Brice, Ah, okay. That’s pretty slick. I have some renaming to do :-).
Thank you, Sam On Oct 10, 2019, at 11:14 AM, Brice Goglin <brice.gog...@inria.fr<mailto:brice.gog...@inria.fr>> wrote: Yes. Once you call HWLOC_SET_SYMBOL_PREFIX(), hwloc takes care of automatically renaming your external callers and our internal hwloc code. The idea is that none of us has to change your code when enabling renaming. That's how OMPI manages to support both renamed embedded hwloc and non-renamed external hwloc with the same OMPI code. Brice Le 10/10/2019 à 18:38, Gutierrez, Samuel K. via hwloc-users a écrit : Sorry, I sent the previous email too quickly. Is the recommended usage of embedded hwloc when using HWLOC_SET_SYMBOL_PREFIX to use the non-renamed symbols and avoid altogether the prefixed ones? Thank you, Sam On Oct 10, 2019, at 10:30 AM, Brice Goglin <brice.gog...@inria.fr<mailto:brice.gog...@inria.fr>> wrote: Le 10/10/2019 à 17:38, Gutierrez, Samuel K. via hwloc-users a écrit : Good morning, I have a question about expected name mangling behavior when using HWLOC_SET_SYMBOL_PREFIX in hwloc v2.1.0 (and perhaps other versions). Say, for example, I do the following in a project embedding hwloc: HWLOC_SET_SYMBOL_PREFIX(foo_internal_) HWLOC_SETUP_CORE(…) ... Now, entry points into hwloc are prefixed with foo_internal_ (e.g., foo_internal_hwloc_topology_init()). This all works great. Next, let’s consider what happens to hwloc-exported constants such as HWLOC_OBJ_MACHINE when using the same setup above. I would expect something like this: HWLOC_OBJ_MACHINE now becomes FOO_INTERNAL_HWLOC_OBJ_MACHINE. Instead, I notice the following curious mangling convention: FOO_INTERNAL_hwloc_OBJ_MACHINE. Is this intentional? If so, that’s fine—functionally, it works as expected. However, this seems like a bug. Hello #define HWLOC_NAME(name) HWLOC_MUNGE_NAME(HWLOC_SYM_PREFIX, hwloc_ ## name) #define HWLOC_NAME_CAPS(name) HWLOC_MUNGE_NAME(HWLOC_SYM_PREFIX_CAPS, hwloc_ ## name) Indeed I don't see any reason not to use HWLOC_ on the second line. It looks like we've been doing this forever. Even if users are supposed to only use official (non-renamed) names, I guess there might exist a ugly hack that explicitly depends on renamed names somewhere in hwloc users' code. So I'd rather not touch this as long as it doesn't break anything. Brice _______________________________________________ hwloc-users mailing list hwloc-users@lists.open-mpi.org<mailto:hwloc-users@lists.open-mpi.org> https://lists.open-mpi.org/mailman/listinfo/hwloc-users _______________________________________________ hwloc-users mailing list hwloc-users@lists.open-mpi.org<mailto:hwloc-users@lists.open-mpi.org> https://lists.open-mpi.org/mailman/listinfo/hwloc-users _______________________________________________ hwloc-users mailing list hwloc-users@lists.open-mpi.org<mailto:hwloc-users@lists.open-mpi.org> https://lists.open-mpi.org/mailman/listinfo/hwloc-users
_______________________________________________ hwloc-users mailing list hwloc-users@lists.open-mpi.org https://lists.open-mpi.org/mailman/listinfo/hwloc-users
