> Well, if change b) has the same effect as change a), that would be
> more confirmation that my understanding of the "problem" is right.
> Change b) affects only the NSL_FORK code, so it retains the benefits
> of the "Avoid checking interrupts" code (whatever they are - I am
> not really convinced that checking for keyboard input is such a "slow
> and expensive operation" that it nees to be rationed.
Well, it certainly was when I added the check - the select's were using
a large percentage of the cpu time (LP may recall some numbers for that -
I recall it as a factor-of-two slowdown in various operations - but the
calls have been revised a few times since then).
Perhaps there's a loop in HTTCP.c which is calling HTCheckForInterrupt
without waiting or otherwise delaying. That would cause this effect, too,
I think.
> With change b), expect lynx to react less immediately to 'z' when
> it's waiting for a nameserver response.
>
> You'd have to only recompile the two files in question, not everything.
>
> Klaus
>
>
--
Thomas E. Dickey
[EMAIL PROTECTED]
http://www.clark.net/pub/dickey
