Re: Re: Cygwin PostgreSQL CVS Patch Question

Peter,

On Tue, Jan 09, 2001 at 05:55:39PM -0500, Jason Tishler wrote:
> On Tue, Jan 09, 2001 at 04:34:57PM -0500, Tom Lane wrote:
> > Jason Tishler <Jason.Tishler@dothill.com> writes:
> > > Why did you remove the following from src/include/port/win.h?
> > >     #if (CYGWIN_VERSION_API_MAJOR >= 0) && (CYGWIN_VERSION_API_MINOR >= 8)
> > >     #define sys_nerr _sys_nerr
> > >     #endif
> > > This definitely breaks Cygwin.
> >
> > Er, why should it?  With the code as it stands, configure should decide
> > that Cygwin doesn't HAVE_SYS_NERR, and so nothing will reference
> > sys_nerr.
>
> Unfortunately, the above is not true:
>
>     $ configure
>     ...
>     checking for sys_nerr... yes
>     ...

I have more details as to why the above is occurring.

Consider the following program, j.c, which is derived from the sys_nerr
test program in configure:

    extern int sys_nerr;
    int main() {
    int x = sys_nerr;
    ; return 0; }

Now compile it with the same (relevant) options used by configure:

    $ gcc -O2 -o j j.c

Now compile it without the -O2 option:

    $ gcc -o j j.c
    /tmp/ccf49pHw.o(.text+0xc):j.c: undefined reference to `sys_nerr'
    collect2: ld returned 1 exit status

The problem is gcc -O2 is being too smart, optimizing away the reference
to sys_nerr.

So when configure compiles it's sys_nerr test program it compiles without
errors and returns 0 which fools configure into thinking that Cygwin has
sys_nerr.

Any ideas on what is the best way to solve this problem?

Thanks,
Jason

--
Jason Tishler
Director, Software Engineering       Phone: +1 (732) 264-8770 x235
Dot Hill Systems Corp.               Fax:   +1 (732) 264-8798
82 Bethany Road, Suite 7             Email: Jason.Tishler@dothill.com
Hazlet, NJ 07730 USA                 WWW:   http://www.dothill.com