Re: AS_EXECUTABLE_P again -- summary (was: bug in texi2dvi, and hack patch)

[Stepan Kasal]

Hello,

On Sat, Jan 22, 2005 at 06:23:25PM +0200, Eli Zaretskii wrote:
> > I proposed to change the implementation of AS_EXECUTABLE_P on platforms
> > supporting `test -x' from
> >        test -f $1 && test -x $1
> > to
> >        test -x $1 && test ! -d $1
> > 
> > But Paul has pointed out the change can bring problems eg. with named
> > pipes in /tmp (some people have /tmp in their path).
> 
> What problems does that cause?  I don't see how named pipes are
> related to looking for executables.

If there was a named pipe /tmp/foo with an executable bit set, the test
could misclasify it as an executable file.

(FYI, I was referring to Pauls mail archived at
http://lists.gnu.org/archive/html/autoconf/2005-01/msg00079.html )

> > 2) On Cygwin, both `test -x foo' and `test -f foo' look for foo.exe.
> 
> That's a bug, IMHO.

I'm not so sure.  In any case, I take this as a fact until someone
convinces them to change the behaviour.

> > I see two options:
> > - change DJGPP's `test -f foo' to search also for foo.exe
> >     (as suggested by Paul)
> > - make a DJGPP-specific change to AS_EXECUTABLE_P
> > 
> > But I don't think any of them is worth the time it would take.
> 
> So what _are_ you suggesting for AS_EXECUTABLE_P to do?

I'm happy with the current code.  (Part of it is my work. :-)
It does `test -f $1 && test -x $1' on most platforms.
See the autoconf CVS, file lib/m4sugar/m4sh.m4, for details.

> Also, you didn't tell what do "test -f" and "test -x" do in non-Cygwin
> Windows ports of Bash and Sh-utils.  I don't think we should ignore
> those.

I don't know what they do.  You are right, I should correct my previous
statement:

The problem might affect running non-autoconf generated m4sh scripts
on non-Cygwin Windows ports (eg. DJGPP).

Developers of these ports can submit a patch.  And the patch should not
affect Unix-style patforms, beacause of the Paul's argument cited above.

Have a nice day,
        Stepan