On 12/07/2014 07:29 PM, Paul Eggert wrote: > Pádraig Brady wrote: >> BTW if free() may reset errno on some platforms then it's >> probably worth augmenting the gnulib free() wrapper >> to restore errno if needed, > > The documented GNU behavior for 'free' allows 'free' to set errno, > right? So why should the corresponding gnulib wrapper guarantee > behavior above and beyond what GNU implements?
Only implicitly (basically, any function that does not explicitly state that errno is left unchanged is free to change errno as a side effect on success). But the POSIX proposal is that free should be one of the functions that guarantees that errno can't be clobbered on success; and GNU should probably implement that now rather than waiting a few years for the POSIX proposal to become standardized. > > It might make sense to have a variant of 'free' that preserves errno, > but we should probably give the variant a different name. 'noerr_free', > say. I'm not sure a variant is needed; if we can get glibc to guarantee the POSIX proposal now, then we would merely be guaranteeing sane behavior on all platforms. -- Eric Blake eblake redhat com +1-919-301-3266 Libvirt virtualization library http://libvirt.org
signature.asc
Description: OpenPGP digital signature
