Re: A few Windows build fixes

[Eli Zaretskii <eliz@gnu.org>]

> From: Stefan Monnier <monnier@iro.umontreal.ca>
> Cc: andrewjmoreton@gmail.com,  emacs-devel@gnu.org
> Date: Mon, 05 Sep 2011 22:49:52 -0400
> 
> >> > It doesn't matter who and how runs "mount -m".  The results of that
> >> > should be used on the C level.  Otherwise, the result will be
> >> I don't think so: we only want to convert cygwin file names when they
> >> "enter" Emacs and everything else (e.g. all primitives that access
> >> files) should not know about them.
> > You cannot possibly hope to plug all the holes this way.  A file name
> > can "enter" Emacs in any number of ways, including as a substring of
> > some string or buffer.  Trying to catch all those loopholes is never
> > going to work reliably.
> 
> We have no chance to plug all hole reliably anyway.

Yes, we do: on the lowest possible level that Emacs handles file names
before it passes them to system APIs.  I pointed out the relevant
function in w32.c where that happens.  If you transform all file names
there, every single feature of Emacs will work as expected by users.

> The closest to
> "handle it everywhere" already exists, it's called cygwin-mount.el.

cygwin-mount is part of the solution, but it isn't all of it.  If we
want a reliable solution, that is (which IMO is what users expect).

> Instead we want something that's less intrusive, even though it will
> miss a few more cases (but by being less intrusive it can work
> better in the cases where it does work).

Something that happens in one or 2 functions cannot be intrusive.