[Emacs-diffs] master a59839d: file-truename now uses realpath for local files

[Emacs-diffs] master a59839d: file-truename now uses realpath for local files

[Eli Zaretskii]

Paul,

I had to revert this commit, as it had multiple issues:

 . It breaks the MS-Windows build, because there's no realpath (and we
   cannot use the Gnulib replacement, due to non-ASCII file names
   Gnulib doesn't support well on Windows).
 . Even when I modified file--truename to always signal an error on
   MS-Windows, Emacs wouldn't build, probably because the
   corresponding change in files.el has another mistake.  Did you test
   the build when realpath signals an error?
 . I don't think it's a good idea to favor GNU/Linux systems by
   punishing the rest.  We should have a better way than signaling an
   error on platforms where realpath doesn't work well enough.  And in
   any case, using Gnulib's realpath will not make file-truename
   faster on those other systems.
 . The code of file--truename incorrectly looked for the trailing
   slash ion the encoded file name, something you can only do reliably
   if the file-name encoding is UTF-8, because other encodings don't
   guarantee a single slash is never part of a multibyte sequence.
 . Last, but not least: there was no documentation of the new
   primitive.

For all of these reasons, I think we need to talk about the issue
before implementing it, and the implementation, if we will have one,
should resolve all of the problems mentioned above.

Re: [Emacs-diffs] master a59839d: file-truename now uses realpath for local files

[Eli Zaretskii]

> Date: Sat, 07 Sep 2019 10:27:05 +0300
> From: Eli Zaretskii <eliz@gnu.org>
> Cc: emacs-devel@gnu.org
> 
> For all of these reasons, I think we need to talk about the issue
> before implementing it, and the implementation, if we will have one,
> should resolve all of the problems mentioned above.

One possible solution would be to have file--truename bound only on
platforms which have the realpath function natively in their libc.
Then files.el could test that with fboundp.  I see no reason to
replace a loop in files.el by a loop in Gnulib's realpath emulation,
so we might as well not pretend realpath exists when it doesn't.