Re: [9front] time v4

[BurnZeZ@feline.systems]

[-- Attachment #1: Type: text/plain, Size: 687 bytes --]

On Sun Aug 12 05:40:12 GMT 2018, alex@musolino.id.au wrote:
> What's the point of the segattach?  Can you not just add RFMEM to the
> rfork flags?
char	output[4096];
output would be shared, and the child process uses it for the exec
path stuff.
I thought about using RFMEM at first, but it ended up being simpler
this way.

> Also, you probably ought to be setting t0 again before the second call
> to exec.
You would miss out on accounting for 9p latency for the first exec()
attempt, but I think you’re right, since the fact that t0 is set
after the fork() indicates we’re not worried about absolute time
from the parent process’ perspective.

Attached diff contains that change.

[-- Attachment #2: Type: text/plain, Size: 1129 bytes --]

diff -r 5c5acc9ab7a5 sys/src/cmd/time.c
--- a/sys/src/cmd/time.c	Sat Aug 11 16:19:32 2018 +0200
+++ b/sys/src/cmd/time.c	Sun Aug 12 18:17:11 2018 +0000
@@ -14,20 +14,27 @@
 	long l;
 	char *p;
 	char err[ERRMAX];
+	vlong *t0, t1;
+	enum { SEC = 1000000000ULL };
 
 	if(argc <= 1){
 		fprint(2, "usage: time command\n");
 		exits("usage");
 	}
 
+	if((t0 = segattach(SG_CEXEC, "shared", nil, sizeof(*t0))) == (void*)-1)
+		sysfatal("segattach: %r");
+	*t0 = 0;
 	switch(fork()){
 	case -1:
 		error("fork");
 	case 0:
+		*t0 = nsec();
 		exec(argv[1], &argv[1]);
 		if(argv[1][0] != '/' && strncmp(argv[1], "./", 2) &&
 		   strncmp(argv[1], "../", 3)){
 			sprint(output, "/bin/%s", argv[1]);
+			*t0 = nsec();
 			exec(output, &argv[1]);
 		}
 		error(argv[1]);
@@ -43,12 +50,14 @@
 			goto loop;
 		error("wait");
 	}
+	t1 = nsec();
 	l = w->time[0];
 	add("%ld.%.2ldu", l/1000, (l%1000)/10);
 	l = w->time[1];
 	add("%ld.%.2lds", l/1000, (l%1000)/10);
 	l = w->time[2];
 	add("%ld.%.2ldr", l/1000, (l%1000)/10);
+	add("%lld.%.9lldt", (t1-*t0)/SEC, (t1-*t0)%SEC);
 	add("\t");
 	for(i=1; i<argc; i++){
 		add("%s", argv[i], 0);