Re: [PATCH] Use statfs64() in shm_open() and posix_getpt().

[Mike Frysinger <vapier at gentoo dot org>]

On Mon 04 Aug 2014 23:50:40 Mike Frysinger wrote:
> On Mon 04 Aug 2014 18:06:58 Andrew Pinski wrote:
> > On Mon, Aug 4, 2014 at 5:49 PM, Mike Frysinger wrote:
> > > On Mon 04 Aug 2014 15:51:15 Joseph S. Myers wrote:
> > >> On Sat, 2 Aug 2014, Mike Frysinger wrote:
> > >> > right, posix_openpt is part of libc, but shm_open is part of librt. 
> > >> > so
> > >> > the
> > >> > former should use __statfs64 to avoid the PLT while the latter should
> > >> > use
> > >> > statfs64.
> > >> 
> > >> No, both should use __statfs64 as statfs64 is not part of the POSIX
> > >> namespace but shm_open is.  As previously discussed, PLT avoidance and
> > >> being namespace-clean are separate issues.
> > > 
> > > why does that matter ?  statfs64 is always exported by libc and thus is
> > > in
> > > the visible ABI namespace.  i don't see why glibc's internals need to be
> > > concerned with conforming to POSIX at all -- they're internals.  what
> > > are
> > > you trying to cater to ?
> > 
> > A program can define a variable called statfs64 which is used instead
> > of the libc.so one when calling from librt.so.
> 
> librt/libpthread/etc... have lots of calls to funcs from libc which can be
> classified in the same way.  they aren't going through GLIBC_PRIVATE nor
> using __ prefixes.  what makes this any different ?

just caught up on another thread from while i was gone.  Carlos wrote up a 
nice explanation here:
https://sourceware.org/glibc/wiki/Style_and_Conventions#Double-underscore_names_for_public_API_functions

so the answer to the other examples i highlighted is that they're wrong too :)
-mike

Attachment: signature.asc
Description: This is a digitally signed message part.