Re: patch to add error checking in term

[Thomas Bushnell, BSG]

Roland McGrath <roland@gnu.org> writes:

> > S_tiocltl_tiocflush, set_state:
> > We clear the queue before notifying the bottom handler of the flush (or
> > abandoning all output).  Does it make sense to ask the bottom handler to
> > flush, and if that fails, not to clear our queue?
> 
> Yeah, maybe.  That is by the theory that if you can't do it all, you
> propagate the error up and leave nothing changed.  There is no extra cost
> to doing it that way.

In the specifica case of tiocflush, I disagree.  We can't promise to
flush everything anyway; we do as much as we can.  So if the lower
level can't do anything (and reports an error even) we should not
worry about it, flush our own queue anyway, and report success.

Thomas