Re: No, I *don't* HAVE_IO_H!

[Pavel Roskin]

Hello, David!

> I find it highly interesting to note that that the first test for the
> <vfork.h> header succeeded, but the very next test configure performed,
> to determine whether vfork() actually works, failed because <vfork.h>
> could not be found.

Thank you for investigating this problem.

The old versions of Autoconf including the latest released version,
Autoconf-2.13, determine whether headers exist by checking whether the
proprocessor prints anything to stderr.

The current CVS version of Autoconf uses a more complicated way to check
for the headers. If the preprocessor can indicate missing includes with
the exit status, the warnings are not honored.

I'm almost sure that the CVS version of Autoconf will work properly for
you. I'm afraid that Autoconf maintainers are not really interested in
disscussing the obsolete code.

However, something must be broken on your system. My guess is that the
warnings from your compiler are either lost or go to stdout instead of
stderr.

Try this and make sure that your results are the same:
$ cat >foo.c <<EOF
> #include <foo.h>
> EOF
$ gcc -E foo.c >/dev/null
foo.c:1: foo.h: No such file or directory
$ echo $?
1

Maybe your Autoconf picks some other compiler. I have no idea.

Feel free to send me config.log

Regards,
Pavel Roskin