>> >Description: fsync() on a pipe fails, returning EINVAL.  This
>> breaks RCS, which expects to fsync all of its I/O before closing
>> it.

Chris> umm, as far as i can tell, every UN*X that i have access to
Chris> (net/2 derivatives, 4.4BSD, Ultrix, SunOS, Domain/OS) does
Chris> exactly this.  it's the expected behaviour, not a bug.  RCS is
Chris> in error here, and the vfs routines should not be "fixed."

The funny thing is that the reason for the bad interaction between RCS
and VC is the fsync() that "we" added long long ago to make sure that
the RCS files were written out to disk during CVS commits.

	--jtc

------------------------------------------------------------------------------