliz@gnu.org> wrote:
> 
>> From: Yuan Fu <casouri@gmail.com>
>> Date: Wed, 11 Dec 2024 22:05:56 -0800
>> Cc: Andrea Corallo <acorallo@gnu.org>,
>> Eli Zaretskii <eliz@gnu.org>,
>> emacs-devel@gnu.org
>> 
>> 
>> 
>>> On Dec 11, 2024, at 4:23 PM, Dmitry Gutov <dmitry@gutov.dev> wrote:
>>> 
>>> On 12/12/2024 01:38, Andrea Corallo wrote:
>>>> The suggestion of having the C functions defined to nops when
>>>> HAVE_TREE_SITTER is not defined is cleaner in principle, but I think
>>>> there are two downsides:
>>>> - We don't tipically do it this way, so it would be an exception
>>> 
>>> We also don't typically require a caller to have a 'declare-function' for 
>>> every function they use from a package that they (require ...) already.
>>> 
>>>> - Also we should do it for all the functions in treesit.c, otherwise we
>>>>  don't solve the original problem of introducing warnings by mistake on
>>>>  non treesit builds.
>>> 
>>> Sure. All non-static ones anyway.
>> 
>> I tried to define all the C functions and it’s not easy to do: doing it in C 
>> seems pretty tedious, we’d need DEFUN form and defsubr for each function (is 
>> there a easier way?). If done in lisp, we need to do it in a separate file 
>> and _load_ it before compiling treesit.el and any other package that uses 
>> tree-sitter. I’m not familiar with Emacs’s build process so I don’t know how 
>> and where to add such a file and make Emacs load it on startup.
>> 
>> Can someone enlighten me on how to define these functions in either C or 
>> lisp?
>> 
>> I think defining them as no-op for Emacs w/o tree-sitter is a good move, 
>> because otherwise if some user compiles an Emacs w/o tree-sitter and 
>> compiles a bunch of packages, and one of the package happens to be based on 
>> tree-sitter, they’ll see a bunch of warnings unless the package author added 
>> declare-function forms or the proposed macro. It’s better to have 
>> tree-sitter functions always defined and save everybody some hassle.
> 
> FWIW, I'd rather we didn't go this way, for the reasons Andrea
> explained (and I agreed).  Inventing some new technique or tricks just
> for this one case, when we don't do anything like that in any other
> similar case, doesn't sound like a good idea, because it would mean
> one more subtle protocol to support in Emacs development.  It gets in
> the way while reviewing patches, for example, because it is not easy
> to remember when special conventions must be observed.  E.g., should
> someone come up with a new DEFUN in treesit.c, someone should remember
> that a no-op function by the same name should be defined at the same
> time.  Or we'd need to #ifdef out the body of each DEFUN, which makes
> the code harder to read.
> 
> Can we find some other way of making this easier, without defining
> no-op functions in treesit.c?

Cool. I don’t want to waste more of everybody’s time on this, so I pushed the 
original patch (sans eval-when-compile) which I believe everybody is ok with.

Yuan