Re: [Patch] autoconf.sh - no FD + chmod at right time

[Tim Van Holder]

> Tim> Which one? I tried 3, 5 and 6 as well; they gave the same errors.
> 
> Can't we go higher?
Doesn't seem to do much good.  Odd though - a simple shell script doing
simple redirection works for FD's 3 through 15.  The same script even
works if run from Perl.
For some reason the same is not true for autoconf; no matter what FD I
use, I get the 'bad file descriptor' message, even though the output is
correct.  I'll have to dig deeper at some point, but I'm not sure
whether it's a problem in Perl, bash, or the C library.

> Tim> Unless using a temp file is less portable, I don't see a problem
> Tim> with that.
> 
> I wouldn't be surprised that for instance you broke -o for traces.
> The use of FD was carefully crafted to avoid using tmp files, and to
> factor most of the code.

Well, the only real problem is that it breaks the testsuite; autoupdate
runs correctly and the output is correct.  It just emits an error message
it shouldn't.
I guess I'll just put a notice in the DJGPP readme so people won't freak
out about the EBADF error message, and I'll probably skip all tests
using autoupdate on DJGPP.