What I noticed is that moving the sizeof check before the library check resulted in the generated GCC command not including the libraries (no -l parameters). Therefore, the sizeof check program did indeed run, and yielded the correct result. Evidently, once the library check is performed, subsequent checks all generate a GCC command line which includes the checked libraries as "-l<libxx>" parameters. You simply need to organize file configure.in so that all tests which do not require extra libraries are executed before any library-check tests. (I assume that the macro language generates a "LIB=" line when a library check is requested, and that this line is not unset after the test -- a probable bug in autoconf or configure).
BTW, I noticed that the library check only compiled test programs -- it did not attempt to run them! (thats why the library check succeeded even though the checked libraries are not in the search path). Strange.
Jack
-----Original Message-----
From: Roland Dreier [mailto:[EMAIL PROTECTED]]
Sent: Thursday, September 15, 2005 8:32 AM
To: Jack Morgenstein
Cc: [email protected]
Subject: Re: [PATCH] [openib-general] Strange configure error in libibcm
Jack> No error message is generated (i.e., the patch fixes the
Jack> problem). autoconf and configure work as they should with
Jack> this patch.
Right, but the program that gets built won't run because ld.so won't
find the library it's linked with. That's what I meant about the
error message being easier to understand: it's easy to see you have to
set LD_LIBRARY_PATH in that case.
- R.
_______________________________________________ openib-general mailing list [email protected] http://openib.org/mailman/listinfo/openib-general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
