Re: [Emacs-diffs] trunk r116230: Fix bug #16558 with w32-shell-execute on remote file names.

[Eli Zaretskii]

> From: Stefan Monnier <address@hidden>
> Cc: address@hidden
> Date: Sun, 02 Feb 2014 21:33:18 -0500
> 
> > Btw, find-file-name-handler returns nil for file-exists-p when the
> > file name specifies a compressed file, so I'm not sure what bothered
> > you in the first place.
> 
> Ha!  Indeed, I had never noticed that `operations' property!

I'm not sure this is an accident.  IMO, any file handler that works
with local files should do the same.

> Then I guess the current code is OK (tho it's still kind of a hack).

I prefer "clever engineering solution".

> >> And the reason we do that is because some file names are "normal" and
> >> others refer to non-files according to some w32 feature which can map
> >> them to some other tools.
> > The reason is described in the comment: if the file name is not
> > absolute and its name is not relative to the directory passed to the
> > system API, the API will fail.
> 
> I'm having trouble understanding the above: can you give an example of
> a file which is neither absolute nor relative to the directory passed
> to the system API?

I think it happened to me with shr.el, since it doesn't set the
current directory of the buffer where it renders an HTML document.

> >> I don't know that w32 feature at all, so it's hard for me to figure out
> >> what should be done, but it seems like file-exists-p is not the right
> >> thing to do anyway since the file name might be "normal" but refer to
> >> a file that doesn't exist yet.
> > Yes, that might happen, but then the file will be created in a
> > directory other than what the user expects it to, perhaps.
> 
> So it would be a problem, right?

Yes; see the comments in the code.  But I don't see any good solution
to this problem; do you?

In any case, this is not worse than what happened before the change
that introduced the call to expand-file-name: then, the ShellExecute
API was always called with a file name as passed to w32-shell-execute.

> Maybe instead of Ffile_exists_p a better option is to use a w32 system
> call along the lines of faccess or stat (after all, this presumed file
> name will be passed to the OS rather than to Emacs functions, so it
> shouldn't pay attention to file-name-handlers and things like that,
> right?).

That's exactly what the current code does:

  handler = Ffind_file_name_handler (absdoc, Qfile_exists_p);
  if (NILP (handler))
    {
      Lisp_Object absdoc_encoded = ENCODE_FILE (absdoc);

      if (faccessat (AT_FDCWD, SSDATA (absdoc_encoded), F_OK, AT_EACCESS) == 0)
        document = absdoc_encoded;
      else
        document = ENCODE_FILE (document);
    }
  else
    document = ENCODE_FILE (document);

> And to deal with "not yet created files" we shouldn't check the file
> itself but its directory.

Its directory is just the default-directory of the current buffer, so
I think we are covered:

  Lisp_Object current_dir = BVAR (current_buffer, directory);;

> >> So, how does w32 decide whether a file name is "normal" or not?
> > We cannot know: it's up to the application that is associated with
> > DOCUMENT and OPERATION.
> 
> So we don't actually know if the string we have is really a file name,
> and Ffile_exists_p is used to try and guess whether that's the case
> (because we need to adjust it somehow if that's the case)?

Indeed.