bug#10733: 24.0.93; w32 file truncation

[Eli Zaretskii <eliz@gnu.org>]

> From: Óscar Fuentes <ofv@wanadoo.es>
> Cc: lekktu@gmail.com,  Takaaki.Ota@am.sony.com,  10733@debbugs.gnu.org
> Date: Tue, 07 Feb 2012 06:22:12 +0100
> 
> Eli Zaretskii <eliz@gnu.org> writes:
> 
> >> Maybe it can be integrated in the
> >> 
> >> if (!(NILP(Vw32_get_true_file_attributes) ...
> >> 
> >> hence reusing the calls to CreateFile and GetFileInformationByHandle and
> >> shortening the patch, but as I don't know what
> >> Vw32_get_true_file_attributes does, preferread to follow the safe way.
> >
> > You did right: w32-get-true-file-attributes can be set by the user to
> > nil, if she wants her file ops faster.
> 
> I was thinking on something like
> 
> diff --git a/src/w32.c b/src/w32.c
> index 3d3d334..418be63 100644
> --- a/src/w32.c
> +++ b/src/w32.c
> @@ -3447,8 +3447,12 @@ stat (const char * path, struct stat * buf)
>  	}
>      }
>  
> -  if (!(NILP (Vw32_get_true_file_attributes)
> -	|| (EQ (Vw32_get_true_file_attributes, Qlocal) && is_slow_fs (name)))
> +  buf->st_size = 0;
> +
> +  if ((!(NILP (Vw32_get_true_file_attributes)
> +         || (EQ (Vw32_get_true_file_attributes, Qlocal) && is_slow_fs (name)))
> +       || ((wfd.dwFileAttributes & FILE_ATTRIBUTE_REPARSE_POINT) &&
> +           (wfd.dwReserved0 == IO_REPARSE_TAG_SYMLINK)))

Then what made you hesitate?  This approach looks fine to me.

> >> if is-symlink?
> >>   use fstat
> >> fi
> >
> > Since fstat is also reimplemented, I'd rather do what it does inline.
> >
> > For that, we need to know which other attributes are reported
> > different.  Or maybe just test for the reparse point up front and do
> > all the work for the target instead.
> 
> Since Emacs' fstat reimplementation is based on
> GetFileInformationByHandle, and that the handle points to the linked
> file (CreateFile follows the link unless told otherwise), we should be
> safe delegating all work to `fstat' when a symlink is detected on `stat'
> (the executable bit must be setted on `stat', but that's no problem.)

Please compare w32.c's `fstat' with `stat', and you will see that the
former does much less than the latter, even with information that can
be gotten by the handle.  To go the way you suggest, we need first to
make `fstat' a proper subset of `stat'.  (I never had time to do it,
and since `fstat' is used much less that `stat' in Emacs, more
important jobs came first.)