Re: glib poll complaint in 1.20 still not resolved

[Owen Taylor <otaylor redhat com>]

Mike Johnson <mjohnson@patriot.or.credence.com> writes:

> Thanks to all who made suggestions about my problem with the poll
> statement in main routine.  For those who are not familiar with what I
> am tracking down, in glib, the configure program hunts for a "poll"
> statement for proper definition in the compile based on the underlying
> OS. For solaris-1, aka sun-os 4.1.4, poll is implemented in the os as a
> macro, not a c library routine.  Currently straight off the ftp site,
> this will compile with minor complaints upto the poll call in the main
> routine.
> 
> I tried replacing all of the following first:
> libtool
> binutils
> autoconf
> automake
> ld
> m4
> 
> Then I tried the re-configure, I tried to undef the poll statements in
> the configure generated header files, then I tried to simply comment
> them out.  But to no avail.  I still get a fatal error on the call to
> poll from gcc.
> 
> If anyone has sucessfully managed to compile glib without a poll,
> (implied by the configure and header files) would you please give me
> some hints about what you did differently.

I suspect you'll have to make some changes to the configure
routines to check for your specific variant of POLL, and
in that case fall back to g_poll() (or some other routine
that wraps your poll() into something with the arguments
that GLib expects.)

Alternatively, as a workaround, after the configure step, 
you could just go into config.h and remove the #define's 
for HAVE_POLL. (Note that there are two of them in there 
now for one reason or the other).

Regards,
                                        Owen