On Wed, Aug 12, 2015 at 12:40 PM, Jeff Squyres (jsquyres) < [email protected]> wrote: > > In this case, however, we're unlikely to fix this issue. The "export > CFLAGS" is there for a reason, and this is a low priority to fix. I.e., > there's no technical *problem* -- it's just that we're now passing -g down > to VampirTrace, and the resulting libraries are a little larger. VT is > going away in the Very Near Future, so this issue will solve itself then. >
Jeff, You seem to be contradicting yourself above. You state that > The "export CFLAGS" is there for a reason and yet the problem Lisandro is reporting began when you *removed* that export. And you also say > we're now passing -g down to VampirTrace when the problem is not because of what you *are* passing, but rather what you have *ceased* to pass (CFLAGS="-g -O2" in the report). Can you please clarify? Also, lets assume that CFLAGS contains something like "-m32". I believe that the current (no-export) case may be broken if Open MPI is built 32-bit and VT is built 64-bit (though I've not verified). -Paul -- Paul H. Hargrove [email protected] Computer Languages & Systems Software (CLaSS) Group Computer Science Department Tel: +1-510-495-2352 Lawrence Berkeley National Laboratory Fax: +1-510-486-6900
