More bugs.

More bugs.

[Boyd]

    From:	avg@postman.ncube.com (Vadim Antonov)
    
    Plan9 already has "line discipline" modules is several places
    (devcons, 81/2, readln in auth utilities, telnetd), so instead
    of replicating it would be useful to keep it in one place.

i don't have a problem with lines discipline as a technology, it's
just their misuse that concerns me.  i've seen that (in the old system)
readln was duplicated in too many places.  that should be addresed.
adding further /dev/cons modes is a road to ruin:

	they're unnecessary in any modern system -- rob
    
    Raw mode and no echo mode are two different things and lumping
    them together only makes all programs which read passwords
    to be equipped with handling of kill and erase.

on plan9 their are bugger all of these things that want to fool
with the cons mode; the interface is the window system, not some
archaic tty interface.  occassionally you are faced with
this stuff: ftpfs, passwd.  but, for the most part you're not.
    
    --vadim
    (kremvax? yes, i used to run _the_ kremvax).
    
yes, i know who you are vadim.  you look just like brucee :-)

More bugs.

[Boyd]

    From:	avg@postman.ncube.com (Vadim Antonov)
    
    The quick fix is appended.  As always, the real fix
    would be to make echo-on and echo-off _separate_
    from raw-on and raw-off.  I remember seeing the same
    kind of problem in Unix version 6.  I guess aliens from
    outer space kept somebody in suspended animation all
    those years :)
    
ohh no, rob has got it right with raw mode.  any other tty mode is
a waste of time; just extra complexity for nothing.  this kind of
thinking will result in a bsd-tty line discipline for plan9, job
control, csh, dbx -- the horror, the horror...

More bugs.

[Vadim]

>i don't have a problem with lines discipline as a technology, it's
>just their misuse that concerns me.  i've seen that (in the old system)
>readln was duplicated in too many places.  that should be addresed.
>adding further /dev/cons modes is a road to ruin:

It still is.  And with different bugs in different places.
Verry funnny.

>	they're unnecessary in any modern system -- rob

Yep.  As long as you don't get to move to a real world where
most of terminals are messy-loss machines, all of them
have vt100 emulator.
    
>on plan9 their are bugger all of these things that want to fool
>with the cons mode; the interface is the window system, not some
>archaic tty interface.  occassionally you are faced with
>this stuff: ftpfs, passwd.  but, for the most part you're not.

Plan9 is a text-based system, just like good 'ol unix.
A window system is merely something to enter text. There is
absolutely *no* reason for making it hard to work from
other systems.  Particularly, my additions to telnetd made
telnet session to feel like 81/2 terminal (though you can't
run sam or do Unicode).
    
--vadim

More bugs.

[Vadim]

>ohh no, rob has got it right with raw mode.  any other tty mode is
>a waste of time; just extra complexity for nothing.  this kind of
>thinking will result in a bsd-tty line discipline for plan9, job
>control, csh, dbx -- the horror, the horror...

Plan9 already has "line discipline" modules is several places
(devcons, 81/2, readln in auth utilities, telnetd), so instead
of replicating it would be useful to keep it in one place.

Raw mode and no echo mode are two different things and lumping
them together only makes all programs which read passwords
to be equipped with handling of kill and erase.

Most of the complexity of BSD and SysV line disciplines comes
from the history of the issue, as in v6 and v7 the hardware-specific
things were lumped together with the line discipline stuff.
Namely, ~ECHO had dual functon -- to handle half-duplex terminals
*and* to allow reading passwords.  Needless to say it was
inconvinient for both :)  RAW was "both ways", meaning no
conversion on output to terminals and no cooked processing
on input. That was very funny on terminals which needed 
output processing for things like Shift-In and Shift-Out
(yes, Virgina, there are people who speak funny languages
with letters like ya and tvyordyi znak).

For the hack value -- "no echo" on half-duplex hardcopy terminals
means typing over many times.  To my knowledge unix never
did that right.

If you don't like line disciplines you'd better not do
telnet-ing around.  There are different things (even TTYs)
on the Internet.  I respect the "simplicity is beauty" maxim
but tins on a string are arguably simplier than telephone.

Much of ugliness of present-day unices comes from the
under-functionality of the original system.  The holes were
plugged by other people, differently, and so the Babylon
started.  Later, some sorely misguided souls tried to "unify"
things by getting all variances into one system instead
of redesigning things to provide clear separation of functions.

Providing no functionality is not an answer, as people who
need to solve real problems in the real life will find ways
around it -- but it won't be pretty.  Not at all.

--vadim
(kremvax? yes, i used to run _the_ kremvax).

More bugs.

[Vadim]

/sys/src/cmd/auth/lib/readln didn't turn raw mode off
before sending newline (which caused some funny behaviour
if the server translates the newline differently in
raw and cooked modes).  Also, in some cases closed
stream could be interpreted as DEL by readln.

The quick fix is appended.  As always, the real fix
would be to make echo-on and echo-off _separate_
from raw-on and raw-off.  I remember seeing the same
kind of problem in Unix version 6.  I guess aliens from
outer space kept somebody in suspended animation all
those years :)


--vadim

naiad% pwd
/sys/src/cmd/auth/lib
naiad% diff readln.c.old readln.c
49,50d48
<               if(*p == 0x7f)
<                       exits(0);
53c51,52
<                       if(raw)
---
>                       if(raw) {
>                               write(ctl, "rawoff", 6);
54a54
>                       }
58a59,60
>               if(*p == 0x7f)
>                       exits(0);
naiad%