Very Computer

I can think of three different ways:
 - The co# and li# entries in /etc/termcap (or the terminfo equivalent)
 - Environment variables COLUMNS and LINES
 - An ioctl call with TIOCGWINSZ

But what is considered the right order to try these methods to get the
real size? Also, I believe ioctl(TIOCGWINSZ...) is not POSIX; is there
a POSIX function for this?

Thanks.

--

Quote:Kenneth Falck writes:
>I'd like to know the correct method(s) of finding out a terminal's
>screen size, considering that it may change as a SIGWINCH occurs
>(under X, for example).

>I can think of three different ways:
> - The co# and li# entries in /etc/termcap (or the terminfo equivalent)
> - Environment variables COLUMNS and LINES
> - An ioctl call with TIOCGWINSZ

>But what is considered the right order to try these methods to get the
>real size? Also, I believe ioctl(TIOCGWINSZ...) is not POSIX; is there
>a POSIX function for this?

Quote:> - An ioctl call with TIOCGWINSZ

Quote:> - The co# and li# entries in /etc/termcap (or the terminfo equivalent)

Quote:> - Environment variables COLUMNS and LINES

I have a few questions of my own?

- Why is it that when I log in from an xterm to a Sun, then resize,
TIOCGWINSZ on the remote sun gives the right answer, but when I log in
to an rs6000, it does not?

- Which headers should be included?
- Which is preferable, or standard?
<sys/ioctl.h>
<sys/termios.h>
<sgtty.h>

Brian

Quote:>>Also, I believe ioctl(TIOCGWINSZ...) is not POSIX; is there
>>a POSIX function for this?

Quote:>- Why is it that when I log in from an xterm to a Sun, then resize,
>TIOCGWINSZ on the remote sun gives the right answer, but when I log in
>to an rs6000, it does not?

Quote:>- Which headers should be included?
>- Which is preferable, or standard?
><sys/ioctl.h>
><sys/termios.h>
><sgtty.h>

If we restrict ourselves to systems with "termio(s)" - which blows away
pre-4.3-reno BSD systems, for example - we find that:

        1) 4.3-reno and Net-2 define TIOCGWINSZ and TIOCSWINSZ in
           <sys/ioctl.h>, and do not automatically include <sys/ioctl.h>
           if you include <sys/termios.h>, so you can't just include
           <sys/termios.h> on those systems - you have to include
           <sys/ioctl.h> (or something that forces it to be included).

        2) SunOS 4.x defines them in <sys/ttycom.h>, which is included
           by <sys/termios.h>, although you have to make sure
           _POSIX_SOURCE is *not* defined (if it *is* defined, symbols
           such as "TIOC[GS]WINSZ" that "pollute the namespace" aren't
           defined).  You can, therefore, get them with <sys/termios.h>.

           You can also get them with <sys/ioctl.h>, as that includes
           <sys/ttold.h>, which includes <sys/ttycom.h> - *however*,
           <sys/ioctl.h>, for compatibility with pre-4.3-reno BSD
           releases, defines stuff for the older V7-flavored tty
           driver that collide with definitions in <sys/termios.h>
           (e.g., ECHO).

        3) SunOS 5.x defines them in <sys/termios.h> as well - again,
           only if _POSIX_SOURCE isn't defined.  <sys/ioctl.h> appears
           to act in the SunOS 4.x fashion only if BSD_COMP is defined.

           I don't have a "more vanilla" SVR4 system handy; however, the
           SVR4 documentation mentions TIOCGWINSZ and TIOCSWINSZ in the
           "termios(7)" manual page, so I suspect it gives you
           TIOC[GS]WINSZ if you include <sys/termios.h> and don't have
           _POSIX_SOURCE defined.  I suspect you may *not* get them if
           you include <sys/ioctl.h> (it may do so if BSD_COMP is
           defined).

        4) AIX 3.2 defines them in <sys/ioctl.h>, although _ALL_SOURCE
           has to be defined to have "struct winsize" defined.

           <sys/ioctl.h> has a comment that it contains "ioctl
           definitions a la 4.xBSD for BSD to AIX porting tools"; I've
           no idea whether or not they're under the delusion that
           TIOC[GS]WINSZ's only reason for existence is BSD compatibility.

           <sys/termio.h> doesn't define them, and there doesn't appear
           to be a <sys/termios.h> - at least not in "/usr/include/sys".

        5) HP-UX 8.07 (or whatever the hell version the machine in
           question is running) defines them in <sys/termio.h>, which is
           included by <sys/termios.h>; I haven't bothered checking what
           stuff has to be defined to turn that on, as their "vi" is
           singularly obnoxious - even though "stty size" correctly
           reports that my window has 66 rows, "vi" seems to insist that
           it doesn't have 66 rows....

           <sys/ioctl.h> includes only <sys/stdsyms.h>, which includes
           nothing; <sys/ioctl.h> doesn't define TIOC[GS]WINSZ.

So systems seem to fall into two camps:

        1) define it in <sys/termios.h> (at least if "strict POSIX
           namespace" isn't turned on), and either don't define it in
           <sys/ioctl.h> or don't encourage <sys/ioctl.h> to coexist
           with <sys/termios.h>;

        2) define it in <sys/ioctl.h> but not in <sys/termios.h>.

I somewhat prefer camp 1) (then again, I'm the one who put SunOS 4.x
*into* that camp); the fact that TIOC[GS]WINSZ pollute the POSIX
namespace doesn't bother me, as plenty of other tty driver features that
belong in <sys/termios.h> do so as well, so you're stuck with a "strict
POSIX" flag in any case.

If camp 1) causes some people a real problem, I'd vote for camp 3) -
have some separate include file, <sys/ttywinsz.h> or something such as
that - as long as *everybody* agrees on it, so you can write software
that uses "termios" and TIOC[GS]WINSZ without having to pollute it with
lots of stupid #ifdefs.