Quoth David Bremner on May 07 at 10:17 pm: > Tomi Ollila <tomi.ollila at iki.fi> writes: > > > On Wed, May 07 2014, Kushal Kumaran <kushal.kumaran+notmuch at gmail.com> > > wrote: > > > >> X?c? <xico at atelo.org> writes: > >> > >>> Also, commit 3c13bc introduced a call to fdatasync() which is not > >>> available on FreeBSD, and probably not either on MacOS at least. > >>> > >> > >> fdatasync is POSIX: > >> http://pubs.opengroup.org/onlinepubs/009695399/functions/fdatasync.html > > > > No wonder it is problematic, then >;) > > > > I seem to recall Austin saying on IRC that this usage was guaranteed to > call fsync anyway. Comments Austin?
Yes, since the size of the file will have definitely changed, the metadata will have to be flushed anyway, so using fdatasync here has no advantage over using fsync.
