> The behavior is perfectly well defined; "open() opens the named file for
> reading and/or writing, as specified by the flags argument".  In the absence
> of any special-case description, I'd expect it to just proceed to work as
> documented.  (What does it *say* happens if you open a regular file with
> O_NDELAY/O_NONBLOCK _clear_?)

It doesn't say anything about O_NDELAY w.r.t. regular files
in open().  After reluctantly rewiring some synapses (it
is getting harder, you know)  I now agree with you on FIFOs
and O_RDWR.

Reading and writing should also work as expected.

> The SVID is a good deal less legalistic that (for example) the ANSI C standard

Not surprising.  Language semantics are a lot more slippery.

--bakul