bug#60453: 29.0.60; treesit-range-rules throw an error without tree-sitter

[Eli Zaretskii]

> From: Wilhelm Kirschbaum <wkirschbaum@gmail.com>
> Cc: 60453@debbugs.gnu.org
> Date: Sat, 31 Dec 2022 18:50:31 +0200
> 
> 
> Eli Zaretskii <eliz@gnu.org> writes:
> 
> >> From: Wilhelm Kirschbaum <wkirschbaum@gmail.com>
> >> Date: Sat, 31 Dec 2022 16:53:08 +0200
> >> 
> >> 
> >> With the following code without tree-sitter library:
> >> 
> >> (defvar elixir-ts-mode--treesit-range-rules
> >>   (treesit-range-rules
> >>    :embed 'heex
> >>    :host 'elixir
> >>    '((sigil (sigil_name) @name (:match "^[H]$" @name) 
> >>    (quoted_content)
> >>   @heex))))
> >> 
> >> upon loading the mode I get the following error:
> >> 
> >> treesit-range-rules: Symbol’s function definition is void:
> >> treesit-query-compile
> >> 
> >> This can easily be mitigated with (when 
> >> (treesit-available-p)...)
> >> but think it should function similar to how 
> >> (treesit-font-lock-rules
> >> work.
> >
> > Why does it make sense to protect treesit.el's code with
> > treesit-available-p?  You aren't supposed to use treesit.el 
> > functions
> > when the tree-sitter library is not available.  IOW, Lisp 
> > programs
> > that want to use treesit-range-rules and other functions from
> > treesit.el should make the treesit-available-p test _before_ 
> > that.
> 
> Okay, that makes sense.  I just saw this comment on
> 
> ;; treesit.el#618
> (defun treesit-font-lock-rules (&rest query-specs)
>   ...
>   ;; Other tree-sitter function don't tend to be called unless
>   ;; tree-sitter is enabled, which means tree-sitter must be 
>   compiled.
>   ;; But this function is usually call in `defvar' which runs
>   ;; regardless whether tree-sitter is enabled.  So we need this
>   ;; guard.
>   (when (treesit-available-p)
> 
> As treesit-range-rules also gets called with defvar and it is a
> consistency issue.  I think the reason why this has not popped up
> before is that no other modes I have seen uses treesit-range-rules
> yet and think it will probably catch people off guard in the future.

It's up to Yuan: if he thinks this is a good idea, he should feel free
to add that test.  But it's slippery slope, IMNSHO: we will very soon
find ourselves adding such tests to every treesit.el function, just
because some code somewhere calls that function without a prior test.
IOW, IMO a single case of such callers is not enough to add a test.
But that's me.