On Tue, 2018-01-09 at 12:25 +0100, Florian Weimer wrote: > On 01/09/2018 11:00 AM, Petr Pisar wrote: > > On 2018-01-05, Florian Weimer <fwei...@redhat.com> wrote: > > > perl-PDL-2.18.0-4.fc27.src.rpm > > > > [...] > > > This is based on relatively current Fedora rawhide/x86_64 and reflects > > > which currently uses svc_/clnt_/xdr_ symbols from glibc. It does not > > > indicate whether these packages can use libtirpc or have bundled > > > replacements. > > How exactly these xdr_ symbols changed? > > They are compat symbols, so existing binaries can still reference them. > > The perl-PDL situation is yet a bit more complex. Rebuilding it with > the new glibc succeeds because the xdr_ symbols actually come from HDF, > which only provides a static library. Our ld does not default to “-z > defs”, so SD.so contains an undefined symbol references to xdr_enum and > similar symbols. This reference is unversioned, so the glibc dynamic > linker will still bind it to the base symbol version in glibc (e.g., > xdr_enum@GLIBC_2.2.5).
Would it help if xdr_enum@GLIBC_2.2.5 etc. were made weak symbols? That way an unversioned reference could resolve to the one in libtirpc if that happens to be in memory. (I _think_ this is how do_lookup_x already works, but it's hard to ever be sure I'm reading it right.) Definitely mostly in favor of -z defs though. ld doesn't have a way to invert that option atm, but once it does that'd nice to have in a mass rebuild. - ajax _______________________________________________ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
