> I saw the recent change to make release not interruptable to fix the
> lost release problem when a writing process is killed. Can there be
> the same change for flush? I rely on flush, rather than release, to
> handle file closing since I can pass an error back via flush.
I think the proper way to do this, is to use flush when possible, so
that the error is not lost, but in addition always write back data on
the last release. I think the flush() operation is not even called if
a process is killed, and returning an error is not meaningful in this