[net.bugs.uucp] uucp vs. getty

mp@mit-eddie.UUCP (Mark Plotnick) (07/28/83)

I got uucp to work over our localnet today.  Thankfully, the problem
wasn't too hard because the net looks like a raw-mode tty, but
one problem that had me debugging for hours was that about 2/3 of
the time, uucico would connect to the foreign machine, receive the
login banner, send its username, and then time out waiting for 
"password" (or anything else, for that matter) to come.

The reason was a nice interaction between getty and uucico, accentuated
by vagaries of UNIX and network buffering.  It seems that getty sleeps
for a second after printing out its banner, then flushes all typeahead.
uucico wisely sleeps for 2 seconds before outputting its username.  But,
UNIX and the network work in strange ways, and 2 seconds of delay wasn't
long enough.  The sleep is only accurate to + or - 1 second, I believe,
and who knows what problems network buffering caused.  It was
interesting that the addition of DEBUG statements inside conn.c
sometimes made the delay long enough so that it worked.

Luckily, the L.sys format is flexible enough so that I could
just kludge a line that works right - something like
	login: uucp assword:-uucp-assword: uucp

So, I didn't have to alter the source for uucp or getty more than
necessary.  But, uucp should probably sleep a little longer, and getty
should probably be changed so that it doesn't flush typeahead on network
terminals.


	Mark

honey@allegra.UUCP (07/28/83)

	... "" @ in:--in: \d@uucp word word
is the best "standard" chat script i've come across (providing
\d == delay in sendthem().)  @ wins over \r for breaking down
getty; as ber explains, "some bits are better than others."
	peter honeyman