* George Bosilca wrote on Thu, Sep 01, 2005 at 06:27:27AM CEST: > I trace this one as far as I could. And the results are mostly unexpected. > On some of the clusters it compiles without any problems and on some > others it doesn't. The difference is ... if there is an ltdl.h installed > in the system directories. I don't think that's the expected behavior for > the compilation, if we have our own ltdl.h file why should we use the > system wide one ...
This is usually up to the user's choice at some point (I don't know about OpenMPI with that respect). > For some component it work as expected because on the revision 7106 the > -I$(top_srcdir)/opal/libltdl has been added to the AM_CPPFLAGS in the > Makefile.am. As most of our code use components there is a dependency > between nearly every file in the ompi source code and the ldtl.h file. > Having to modify all the Makefile.am is a herculean task ... Ah, ok, I was blind then, in thinking only the parts that compiled over here used ltdl.h. There are a couple of ways to add include paths to many Makefile.am's: You could leverage config/Makefile.options, and set a default value for AM_CPPFLAGS (surely you have to change the Makefile.am's once to only add to AM_CPPFLAGS afterwards; or, maybe better, to add some other variable CPPFLAGS_LTDL set in Makefile.options). But surely, once the work has to be done. I suggested that change because before, the code would've broken with Libtool-2.x anyway. Cheers, Ralf
