[9front] patch ssh.c extract window resolution

[9front] patch ssh.c extract window resolution

[Steve Simon]

changes to ssh.c to extract the current window resolution,
LINES and COLS still have precidence.

I use this as I ssh to osx without running vt(1) - I want the rio
line editing experience and have no need of cursor addressing.

I am not sure I like pulling draw.h into ssh to do this,
but I like a pipe to an external program even less.

Does not track window changes (that feels a complexity too far).

-Steve

yesterday -d ssh.c
diff /n/dump/2021/0202/sys/src/cmd/ssh.c /sys/src/cmd/ssh.c
6a7
> #include <draw.h>
307c308
< 		sysfatal("write: %r");
---
> 		sysfatal("sendpkt write fmt=%s #=%d: %r", fmt, n);
1165a1167,1188
> 
> static Point
> fontsize(void)
> {
> 	Font *f;
> 	Point sz;
> 	char *fontname;
> 
> 	if((fontname = getenv("font")) == nil)
> 		return Pt(8, 12);
> 
> 	if((f = openfont(nil, fontname)) == nil){
> 		fprint(2, "%s: %s cannot open - %r\n", argv0, fontname);
> 		free(fontname);
> 		return Pt(8, 12);
> 	}
> 	sz = stringsize(f, "0");
> 	freefont(f);
> 	free(fontname);
> 	return sz;
> }
> 
1166a1190,1233
> screensize(void)
> {
> 	int n, wfd;
> 	Point sz;
> 	char *a[6], buf[64];
> 
> 	if((wfd = open("/dev/wctl", OREAD)) < 0){
> 		return;
> 	}
> 
> 	/* wait for event, but don't care what it says */
> 	if((n = read(wfd, buf, sizeof buf)) < 0){
> 		fprint(2, "%s: /dev/wctl read failed - %r\n", argv0);
> 		close(wfd);
> 		return;
> 	}
> 	close(wfd);
> 
> 	buf[n-1] = 0;
> 	if((n = tokenize(buf, a, nelem(a))) < 4){
> 		fprint(2, "%s: /dev/wctl too few tokens (%d<4)\n", argv0, n);
> 		return;
> 	}
> 
> 	sz = fontsize();
> 
> 	/* This code lifted from mc.c, and is correct for rio(1) windows.
> 	 *
> 	 * 4 pixels left edge
> 	 * 1 pixels gap
> 	 * 12 pixels scrollbar
> 	 * 4 pixels gap
> 	 * text
> 	 * 4 pixels right edge
> 	 *
> 	 * 4 pixels top and bottom edges
> 	 */
> 	tty.ypixels = atoi(a[3]) - atoi(a[1]);
> 	tty.xpixels = atoi(a[2]) - atoi(a[0]);
> 	tty.lines = (tty.ypixels - (4+4)) / sz.y;
> 	tty.cols = (tty.xpixels - (4+1+12+4+4)) / sz.x;
> }
> 
> void
1170a1238,1239
> 	screensize();
> 
1186a1256
> 	fprint(2, "'%s' %d %d, %d %d \n", tty.term, tty.cols, tty.lines, tty.xpixels, tty.ypixels);
1410a1481
> 

Re: [9front] patch ssh.c extract window resolution

[kvik]

> sz = stringsize(f, "0");

This is going to be quite wrong for variable width fonts.


It seems that we could avoid this code (also in mc, etc.) if rio
provided these metrics in a per-window file.

Re: [9front] patch ssh.c extract window resolution

[Steve Simon]

hi all

i realise i didn't say this patch is of use.

when i ssh onto unix or osx i don't use vt(1).
i really like using rio to generate commands using cut and paste, send etc. i like my unix to look and feel like plan9

this means the environment variables ssh expects from vt dont exist. as a result some unix apps like ls and man don’t format their output correctly. this patch allows them to work as expected

re: variable width fonts
as you say, window width has no meaning under these circumstances, however if the app uses tabs you can still get sane formatting.

-Steve

Re: [9front] patch ssh.c extract window resolution

[Lyndon Nerenberg]

[-- Attachment #1: Type: text/plain, Size: 650 bytes --]

> this means the environment variables ssh expects from vt dont exist. as =
a re=3D
> sult some unix apps like ls and man don=3DE2=3D80=3D99t format their out=
put correc=3D
> tly. this patch allows them to work as expected

In my UNIX .env (only sourced for "interactive" shells), if $TERM
is unset I set TERM=3Ddumb, PAGER=3D'col -b', and EDITOR=3Ded.  That seems
to get around most of the terminal-related issues.  And if you need
more than 80 columns of output, setting cols via stty(1) will take
care of that for programs that pay attention.

I also have a tiny UNIX program that prints column rulers, to help
with this sort of thing.

--lyndon

[-- Attachment #2.1: Type: text/plain, Size: 380 bytes --]

from postmaster@1ess:
The following attachment had content that we can't
prove to be harmless.  To avoid possible automatic
execution, we changed the content headers.
The original header was:

	Content-Type: text/x-c; name="ruler.c"; charset="us-ascii"
	Content-Description: ruler.c
	Content-Disposition: attachment; filename="ruler.c"
	Content-Transfer-Encoding: quoted-printable

[-- Attachment #2.2: ruler.c.suspect --]
[-- Type: application/octet-stream, Size: 874 bytes --]

/* ruler [cols] -- print a column-count ruler */

#include <math.h>
#include <stdio.h>
#include <stdlib.h>
#include <sys/ioctl.h>

int
main (int argc, char **argv)
{

	int cols, rows, i, j, skip;
	char *buf, *p;

	cols = -1;
	if (argc > 1) {
		cols = atoi(argv[1]);
		if (cols == 0)
			exit(1);
	}
	if (cols == -1) {
		struct winsize ws;
		if (ioctl(0, TIOCGWINSZ, &ws) == -1)
			exit(1);
		if (ws.ws_col > 0)
			cols = ws.ws_col;
	}
	if (cols <= 0)
		cols = 80;
		
	buf = malloc(cols + 1);
	if (buf == NULL) {
		fprintf(stderr, "out of memory");
		exit(1);
	}

	rows = log10(cols);
	for (i = rows; i >= 0 ; i--) {
		skip = pow(10, i);
		p = buf;
		for (j = 1; j <= cols; j++) {
			if (j % skip)
				*p++ = ' ';
			else
				sprintf(p++, "%d", (j / skip) % 10);
		}
		buf[cols] = '\0';
		printf("%s\n", buf);
	}
	exit(0);
}

Re: [9front] patch ssh.c extract window resolution

[ori]

Quoth kvik@a-b.xyz:
> 
> It seems that we could avoid this code (also in mc, etc.) if rio
> provided these metrics in a per-window file.
> 

I don't think that rio should be providing metrics that make no
sense if you use a variable width font. Not sure how I feel
about making ssh provide them either, though.

Since ssh uses getenv and interrupts, perhaps this can go into
a program outside of ssh that runs in the same env?

	setsize ssh yourhost

where sizehack() is essentially:

	void
	threadmain(int arc, char **argv)
	{
		pid = spawn(argv+1, argc-1);
		while(1){
			recv(mc->resizec, nil);
			sz = windowsize();
			setenv(...);
			write("/proc/%d/note", pid, "interrupt");
	}