On 28 Oct 2013 19:41, "Christos Zoulas" <christos%astron.com@localhost> wrote:
> In article <CAK4o1WzX3aG=pyN43ZJ-j0dqS5EM4JKLuu30NtGJSE10dtz3HQ%mail.gmail.com@localhost>,
> Justin Cormack <justin%specialbusservice.com@localhost> wrote:
> Heh, glibc talking about standards. man strerror_r. Please explain
> how one can compile a program that uses additional features not in
> the POSIX standard (meaning it requires _GNU_SOURCE) -- common
> features in all Unixes, but at the same time does not want to
> violate the standard so it can compile on other systems avoiding
> ifdefs/autoconf tests (wants strerror_r to return int).
> The glibc standard is whatever is convenient for glibc. I am not saying
> this is not true for other software, but...
> The problem here is that we are trying to mix two sets of headers,
> one that assumes that __unused is an attribute that is optional,
> and the other that assumes that it is an identifier and is mandatory.
> The easiest fix is to follow what was suggested in the thread.
> Something that glib is already doing:
> Using inconvenient long unique names like __malloc_attribute__.
> And apply it in the __unused case, but that would make glibc more
> consistent, and improve coexistance with other operating systems
> outside the glibc ecosystem so it is not going to be done.
The fix to remove use of __unused has now been committed to glibc and should be in the next release.
We should look at doing the same for tools...