Re: Is statfs() still supposed to be in libc?

[Jason Bacon <outpaddling%yahoo.com@localhost>]

On 03/27/18 10:54, Kamil Rytarowski wrote:
> On 27.03.2018 17:29, Jason Bacon wrote:
> > While upgrading ncbi-blast+, I looked deeper into a workaround for a
> > statfs() issue.  The blast configure script incorrectly determines that
> > statfs() is supported on NetBSD, when in reality is was deprecated and
> > removed from the API a long time ago, superseded by statvfs().
> >
> > However, the statfs() function is still in libc and this is all the
> > configure script checks for.
> >
> > Is statfs() still needed internally, or should it be removed from libc
> > at this point?
> statfs() is supported internally as a compat symbol and shall not be
> used by a newly compiled code.
>
> Every user is mandated to migrate to statvfs().
>
> With a libc major bump, statfs() will likely disappear, although we will
> keep maintaining it in the kernel compat layer.
Thanks.  I'll propose a patch to NCBI to make their configure script a 
little smarter.  Their code already supports statvfs(), but it tries to 
use statfs() on NetBSD simply because it's present in libc.

For now, we hack around it with

.if ${OPSYS} == NetBSD
# statfs is deprecated on NetBSD, but still exists in libc
# configure checks only for undefined reference, not for usability
CONFIGURE_ENV+=         ac_cv_func_statfs=no
.endif