Hello, Maxim!

Thanks for your effort in this, some comments with the quotes for
context.

Maxim Cournoyer <maxim.cournoyer@gmail.com> writes:
>> This fails if geiser-guile-load-path does not exist (void-variable).
>> The cleanup removes other guixes, isn't it?  I suggest making the
>> variable buffer-local and forget about hard-coded names. :-)
>
> That's a good suggestion!  I toyed with it and it's a bit tricky but I
> think the v2 patch I'll send as a follow-up does the trick.  My concern
> was also supporting when a user has previously set
> geiser-guile-load-path in their .emacs init file, e.g.:
>
> (setq geiser-guile-load-path (list (expand-file-name "~/src/guix")
>                                    (expand-file-name "~/src/shepherd")))
>
> That would mean their entries don't get cleaned up (it seems this
> doesn't matter in my latest tests with the v2 patch though!).

That cleanup seems to me responsibility of that .emacs maintainer
instead of something to take into account in .dir-locals. ;-)

>> (eval . (setq guix-directory
>>           (locate-dominating-file default-directory ".dir-locals.el")))
>> (eval . (when (boundp 'geiser-guile-load-path)
>
> This check makes it so that if geiser-guile-load-path is not already
> defined, nothing happens.  It is likely that this is the case, as when
> relying on just Geiser's autoloads, it is not loaded.  The user would
> have to either set explicitly before hand or call (require
> 'geiser-guile), which we can't rely on.  But we can drop this check.

You're right, as you can only bind the keys and enable it when used, not
at file load as I do, great catch. :-)

> One thing that worried me was the %load-compiled-path not appearing in
> the order defined from guile-geiser-load-path, but in my latest tests as
> mentioned above it didn't matter.

With the right directories (meaning no-conflicts between modules) it
shouldn't matter, but it's weird.  Looking into geiser-guile.el the load
path is provided to guile through -L parameters in
geiser-guile--parameters, and the extra path for Geiser code is added
with geiser-guile--set-geiser-load-path (that's why it is not added to
%load-compiled-path)...  I'd have to check Guile to be sure why are the
differences, but they seem harmless.

> +               (unless (fboundp 'geiser-guile-load-path)
> +                 (defvar geiser-guile-load-path '()))

This checks the function definition, not the variable, as Emacs Lisp has
two separate namespaces.

> I ended up using `cl-pushnew' here instead of push, as otherwise
> repeated entries were accumulated.

I wanted to avoid exactly that check, as the variable should end up with
duplicated entries only when you call it twice from the same buffer, how
could that happen?

> +               (make-local-variable 'geiser-guile-load-path)
> +               (cl-pushnew root-dir* geiser-guile-load-path
> +                           :test #'string-equal)))))

If there is some way this may happen, then this call is OK, but I'd try
to stay with a cheaper push unless it's really needed, as O(1) < O(n),
for almost every n. :-)

Again, thank you very much for taking care of this, as it would make
life easier for everybody of us who uses Geiser.

Happy hacking!
Miguel