[9fans] Is POSIX:clock_gettime(2) in APE somewhere/anywhere?

[9fans] Is POSIX:clock_gettime(2) in APE somewhere/anywhere?

[tlaronde]

Hello,

LaTeX now requires primitives (extensions) that are not provided by TeX
nor e-TeX.

I have hence developed a new engine: Prote, that adds the required
extensions and that will be the main target for the future enhancements.

(The main purpose is to have one and only one engine, TeX compatible,
e-TeX compatible and LaTeX compatible, with only libc requirements and
BSD like licence, in order to set TeX as the OSes typographical engine.)

Amongst the extensions I had to add, there are two bits that go beyond
ISO C libc. One is a way to get elapsed time and unfortunately clock(3)
is not the answer. I then used POSIX.1:clock_gettime(2) with CLOCK_MONOTONIC
or CLOCK_REALTIME if the previous is not defined.

But is clock_gettime(2) available in APE in any instance of Plan9 or is
there a routine achieving the same purpose?

TIA,
-- 
        Thierry Laronde <tlaronde +AT+ polynum +dot+ com>
                     http://www.kergis.com/
                    http://kertex.kergis.com/
                       http://www.sbfa.fr/
Key fingerprint = 0FF7 E906 FBAF FE95 FD89  250D 52B1 AE95 6006 F40C

------------------------------------------
9fans: 9fans
Permalink: https://9fans.topicbox.com/groups/9fans/T1992d89fd8b382c7-M8861017ed3da89d5efd38e78
Delivery options: https://9fans.topicbox.com/groups/9fans/subscription

Re: [9fans] Is POSIX:clock_gettime(2) in APE somewhere/anywhere?

[ori]

Quoth tlaronde@polynum.com:
> Hello,
> 
> LaTeX now requires primitives (extensions) that are not provided by TeX
> nor e-TeX.
> 
> I have hence developed a new engine: Prote, that adds the required
> extensions and that will be the main target for the future enhancements.
> 
> (The main purpose is to have one and only one engine, TeX compatible,
> e-TeX compatible and LaTeX compatible, with only libc requirements and
> BSD like licence, in order to set TeX as the OSes typographical engine.)
> 
> Amongst the extensions I had to add, there are two bits that go beyond
> ISO C libc. One is a way to get elapsed time and unfortunately clock(3)
> is not the answer. I then used POSIX.1:clock_gettime(2) with CLOCK_MONOTONIC
> or CLOCK_REALTIME if the previous is not defined.
> 
> But is clock_gettime(2) available in APE in any instance of Plan9 or is
> there a routine achieving the same purpose?
> 

Currently, no. It can be added without too much
trouble, though we'd need to create /dev/monotime
to get properly monotonic time.


------------------------------------------
9fans: 9fans
Permalink: https://9fans.topicbox.com/groups/9fans/T1992d89fd8b382c7-M180ef7ef96fa06ad00c452eb
Delivery options: https://9fans.topicbox.com/groups/9fans/subscription

Re: [9fans] Is POSIX:clock_gettime(2) in APE somewhere/anywhere?

[Sigrid Solveig Haflínudóttir]

I've been using cycles() with _tos on amd64/arm64 with 9front for my
porting needs, specifically to get monotonic time:
https://git.sr.ht/~ft/npe/tree/master/item/libnpe/_npe.c#L15

------------------------------------------
9fans: 9fans
Permalink: https://9fans.topicbox.com/groups/9fans/T1992d89fd8b382c7-M6470aec9234ed1849753fcd8
Delivery options: https://9fans.topicbox.com/groups/9fans/subscription

Re: [9fans] Is POSIX:clock_gettime(2) in APE somewhere/anywhere?

[tlaronde]

Le Mon, Oct 11, 2021 at 03:12:11PM -0400, ori@eigenstate.org a écrit :
> Quoth tlaronde@polynum.com:
> > Hello,
> > 
> > LaTeX now requires primitives (extensions) that are not provided by TeX
> > nor e-TeX.
> > 
> > I have hence developed a new engine: Prote, that adds the required
> > extensions and that will be the main target for the future enhancements.
> > 
> > (The main purpose is to have one and only one engine, TeX compatible,
> > e-TeX compatible and LaTeX compatible, with only libc requirements and
> > BSD like licence, in order to set TeX as the OSes typographical engine.)
> > 
> > Amongst the extensions I had to add, there are two bits that go beyond
> > ISO C libc. One is a way to get elapsed time and unfortunately clock(3)
> > is not the answer. I then used POSIX.1:clock_gettime(2) with CLOCK_MONOTONIC
> > or CLOCK_REALTIME if the previous is not defined.
> > 
> > But is clock_gettime(2) available in APE in any instance of Plan9 or is
> > there a routine achieving the same purpose?
> > 
> 
> Currently, no. It can be added without too much
> trouble, though we'd need to create /dev/monotime
> to get properly monotonic time.

