Re: Last use of defadvice in Emacs

[Alan Mackenzie]

Hello, Stefan.

On Thu, Apr 07, 2022 at 14:37:14 -0400, Stefan Monnier wrote:
> > I want more than to "perform the test" at compile time.  I want a Lisp
> > form that will check whether that variable is bound, and if so, not even
> > compile the sub-form.  Something like C's #ifndef preprocessor form.  It
> > would look something like

> >     (hash-if (not
> >               (boundp 'font-lock-extend-after-change-region-function))
> >              (progn
> >           (defmacro c-advise-fl-for-region (function) .....)
> >               (c-advise-fl-for-region ....)
> >           ....
> >           ))

> > ..  Here the progn form would be neither evaluated nor compiled if that
> > font-lock-... variable were boundp.  We don't have this at the moment.
> > Not that it's all that important in the current case, but it might be
> > handy to have, perhaps, in other version dependent code.

> The patch I send does obey the requirement that the `defadvice` will be
> "neither evaluated nor compiled" if the variable exists at compile-time.

That's different from, and less demanding than, the "challenge" I set, I
think.  It's not at all important to Emacs's health, but .....

> It is not a separate "hash-if", OTOH.  We can define such a "hash-if",
> as seen for example in the `url-http-ntlm` GNU ELPA package:

>     (defmacro url-http-ntlm--if-when-compile (cond &rest body)
>       (declare (debug t) (indent 1))
>       (when (eval cond)
>         `(progn ,@body)))

>     ;; Remove authorization after redirect.
>     (url-http-ntlm--if-when-compile
>         (and (boundp 'emacs-major-version)
>        (< emacs-major-version 25))
>       ...
>       ... Various code, including, incidentally, a `defadvice` ...
>       ...)

Here, a piece of `byte-code' gets compiled for the
url-http-ntlm--if-when-compile call.  This is useless code without any
useful function.  (I've just tried it.)

The action I asked for (or, at least, meant to ask for) was for _no_ code
to get compiled when a condition was not met.  I still believe this is
not possible at the moment.  But the C preprocessor can do it with
#ifndef.

One use for this could be writing unit tests in the same file.el as what
they're testing.  They would get compiled only for a test run.

> Along similar lines, there's AUCTeX's `TeX--if-macro-fboundp`:

>     (defmacro TeX--if-macro-fboundp (name then &rest else)
>       "Execute THEN if macro NAME is bound and ELSE otherwise.
>     Essentially,
    
>       (TeX--if-macro-fboundp name then else...)
    
>     is equivalent to
    
>       (if (fboundp 'name) then else...)
    
>     but takes care of byte-compilation issues where the byte-code for
>     the latter could signal an error if it has been compiled with
>     emacs 24.1 and is then later run by emacs 24.5."
>       (declare (indent 2) (debug (symbolp form &rest form)))
>       (if (fboundp name)             ;If macro exists at compile-time, just 
> use it.
>           then
>         `(if (fboundp ',name)               ;Else, check if it exists at 
> run-time.
>              (eval ',then)                  ;If it does, then run the then 
> code.
>            ,@else)))

I haven't been able to find AUCTeX's source code yet, but I'm not sure
this macro would meet the "no code at all" criterion, either.

> No such macro has reached ELisp's core yet, probably because the
> precise requirements tend to be subtly different (often depending on
> the authors's own preferences about what they want to consider as
> legitimate or important use-cases, as is the case in the cc-mode code:
> what should happen when compiled on Emacs-NN but run on Emacs-MM?).

>         Stefan

-- 
Alan Mackenzie (Nuremberg, Germany).