Re: master 19a3b499f84: ; * lisp/loadup.el: Don't prohibit advice when ls-lisp is loaded.

[Stefan Monnier]

> Uh.  I'm still struggling to understand all bits, but here are already
> some comments.
>
> - The docstring of `ls-lisp--insert-directory' still makes references
>   to its previous advice nature, this probably should be ironed out.

I haven't really attacked the task of updating the docs yet (neither
docstrings nor manual).

> - The docstring of `insert-directory-program' does not very explicitly
>   mention that its value could be nil.  (Nothing introduced by your
>   patch, just noticed it.)
>
> - And the custom spec of `insert-directory-program' isn't prepared for
>   nil as well, but that is probably OK since it should be used for a
>   generated default value only?

While I did change the code so that setting `insert-directory-program`
to nil forces the use of ls-lisp, I did it in a minimalist way so far.
I'm not sure if we want to encourage such a setting.  In a sense, it
"comes for free" and it's nice to make the code robust against such
a setting.

> - AFAICT you call `files--insert-directory-program' as a predicate
>   only.  Probably we should rename it to `...-p' or
>   `files--use-insert-directory-program'?

Yeah, it wasn't intended that way but that's indeed how things turned
out.

> - Not sure whether that makes a difference, but: Without your patch the
>   advice was conditional on whether ls-lisp was loaded, now the usage of
>   `ls-lisp--insert-directory' is conditional on whether
>   `ls-lisp-use-insert-directory-program' is bound.  What if a user has
>   a .emacs with a customization or `setq' of `ls-l-u-i-d-p' to nil, which
>   she uses both on Windows and Unix?

Indeed, this is a change in behavior.
Note that those users already encountered this behavior in the past when
their non-Windows system loaded `ls-lisp` at run time, e.g. because of the
use of Tramp or via `help-enable-autoload`, or ...

> - Your fix in `file-expand-wildcards'
>
>           (if (equal "" nondir)
>               (push (or dir nondir) contents)
>
>
>   is hard to digest at 11PM, so a comment won't hurt here, I guess.

Good point, thanks.

>   Plus it seems to violate the order promise done in the docstring of
>   the function ("... sorted in the `string<' order").

Oh, indeed I completely missed that, thank you.

> - Now the logic gets harder and harder, so I may be wrong here.  The
>   docstring of `dired-insert-directory' seems to handle FILE-LIST
>   exclusive to WILDCARD, as if WILDCARDS do not get expanded if
>   FILE-LIST is non-nil:
>
>     If FILE-LIST is non-nil, list only those files.
>     Otherwise, if WILDCARD is non-nil, expand wildcards;
>
>   but your patch changes that, no?

Hmm... IIUC the current code treats a "*" in the directory part as
a wildcard regardless of FILE-LIST and WILDCARD, and my patch tries to
fix that so as to better obey the docstring which says that "*" should
only be treated as wildcards if WILDCARD is non-nil and FILE-LIST
is nil.

Hopefully this will get more clear once I clean up the patch and cut it
into meaningful commits.


        Stefan