Probably all the state should be gathered up and put into _screen but that does not have to be done right now. Rin's fix will make aspell work which is a win already.
I am not so sure that our curses is multi-thread safe in a lot of areas. I think that multiple terminals controlled by curses is allowed for but I have never tested it and I wouldn't be surprised if it failed spectacularly.
----- Original Message -----
From:
"Valery Ushakov" <uwe%stderr.spb.ru@localhost>
To:
<tech-userlevel%netbsd.org@localhost>
Cc:
"Brett Lymn" <blymn%internode.on.net@localhost>, "Rin Okuyama" <rokuyama.rk%gmail.com@localhost>
Sent:
Tue, 12 Mar 2019 13:25:13 +0300
Subject:
Re: namespace pollution by curses
On Tue, Mar 12, 2019 at 11:41:39 +1030, Brett Lymn wrote:
> I am happy with either the rename of the static definition or
> including the state variable in _cursesi_screen which already holds
> the tty information anyway. Mind you, if you put the state variable
> into _cursesi_screen then you may get the situation that someone
> swtiches screens part way through a multi-character key sequence which
> will result in a partial escape sequence being delivered to the new
> screen which is unlikely but undesirable. If we stick with the
> static variable then the worst that happens is the assembled key code
> will be delivered to the new screen.
>
> To be honest, I think the screen switching scenario is so unlikely
> that putting the variable there would be fine.
Doesn't that just mean that inbuf &c should be in __screen too? As I
said, since infd is in __screen, shouldn't everything down the data
path be also per screen?
Admittedly, I'm not sure about the usage. E.g. in wscons case you can
press a modifier on one keyboard and the key on another and it should
work. But in case of curses, do the users really expect to be able to
input the beginning of a multi-byte sequence on one screen (via that
screen-private infd) and the end of if on another screen (via its own
screen-private infd) and still get a composed (via static
state/inbuf/...) wide character sent to the second screen?
-uwe