Re: [9front] vt: fix scrolling over ssh, add mouse scrolling

[unobe@cpan.org]

Quoth ori@eigenstate.org:
> Testing requested.
> 
> We were setting 'resize_flag = 1' in backup(), which was
> causing resized() to get called, which was causing us to
> call exportsize(), which was eventually singalling sshd,
> which forwarded the signal over to the process on the
> other end of the ssh pipe.
> 
> This was causing a SIGINT to get sent, which was screwing
> up the scrolling.

Reviewing the logic of what resize_flag does, your patch makes sense:
when can you ever resize the window when scrolling/paging back or
forward?  I don't see how that can be done simultaneously, but maybe
there's something else afoot.

I want to make sure I understand what you're experiencing so I can
test it properly.  Here's what I've pieced togeter without running
anything:

1.  You start vt(1)

2.  You ssh(1) into a foreign host.  Since vt(1) sets the TERM
environment variable, raw mode is on in ssh(1), which sets the vt(1)
console winch to 1 @ /sys/src/cmd/vt/fs.c:/winchon due to the write @
/sys/src/cmd/ssh.c:/^rawon .

3.  You start a long-running process that spews a lot of text.

4.  You then use 'backup'.  That signals to the foreign host to set
the environment variables, which is performed.  No problem there.

5.  However, at the end of exportsize() an interrupt is sent because
ssh(1) had set winch to 1:
	if(cs->winch)
		send_interrupt();
   Thus the problem?

Related to this, I see resize_flag is set to 1 in only one other
scenario, and that deals with the page menu item and when blocked is
true.  Blocked is only set to 1 in /sys/src/cmd/vt/main.c:^newline
from what I see.  That's only done after newline() is called in
/sys/src/cmd/vt/vt.c and only when:

a) pagemode is true; and

b) the number of lines goes beyond the screen size max.

So perhaps the resize_flag set to 1 for 'backup' is to handle a
special case that we're not aware of currently?  Maybe ssh.c is being
too aggressive in setting winchon? Or maybe your fix is the best.