new lsub nix patch: nsec

[Fco. J. Ballesteros]

http://lsub.org/magic/webls?dir=/nixpatch/nsec

cleanup nsec not to cache anything.
files:
/sys/src/libc/9sys/nsec.c nsec.c
removed:

/sys/src/libc/9sys/nsec.c:
nsec.c.orig:19,24 d nsec.c:18
< static int fd = -1;
< static struct {
< int pid;
< int fd;
< } fds[64];
<
nsec.c.orig:30 c nsec.c:24
< int pid, i, f, tries;
---
> int f;
nsec.c.orig:32,74 c nsec.c:26,33
< /*
< * Threaded programs may have multiple procs
< * with different fd tables, so we may need to open
< * /dev/bintime on a per-pid basis
< */
<
< /* First, look if we've opened it for this particular pid */
< pid = _tos->pid;
< do{
< f = -1;
< for(i = 0; i < nelem(fds); i++)
< if(fds[i].pid == pid){
< f = fds[i].fd;
< break;
< }
< tries = 0;
< if(f < 0){
< /* If it's not open for this pid, try the global pid */
< if(fd >= 0)
< f = fd;
< else{
< /* must open */
< if((f = open("/dev/bintime", OREAD|OCEXEC)) < 0)
< return 0;
< fd = f;
< for(i = 0; i < nelem(fds); i++)
< if(fds[i].pid == pid || fds[i].pid == 0){
< fds[i].pid = pid;
< fds[i].fd = f;
< break;
< }
< }
< }
< if(pread(f, b, sizeof b, 0) == sizeof b){
< be2vlong(&t, b);
< return t;
< }
< close(f);
< if(i < nelem(fds))
< fds[i].fd = -1;
< }while(tries++ == 0); /* retry once */
< USED(tries);
< return 0;
---
> f = open("/dev/bintime", OREAD);
> if(f < 0)
> return 0;
> t = 0;
> if(pread(f, b, sizeof b, 0) == sizeof b)
> be2vlong(&t, b);
> close(f);
> return t;

[erik quanstrom]

lgtm.

also there was a lot of back-and-forth about timing
and time bases. i'd like to gather up something concrete
that allows for
- arch independent interface
- nanosecond resolution timing
- known precision.
- unspecified units & post division.
- maximum range ~ ±2 years.

perhaps
vlong tsprec()
vlong ts(void);
vlong tstonsec();

i realize the cool thing to do these days is to parameterize everything,
but i just spent a few days suffering a set of bugs that was caused by changing
the parameterized type in a non-obvious way, and with no easy way to
revert, it was a real mess. i don't see why vlong can't be good enough.

- erik

[Nemo]

yes. I think I agree.

;; patch/Apply nsec
applying to /n/src/nix...
/dist/patch/nsec
merge...backup...copy...
cpfile nsec.c /n/dist/sys/src/libc/9sys/nsec.c
done

[Charles Forsyth]

It isn't just cool, it's because having to wade through years later working out

just how big each and every ulong needs to be (or what it really represents)

remains as tedious as it was going from 6th to 7th Edition  (in a smaller source).

[erik quanstrom]

On Sun Apr 15 13:11:36 EDT 2012, charles...@gmail.com wrote:

> It isn't just cool, it's because having to wade through years later working
> out
> just how big each and every ulong needs to be (or what it really represents)
> remains as tedious as it was going from 6th to 7th Edition (in a smaller
> source).

i wasn't alluding to anything in the plan 9 or nix kernel, but
rather some recent mistakes in ken's kernel such as Timet ticks
(Timet is long) replacing ulong ticks.

- erik