Re: conftest files

[Christopher Hulbert]

On 8/25/06, Ralf Wildenhues <address@hidden> wrote:
> This thread seems to have died down without conclusion.
> Christopher had a problem with configure link test file leftovers
> spoiling following tests, with some PGI compiler:
>
> * Christopher Hulbert wrote on Tue, Aug 15, 2006 at 11:19:38PM CEST:
> > On 8/13/06, Stepan Kasal <address@hidden> wrote:
> > >> >On Fri, Aug 11, 2006 at 08:52:00AM -0400, Christopher Hulbert wrote:
> > >> >> information, so the PGI compiler creates some updated IPA information
> > >> >> for the conftest. When testing for the next library compiled without
> > >> >> IPA information, it doesn't overwrite the original IPA file and gets
> > >> >> an undefined symbol resulting from the first library.
> > >
> > >at the end of AC_LINK_IFELSE (formerly AC_TRY_LINK), the generated
> > >files are deleted.  What is the name of the filei(s) with IPA
> > >information?  Perhaps it should be deleted, too.
>
> > I guess the problem from an autoconf standpoint is it's compiler
> > dependent and even file dependent. In this case the filename is
> > conftest_ipa8_conftest.oo. The lastpart is the name of the output
> > executable (conftest). The 8 is the length of the executable filename.
> > The extension .oo is the optimized object file so that linking *.o
> > wouldn't pull in the ipo optimized objects.
>
> and PGI compilers 5.2, 6.0, 6.1, on GNU/Linux, and
>   ./configure CC=pgcc CFLAGS=-Mipa
>
> Also, the manpages state that IPA files are named:
> |      file.ipa    InterProcedural Analyzer (IPA) file
> |      file.ipo    InterProcedural Analyzer (IPA) file
>
> but I find neither of them lying around

Unfortunately, I haven't had the time to try the patch yet (maybe this
weekend). Indeed the manpage does not list the .oo files (maybe I'll
file a bug report aginst PGI). In my case I had  a VSIPL library
compiled with IPA. When checking for it, this is when I got the
problem.

Here's a simple example that should show you the .oo files (PGI 6.1-6):

address@hidden ~]$ cat test2.c
#include <stdlib.h>

double multiply(double x,double y)
{
   return x*y;
}
address@hidden ~]$ cat test.c
#include <stdlib.h>
#include <stdio.h>

int
main(void)
{
   double z = multiply(2.0,3.0);
   printf("%f\n",z);
   return EXIT_SUCCESS;
}
address@hidden ~]$ pgcc -Mipa -Minform=inform -Minfo=all -c test2.c
PGC-I-0222-Redundant definition for symbol __THROW
(/usr/include/sys/cdefs.h: 62)
address@hidden ~]$ pgcc -Mipa -Minform=inform -Minfo=all test.c test2.o
test.c:
PGC-I-0222-Redundant definition for symbol __THROW
(/usr/include/sys/cdefs.h: 62)
IPA: Recompiling test.o: new IPA information
IPA: Recompiling test2.o: new IPA information
address@hidden ~]$ ls *oo
test2_ipa5_a.out.oo  test_ipa5_a.out.oo



> * Paul Eggert wrote on Wed, Aug 16, 2006 at 07:27:22PM CEST:
> > Stepan Kasal <address@hidden> writes:
> > > 2006-08-16  Stepan Kasal  <address@hidden>
> > >
> > >     * lib/autoconf/general.m4 (_AC_LINK_IFELSE): Remove the IPA/IPO
> > >     file created by the PGI compiler.
> >
> > That'd be fine with me, if it does fix the problem.  Thanks.