[Twisted-Python] Screen clearing in the Manhole

[Peter Westlake]

I'm making a manhole service in the usual way:

class PF(ServerProtocol):        def protocolFactory(*a, **kw):
return manhole.Manhole(namespace)

realm = manhole_ssh.TerminalRealm()    realm.chainedProtocolFactory = PF
mh_portal = portal.Portal(realm)    mh_portal.registerChecker(checker)

manhole_service = TCPServer(0, manhole_ssh.ConchFactory(mh_portal))
manhole_service.startService()

and I'd like to print a message when the terminal is opened. This
change does it:

class PF(ServerProtocol):        def protocolFactory(*a, **kw):
class Manhole(manhole.Manhole):                def connectionMade(self):
super(Manhole, self).connectionMade()
self.terminal.write('Hello')            return
manhole.Manhole(namespace)

and so does this version, which is better because
t.c.i.i.ServerProtocol.connectionMade says it is the place to
write messages:

class PF(ServerProtocol):        def protocolFactory(*a, **kw):
return manhole.Manhole(namespace)

def connectionMade(self):            ServerProtocol.connectionMade(self)
self.write('Hello')

But in both cases, after the message the screen is cleared by what looks
like a form feed, and another prompt is printed. So far I can't find
where in the code this happens. Is there a way to print a message and
not have it scrolled off the screen?

Thanks,

Peter.

[Brian Costlow]

As to where it happens, look at
twisted.conch.recvline.RecvLine.connectionMade() and
twisted.conch.insults.insults.TerminalProtocol.makeConnection()

You could step on the RecvLine. initializeScreen(), but I have no idea what
the side effects are. (And of course if you bypass all or part of it, even
if it works now, it's alway subject to future breakage.

Post by Peter Westlake
return manhole.Manhole(namespace)
realm = manhole_ssh.TerminalRealm()
realm.chainedProtocolFactory = PF
mh_portal = portal.Portal(realm)
mh_portal.registerChecker(checker)
manhole_service = TCPServer(0, manhole_ssh.ConchFactory(mh_portal))
manhole_service.startService()
and I'd like to print a message when the terminal is opened. This change
super(Manhole, self).connectionMade()
self.terminal.write('Hello')
return manhole.Manhole(namespace)
and so does this version, which is better because
t.c.i.i.ServerProtocol.connectionMade says it is the place to write
return manhole.Manhole(namespace)
ServerProtocol.connectionMade(self)
self.write('Hello')
But in both cases, after the message the screen is cleared by what looks
like a form feed, and another prompt is printed. So far I can't find where
in the code this happens. Is there a way to print a message and not have it
scrolled off the screen?
Thanks,
Peter.
_______________________________________________
Twisted-Python mailing list
http://twistedmatrix.com/cgi-bin/mailman/listinfo/twisted-python

[Peter Westlake]

Post by Brian Costlow
As to where it happens, look at
twisted.conch.recvline.RecvLine.connectionMade() and
twisted.conch.insults.insults.TerminalProtocol.makeConnection()

My subclass of Manhole first calls Manhole.connectionMade, which calls
HistoricRecvLine.connectionMade, which calls RecvLine.connectionMade,
which resets the screen exactly as you say. Then, after the screen has
been reset, I print the message. But then the screen is cleared a
second time!

Peter.

[John Santos]

Someone wrote: (Sorry, reply in my mail client doesn't give me an
editable text version of the mail. Instead, it is including it as an
attachment. I don't know why this happens. I do know from bitter
experience that if I don't delete the attachments from my reply, users
of brain-dead mail clients (I'm looking at you, Outlook) can't see
me reply but just see the attached copy of the original message. So
I cut and pasted some context so people know what I'm talking about.)

Post by Brian Costlow
As to where it happens, look at
twisted.conch.recvline.RecvLine.connectionMade() and
twisted.conch.insults.insults.TerminalProtocol.makeConnection()
You could step on the RecvLine. initializeScreen(), but I have no idea what
the side effects are. (And of course if you bypass all or part of it, even
if it works now, it's alway subject to future breakage.
On Tue, Aug 25, 2015 at 12:27 PM, Peter Westlake

[...]

t.c.r.R.connectionMade() calls self.initializeScreen(), which in turn
calls self.terminal.reset()

I'm pretty sure s.t.reset() is actually twisted.conch.insults.reset(),
which has a bug (or an undesirable feature), it sends a "hard reset"
escape sequence to the terminal (<ESC>c). This freaks out many terminal
emulators, and real, VT100-compatible terminals, in many different ways.

I've tested in Kermit-95, PuTTy, xterm, DECterm (a VT emulator written for
DECwindows, DEC's X implementation) and on real DEC VT420 and VT520
terminals. They act differently, but none are pleasant. For example, a
VT520 does a power-cycle on the CRT (which takes about 10 seconds),
reloads the saved configuration, and (if you have multiple sessions
enabled) switches to Session 1.

