Re: RFA: Support Windows extended error numbers in safe_strerror

[Daniel Jacobowitz <drow@false.org>]

On Sun, Feb 05, 2006 at 04:50:36PM -0500, Christopher Faylor wrote:
> On Sun, Feb 05, 2006 at 03:22:22PM -0500, Daniel Jacobowitz wrote:
> >On Sun, Feb 05, 2006 at 10:19:12PM +0200, Eli Zaretskii wrote:
> >> > The shell is needed for the "shell" command.  The shell is not used for
> >> > running programs.
> >> 
> >> The latter is unlike on Posix platforms, right?
> >
> >That is an implementation artifact that, IMO, we desperately need to
> >move away from.  It's a continual source of trouble from e.g.
> >unexpected variable settings in shell initialization files, and it's
> >an aggravating (to me) external dependency for GDB.
> 
> I did try to implement this artifact in Cygwin once but I never got
> it right.
> 
> Isn't the shell used to allow redirection and shell-like quoting from
> the gdb command-line?  Lack of redirection, at least, is an occasional
> complaint about gdb-under-cygwin.

That's right; and we certainly can't ditch that support.  But what we
do today is pass the command line arguments directly to a shell, which
means all sorts of worrying about quoting, and also means that we get
Bourne shell behavior if the user has an appropriate $SHELL, or C shell
if they're using csh/tcsh, et cetera.  I don't know if anyone takes
advantage of that as a feature, but it's sure confusing!

I keep telling myself we ought to somehow get rid of this, but I'm
fuzzy on the details.

-- 
Daniel Jacobowitz
CodeSourcery