Re: [PATCH v3 08/37] nptl: Move pthread_once and __pthread_once into libc

[Florian Weimer <fweimer@redhat.com>]

* Adhemerval Zanella:

> On 17/03/2021 10:30, Adhemerval Zanella wrote:
>> 
>> 
>> On 16/03/2021 14:28, Florian Weimer via Libc-alpha wrote:
>>> And also the fork generation counter, __fork_generation.  This
>>> eliminates the need for __fork_generation_pointer.
>>>
>>> call_once remains in libpthread and calls the exported __pthread_once
>>> symbol.
>>>
>>> pthread_once and __pthread_once have been moved using
>>> scripts/move-symbol-to-libc.py.
>> 
>> LGTM, I just don't see why we need a GLIBC_2.34 __pthread_once.
>
> Ok, it is called by call_once.  I wonder if we could make a GLIBC_PRIVATE
> instead so we can remove it once we moce call_once to libc.

nss_ldap-253-16.el4 and nss_ldap-253-52.el5_11.2 use it, but those are
doubly obslete (old releases, nss_ldapd is a secure replacement of
nss_ldap).

We could turn __pthread_once into a compat symbol, but maybe we should
handle this in consistent fashion for all libpthread __ symbols that do
not have apparent external users?

From a tactical perspective, maybe we should postpone this conversion to
some pointer later in this release cycle.  We would have to change the
symbol name in glibc because it wouldn't make sense to have a non-compat
GLIBC_PRIVATE symbol and a compat symbol at the same version (users
would just link to the GLIBC_PRIVATE symbol).  If everything is in libc,
we can simply use libc_hidden_proto, without renaming internal use.

Thanks,
Florian