The VT420 and VT520 manuals say about "hard reset", basically, "Don't
do that!!"

I have a patch to change it to send a soft reset (<ESC>[!p) and to set the
terminal into "replace mode" (<ESC>[4l) (sometimes the software thought it
was in overstrike mode, but the terminal thought it was in insert mode,
with bad results. Ideally, it would set both the software and hardware to
what ever mode the user desired. The hard reset would set the terminal to
whichever mode was stored in its permanent settings. For a software
terminal emulator, it was a dice roll.)

Depending on the terminal emulator or physical terminal you are using,
this probably explains your symptoms.

See ticket # 7514: https://twistedmatrix.com/trac/ticket/7514

[J Todd Ulrich]

Can you remove me please. I am in NO WAY affiliated with your organization. Virus?

Walk Tall. 

J Todd Ulrich
m. 416.625.3209

Please excuse brevity as this message sent from my BlackBerry 10 smartphone on the Rogers network.
  Original Message  
From: John Santos
Sent: Tuesday, August 25, 2015 5:58 PM
To: Twisted general discussion
Reply To: Twisted general discussion
Subject: Re: [Twisted-Python] Screen clearing in the Manhole



Someone wrote: (Sorry, reply in my mail client doesn't give me an
editable text version of the mail. Instead, it is including it as an
attachment. I don't know why this happens. I do know from bitter
experience that if I don't delete the attachments from my reply, users
of brain-dead mail clients (I'm looking at you, Outlook) can't see
me reply but just see the attached copy of the original message. So
I cut and pasted some context so people know what I'm talking about.)

Post by Brian Costlow
As to where it happens, look at
twisted.conch.recvline.RecvLine.connectionMade() and
twisted.conch.insults.insults.TerminalProtocol.makeConnection()
You could step on the RecvLine. initializeScreen(), but I have no idea what
the side effects are. (And of course if you bypass all or part of it, even
if it works now, it's alway subject to future breakage.
On Tue, Aug 25, 2015 at 12:27 PM, Peter Westlake

[...]

t.c.r.R.connectionMade() calls self.initializeScreen(), which in turn
calls self.terminal.reset()

I'm pretty sure s.t.reset() is actually twisted.conch.insults.reset(),
which has a bug (or an undesirable feature), it sends a "hard reset"
escape sequence to the terminal (<ESC>c). This freaks out many terminal
emulators, and real, VT100-compatible terminals, in many different ways.

I've tested in Kermit-95, PuTTy, xterm, DECterm (a VT emulator written for
DECwindows, DEC's X implementation) and on real DEC VT420 and VT520
terminals. They act differently, but none are pleasant. For example, a
VT520 does a power-cycle on the CRT (which takes about 10 seconds),
reloads the saved configuration, and (if you have multiple sessions
enabled) switches to Session 1.

The VT420 and VT520 manuals say about "hard reset", basically, "Don't
do that!!"

I have a patch to change it to send a soft reset (<ESC>[!p) and to set the
terminal into "replace mode" (<ESC>[4l) (sometimes the software thought it
was in overstrike mode, but the terminal thought it was in insert mode,
with bad results. Ideally, it would set both the software and hardware to
what ever mode the user desired. The hard reset would set the terminal to
whichever mode was stored in its permanent settings. For a software
terminal emulator, it was a dice roll.)

Depending on the terminal emulator or physical terminal you are using,
this probably explains your symptoms.

See ticket # 7514: https://twistedmatrix.com/trac/ticket/7514

[Peter Westlake]

On Tue, 25 Aug 2015, at 23:58, John Santos wrote:
...

Post by John Santos
t.c.r.R.connectionMade() calls self.initializeScreen(), which in turn
calls self.terminal.reset()
I'm pretty sure s.t.reset() is actually twisted.conch.insults.reset(),
which has a bug (or an undesirable feature), it sends a "hard reset"
escape sequence to the terminal (<ESC>c). This freaks out many terminal
emulators, and real, VT100-compatible terminals, in many different ways.

...

Post by John Santos
Depending on the terminal emulator or physical terminal you are using,
this probably explains your symptoms.
See ticket # 7514: https://twistedmatrix.com/trac/ticket/7514

This does sound like a scary bug, especially on a real terminal, but I
tried your patch and it didn't affect my symptom of the screen being
cleared again. That's more about "when" rather than "how".

Peter.

[Peter Westlake]

I found it!

Manhole.terminalSize, inherited from RecvLine, does
self.terminal.eraseDisplay().
t.c.manhole_ssh.TerminalSessionTransport.__init__ calls terminalSize
*after* makeConnection.

Since setting the terminal size is a sensible thing to do, and it can't
be done before the connection is made, I'm not sure that much can be
done about this. I was able to make my message stay visible by
overriding terminalSize with a version that didn't erase, but that isn't
very clean or general.

Peter.