[mpich2-dev] bugs in test/mpi/configure.in

David Gingold david.gingold at sicortex.com
Tue Feb 17 09:38:08 CST 2009


MPICH2 folks --

I spent some time last week puzzling over why, when configuring the  
MPICH2 tests, the configure script was doing this:

	checking that we can build MPI programs with Fortran 90... no

when in fact mpif90 was working fine.  The cause appears to be a bug  
in test/mpi/configure.in.

The patch below (to the 1.1a1 code) has three changes.  I think it's  
the middle change that fixed my problem, but the other two seemed, on  
their face, to be bugs as well.

Please review this with skepticism.  I am no autoconf wizzard!

Note that I'm attempting to get the tests to run standalone against  
the installed MPI library on our system.  I am configuring them, for  
example, this way:

$ CC=gcc CXX=g++ F77=pathf95 F90=pathf95 ./configure --disable-spawn  
--disable-romio --enable-f77 --enable-f90 --enable-cxx --enable- 
threads=serialized --enable-checkerrors --with-mpi=/usr

-dg

....

Index: configure.in
===================================================================
--- configure.in	(revision 70089)
+++ configure.in	(working copy)
@@ -182,6 +182,16 @@
          enable_f77=no
      fi
  fi
+if test "$enable_f90" = yes ; then
+    AC_PROG_F90
+    if test -z "$BASEF90" ; then
+        BASEF90="$F90"
+    fi
+    # If there is no working F90, then set enable_f90 to no
+    if test -z "$F90" ; then
+        enable_f90=no
+    fi
+fi
  if test "$enable_cxx" = yes ; then
      AC_PROG_CXX
      if test -z "$BASECXX" ; then
@@ -227,6 +237,11 @@
      else
          F77=$MPIF77
      fi
+    if test -z "$MPIF90" ; then
+        F90=$with_mpi/bin/mpif90
+    else
+        F90=$MPIF90
+    fi
      if test -z "$MPICXX" ; then
          CXX=$with_mpi/bin/mpicxx
      else
@@ -659,7 +674,9 @@
      elif test -z "$BASEF90" ; then
          BASEF90=$F90
      fi
-    F90=$saveF90
+    if test -n "$BASEF90" ; then
+        F90=$saveF90
+    fi
  fi

  if test "$enable_f90" = yes ; then



More information about the mpich2-dev mailing list