Would (Eval-when-compile (require 'treesit)) eliminate the need for (declare-function)s?

Would (Eval-when-compile (require 'treesit)) eliminate the need for (declare-function)s?

[Yuan Fu]

I’ve made the mistake of forgetting to add (declare-function treesit-xxx) when using functions from treesit.c far too many times (sorry!). Since treesit.el contains declare-function’s for every function in treesit.c, if we use (eval-when-compile (require ’treesit)), would it eliminate the need for adding declare-function’s? Should we do that?

Yuan

Re: Would (Eval-when-compile (require 'treesit)) eliminate the need for (declare-function)s?

[Ruijie Yu via Emacs development discussions.]

Yuan Fu <casouri@gmail.com> writes:

> I’ve made the mistake of forgetting to add (declare-function treesit-xxx) when
> using functions from treesit.c far too many times (sorry!). Since treesit.el
> contains declare-function’s for every function in treesit.c, if we use
> (eval-when-compile (require ’treesit)), would it eliminate the need for adding
> declare-function’s? Should we do that?
>
> Yuan

According to Eli
(https://lists.gnu.org/archive/html/emacs-devel/2023-03/msg00673.html),
no, it wouldn't work.

[ Also, for some reason I cannot find the message using the msgid,
83ilezegs1.fsf@gnu.org? ]

-- 
Best,


RY

Re: Would (Eval-when-compile (require 'treesit)) eliminate the need for (declare-function)s?

[Philip Kaludercic]

Ruijie Yu via "Emacs development discussions." <emacs-devel@gnu.org>
writes:

> Yuan Fu <casouri@gmail.com> writes:
>
>> I’ve made the mistake of forgetting to add (declare-function treesit-xxx) when
>> using functions from treesit.c far too many times (sorry!). Since treesit.el
>> contains declare-function’s for every function in treesit.c, if we use
>> (eval-when-compile (require ’treesit)), would it eliminate the need for adding
>> declare-function’s? Should we do that?
>>
>> Yuan
>
> According to Eli
> (https://lists.gnu.org/archive/html/emacs-devel/2023-03/msg00673.html),
> no, it wouldn't work.

Strictly speaking, he was referring to removing `declare-function',
without replacing it with anything else.  Replacing that with

  (eval-when-compile (require 'treesit))
        ^
        or "eval-and-compile"?

would be a different matter, since that file contains all the treesit
definitions, which I *believe* should be inherited by the requiring
file.

> [ Also, for some reason I cannot find the message using the msgid,
> 83ilezegs1.fsf@gnu.org? ]

[ I could access the message, but I am accessing the list via NNTP ]

Re: Would (Eval-when-compile (require 'treesit)) eliminate the need for (declare-function)s?

[Eli Zaretskii]

> From: Yuan Fu <casouri@gmail.com>
> Date: Sun, 16 Apr 2023 20:12:22 -0700
> 
> I’ve made the mistake of forgetting to add (declare-function treesit-xxx) when using functions from treesit.c far too many times (sorry!). Since treesit.el contains declare-function’s for every function in treesit.c, if we use (eval-when-compile (require ’treesit)), would it eliminate the need for adding declare-function’s? Should we do that?

Even if it does work reliably, I'm not sure we want that.  It is
better to have declare-function where the function is used, as that
makes maintenance easier and less error-prone.

Lars, Stefan: WDYT?

Re: Would (Eval-when-compile (require 'treesit)) eliminate the need for (declare-function)s?

[Stefan Monnier]

>> I’ve made the mistake of forgetting to add (declare-function treesit-xxx)
>> when using functions from treesit.c far too many times (sorry!). Since
>> treesit.el contains declare-function’s for every function in treesit.c, if
>> we use (eval-when-compile (require ’treesit)), would it eliminate the need
>> for adding declare-function’s?

I hope it doesn't: `declare-function` is supposed to affect only the
"current scope", like `(defvar <VAR>)`.

>> Should we do that?

IIUC one way to look at this is that `treesit.el` wants to "re-export"
the functions provided by `treesit.c` so users of `treesit.el` don't
need to know whether the function they call is implemented in C or
in ELisp?

That makes a lot of sense, and I don't think we have a good way to do
that currently :-(

I suspect in most cases we avoid the problem in one of two ways:
- Always define the C function (rather than being conditional on some
  compilation flag as is the case for `treesit.c` functions).
- Export from C only "internal functions" so all the non-internal API is
  implemented in ELisp.

> Even if it does work reliably, I'm not sure we want that.  It is
> better to have declare-function where the function is used, as that
> makes maintenance easier and less error-prone.

But it makes sense for packages that do

    (require 'treesit)

to be able to use those functions implemented in `treesit.c` without
bothering with `declare-function`.


        Stefan

Re: Would (Eval-when-compile (require 'treesit)) eliminate the need for (declare-function)s?

[Eli Zaretskii]

> From: Stefan Monnier <monnier@iro.umontreal.ca>
> Cc: Yuan Fu <casouri@gmail.com>,  Lars Ingebrigtsen <larsi@gnus.org>,
>   emacs-devel@gnu.org
> Date: Mon, 17 Apr 2023 13:25:26 -0400
> 
> > Even if it does work reliably, I'm not sure we want that.  It is
> > better to have declare-function where the function is used, as that
> > makes maintenance easier and less error-prone.
> 
> But it makes sense for packages that do
> 
>     (require 'treesit)
> 
> to be able to use those functions implemented in `treesit.c` without
> bothering with `declare-function`.

That's what I'm not sure about.

Re: Would (Eval-when-compile (require 'treesit)) eliminate the need for (declare-function)s?

[Yuan Fu]

> 
> IIUC one way to look at this is that `treesit.el` wants to "re-export"
> the functions provided by `treesit.c` so users of `treesit.el` don't
> need to know whether the function they call is implemented in C or
> in ELisp?

Not quite IMO, see below.

>> 
>>> Even if it does work reliably, I'm not sure we want that.  It is
>>> better to have declare-function where the function is used, as that
>>> makes maintenance easier and less error-prone.
>> 
>> But it makes sense for packages that do
>> 
>>    (require 'treesit)
>> 
>> to be able to use those functions implemented in `treesit.c` without
>> bothering with `declare-function`.
> 
> That's what I'm not sure about.

If emacs isn’t built with tree-sitter it doesn’t make much sense to use functions in treesit.c anyway. The real problem is that some files need to be byte-compiled even when emacs isn’t built with tree-sitter. And it’s easy to miss adding declare-function’s. 

Is there any other way to help the situation that we can do in Lisp? Ie, without making all functions always defined in treesit.c? Because I played it in my head a bit, and it looks really messy.

Yuan