Re: [PATCH v3] nptl: Export libthread_db-used symbols under GLIBC_PRIVATE

[Szabolcs Nagy <szabolcs.nagy@arm.com>]

The 06/29/2021 11:09, Florian Weimer via Libc-alpha wrote:
> * Szabolcs Nagy:
> 
> > The 06/28/2021 08:59, Carlos O'Donell via Libc-alpha wrote:
> >> On 6/28/21 8:41 AM, Florian Weimer wrote:
> >> > This allows distributions to strip debugging information from
> >> > libc.so.6 without impacting the debugging experience.
> >> 
> >> This looks good, no redundant checks. The only *MAIN* symbol
> >> left in structs.def is from ld.so and we skip checking it
> >> (because the test framework only processes libc.so's symbols).
> >> 
> >> Reviewed-by: Carlos O'Donell <carlos@redhat.com>
> >
> > after the previous two nptl_db commits i see
> >
> > FAIL: nptl_db/db-symbols
> >
> > i thought this third patch would fix it, but it didn't:
> >
> > $ head nptl_db/db-symbols.out
> > _thread_db_pthread_eventbuf_eventmask_event_bits@@GLIBC_PRIVATE ***MISSING***
> > _thread_db_pthread_start_routine@@GLIBC_PRIVATE ***MISSING***
> > _thread_db_sizeof_list_t@@GLIBC_PRIVATE ***MISSING***
> > _thread_db_pthread_schedparam_sched_priority@@GLIBC_PRIVATE ***MISSING***
> > _thread_db_td_eventbuf_t_eventdata@@GLIBC_PRIVATE ***MISSING***
> > _thread_db_list_t_prev@@GLIBC_PRIVATE ***MISSING***
> > _thread_db_sizeof_dtv_slotinfo@@GLIBC_PRIVATE ***MISSING***
> > _thread_db_pthread_cancelhandling@@GLIBC_PRIVATE ***MISSING***
> > _thread_db___pthread_keys@@GLIBC_PRIVATE ***MISSING***
> > _thread_db_rtld_global__dl_tls_dtv_slotinfo_list@@GLIBC_PRIVATE ***MISSING***
> >
> > is this expected?
> 
> It is not expected.  The test runs even when cross-compiling, and I
> don't see this failure in a default build-many-glibcs.py run.
> 
> Are these symbols present as dynamic symbols in your build of libc.so.6?
> 
> I suspect your readelf has different output not expected by the script.
> Apparently older versions do not print symbol versioning information
> with -D -s. 8-( I guess we should parse ELF directly, rather than
> readelf output, because it is easier to maintain.
> 
> As a stop-gap measure, we should probably switch to objdump -T.

yes i can confirm that my readelf does not print symbol versions
with -D -s.

GNU readelf (GNU Binutils for Ubuntu) 2.31.1

i can try to update it if it's considered too old.