[mvapich-discuss] mvapich2-1.7a: configure: error: Unable to convert MPI_SIZEOF_AINT to a hex string

[Anthony Chan]

I thought the misleading MPI_SIZEOF_AINT error message was
due to the generic MPICH2 configure macros.  After inspecting
your config.log and mvapich2-1.7a's configure.in, it seems the bug
is related to the checks that test for LiMIC2 header and
library. These tests uses AC_CHECK_HEADER and AC_CHECK_LIB
which only do compile and link tests, no runtime test, so
it can't detect the incorrect LD_LIBRARY_PATH.  It seems the
simplest solution is to add a runtime test in/after AC_CHECK_LIB
to check if executable linked with limic2 lib actually works.
(Make sure to disable the test when cross-compiling).

BTW, I notice mvapich2-1.7a's configure is generated with
autoconf-2.66 which isn't recommended by MPICH2 because
2.66 has bug in AC_CHECK_SIZEOF and AC_CONFIG_SUBDIRS.
(MPICH2's maint/updatefiles states that and explicitly checks
for autoconf version.)  Don't remember the exact nature of the bugs
in these macros, it could be related to cross-compilation support...

A.Chan


----- Original Message -----
> 2011/6/1 Anthony Chan <chan at mcs.anl.gov>
> 
> >
> >
> > MPICH2's configure uses autoconf's AC_CHECK_SIZEOF to determine the
> > sizeof
> > C's basic types. Whenever an invalid compile command is used,
> > AC_CHECK_SIZEOF
> > sets the corresponding ac_cv_sizeof_* to zero and that is how the
> > macro is
> > defined.
> > What it should have happen when LD_LIBRARY_PATH is incorrect,
> > AC_PROG_CC
> > should
> > have aborted and emited error message before configure reaches
> > AC_CHECK_SIZEOF.
> > Could you send us the config.log so we can take a look to see
> > anything in
> > PAC_PROG_CC should be improved ?
> >
> > I have attached my (compressed) config.log file. I have added the
> --with-limic2=/home/maik/90183/limic2 option to configure, however,
> /home/maik/90183/limic2/lib is not in my LD_LIBRARY_PATH.
> 
> 
> > BTW, what is corresponding mpich2 version for mvapich2-1.7a ?
> >
> 
> According to the CHANGELOG, it's based on MPICH2-1.3.2p1.
> 
> Maik
> 
> 
> > > After tracing the issue, I found that MPI_SIZEOF_AINT is 0, and so
> > > is
> > > ac_cv_sizeof_int, ac_cv_sizeof_double, etc.
> > >
> > > The error occurs when 'configure' determines the ac_cv_size_*
> > > sizes.
> > > Since
> > > my LD_LIBRARY_PATH did not contain the proper paths, a generated
> > > executable
> > > does not properly run, and 'configure' sets the size to 0 instead
> > > of
> > > printing an error and aborting.
> > >
> > > Even though config.log contains multiple linker and runtime errors
> > > about
> > > missing libraries, the configure script seems to ignore those...
> > >
> > > So, if you get MPI_SIZEOF_AINT errors, check your config.log for
> > > missing
> > > library errors and check your LD_LIBRARY_PATH!
> > >
> > > Maik
> > >
> > > _______________________________________________
> > > mvapich-discuss mailing list
> > > mvapich-discuss at cse.ohio-state.edu
> > > http://mail.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss
> >
> 
> 
> 
> --
> [image: cv_sig.gif]
> Dr. Maik Nijhuis
> HPC Benchmark Specialist
> 
> 
> Direct: +31 20 407 7556
> Skype: maiknijhuis
> maik.nijhuis at clustervision.com
> <marco.vos at clustervision.com>
> 
> ClusterVision BV
> Nieuw-Zeelandweg 15B
> 1045 AL Amsterdam
> The Netherlands
> Tel: +31 20 407 7550
> Fax: +31 84 759 8389
> www.clustervision.com
> 
> _______________________________________________
> mvapich-discuss mailing list
> mvapich-discuss at cse.ohio-state.edu
> http://mail.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss