Should write-contents-functions be permanent-local?

Should write-contents-functions be permanent-local?

[Joseph Turner]

In hyperdrive.el, we set write-contents-functions in a local minor mode
so that buffer contents can be written to a hyperdrive "file." In our
use-case, we want write-contents-functions to stay the same even when
the user manually changes the buffer's major mode. AFAIK, there's no way
to make a variable permanent-local on a per-buffer basis, and we don't
dare to make write-contents-functions permanent-local globally.

The solution we came up with in hyperdrive.el is to hook into
after-change-major-mode-hook with a function (with permanent-local-hook)
that sets write-contents-functions after kill-all-local-variables:

https://git.sr.ht/~ushin/hyperdrive.el/tree/master/item/hyperdrive.el#L320

Does anyone have a better solution?

WDYT about making write-contents-functions permanent-local?

Joseph

Re: Should write-contents-functions be permanent-local?

[Eli Zaretskii]

> From: Joseph Turner <joseph@breatheoutbreathe.in>
> Date: Mon, 14 Aug 2023 23:35:44 -0700
> 
> WDYT about making write-contents-functions permanent-local?

That cannot be done, sorry.  Some modes set this to mode-specific
values, you can find this even in the Emacs source tree.  But even
without those examples, a variable that were not permanent-local for
so long cannot be suddenly made permanent-local, because it will
almost certainly break something somewhere.

In any case, if you are okay with making the variable permanent-local,
then why not just

  (put 'write-contents-functions 'permanent-local t)

in your package, and be done with it?  The effect of the above is the
same global effect as if we would do that by default in Emacs, right?

Re: Should write-contents-functions be permanent-local?

[Joseph Turner]

Eli Zaretskii <eliz@gnu.org> writes:

>> From: Joseph Turner <joseph@breatheoutbreathe.in>
>> Date: Mon, 14 Aug 2023 23:35:44 -0700
>>
>> WDYT about making write-contents-functions permanent-local?
>
> That cannot be done, sorry.  Some modes set this to mode-specific
> values, you can find this even in the Emacs source tree.  But even
> without those examples, a variable that were not permanent-local for
> so long cannot be suddenly made permanent-local, because it will
> almost certainly break something somewhere.

Thank you for the explanation!

> In any case, if you are okay with making the variable permanent-local,
> then why not just
>
>   (put 'write-contents-functions 'permanent-local t)
>
> in your package, and be done with it?  The effect of the above is the
> same global effect as if we would do that by default in Emacs, right?

I was concerned that would break things. (whereas I thought if we made
the change in Emacs, we'd have a chance to ensure that it did not.)

Probably best not to change things - our current solution seems to work:

(defun hyperdrive--hack-write-contents-functions ()
  (cl-pushnew #'hyperdrive--write-contents write-contents-functions))
(put 'hyperdrive--hack-write-contents-functions 'permanent-local-hook t)

(define-minor-mode hyperdrive-mode
  "Minor mode for buffers opened from hyperdrives."
  (if hyperdrive-mode
      (progn
        (cl-pushnew #'hyperdrive--write-contents write-contents-functions)
        (add-hook 'after-change-major-mode-hook
                  #'hyperdrive--hack-write-contents-functions nil 'local))
    (setq-local write-contents-functions
                (remove #'hyperdrive--write-contents write-contents-functions))
    (remove-hook 'after-change-major-mode-hook
                 #'hyperdrive--hack-write-contents-functions 'local)))