Thanks for the answer;

For the time being, I will then simply make the "timer" with a
granularity of 1 second only simply diffing times on plan9.
-- 
        Thierry Laronde <tlaronde +AT+ polynum +dot+ com>
                     http://www.kergis.com/
                    http://kertex.kergis.com/
                       http://www.sbfa.fr/
Key fingerprint = 0FF7 E906 FBAF FE95 FD89  250D 52B1 AE95 6006 F40C

------------------------------------------
9fans: 9fans
Permalink: https://9fans.topicbox.com/groups/9fans/T1992d89fd8b382c7-M4dd508f8a596794104d451af
Delivery options: https://9fans.topicbox.com/groups/9fans/subscription

Re: [9fans] Is POSIX:clock_gettime(2) in APE somewhere/anywhere?

[tlaronde]

Le Mon, Oct 11, 2021 at 09:20:31PM +0200, Sigrid Solveig Haflínudóttir a écrit :
> I've been using cycles() with _tos on amd64/arm64 with 9front for my
> porting needs, specifically to get monotonic time:
> https://git.sr.ht/~ft/npe/tree/master/item/libnpe/_npe.c#L15

Thank you for the hint. I will take a look.
-- 
        Thierry Laronde <tlaronde +AT+ polynum +dot+ com>
                     http://www.kergis.com/
                    http://kertex.kergis.com/
                       http://www.sbfa.fr/
Key fingerprint = 0FF7 E906 FBAF FE95 FD89  250D 52B1 AE95 6006 F40C

------------------------------------------
9fans: 9fans
Permalink: https://9fans.topicbox.com/groups/9fans/T1992d89fd8b382c7-M46eab90c250799d3e308a8b9
Delivery options: https://9fans.topicbox.com/groups/9fans/subscription

Re: [9fans] Is POSIX:clock_gettime(2) in APE somewhere/anywhere?

[tlaronde]

Le Mon, Oct 11, 2021 at 08:39:31PM +0200, tlaronde@polynum.com a écrit :
>[...] 
> Amongst the extensions I had to add, there are two bits that go beyond
> ISO C libc. One is a way to get elapsed time and unfortunately clock(3)
> is not the answer. I then used POSIX.1:clock_gettime(2) with CLOCK_MONOTONIC
> or CLOCK_REALTIME if the previous is not defined.
> 
> But is clock_gettime(2) available in APE in any instance of Plan9 or is
> there a routine achieving the same purpose?

For what I need, in fact I can simply use nsec(2), diffing nanoseconds
times! (I have the printed manuals from Vita Nuova, I should remember
to read the man pages first in printed form at leisure...)
-- 
        Thierry Laronde <tlaronde +AT+ polynum +dot+ com>
                     http://www.kergis.com/
                    http://kertex.kergis.com/
                       http://www.sbfa.fr/
Key fingerprint = 0FF7 E906 FBAF FE95 FD89  250D 52B1 AE95 6006 F40C

------------------------------------------
9fans: 9fans
Permalink: https://9fans.topicbox.com/groups/9fans/T1992d89fd8b382c7-M7450a2a55a3baa87b5a2f993
Delivery options: https://9fans.topicbox.com/groups/9fans/subscription

Re: [9fans] Is POSIX:clock_gettime(2) in APE somewhere/anywhere?

[Thaddeus Woskowiak]

[-- Attachment #1: Type: text/plain, Size: 682 bytes --]

The Vita Nuova man pages are woefully outdated. You might want to consider
purchasing the printed 9front man pages. http://9front.org/propaganda/books

Of course the most up-to-date man pages are available via man after a
sysupdate.


> For what I need, in fact I can simply use nsec(2), diffing nanoseconds
> times! (I have the printed manuals from Vita Nuova, I should remember
> to read the man pages first in printed form at leisure...)
>

------------------------------------------
9fans: 9fans
Permalink: https://9fans.topicbox.com/groups/9fans/T1992d89fd8b382c7-M396f6268e903dd155be665ea
Delivery options: https://9fans.topicbox.com/groups/9fans/subscription

[-- Attachment #2: Type: text/html, Size: 1449 bytes --]

Re: [9fans] Is POSIX:clock_gettime(2) in APE somewhere/anywhere?

[ori]

Quoth tlaronde@polynum.com:
> But is clock_gettime(2) available in APE in any instance of Plan9 or is
> there a routine achieving the same purpose?

Here's a quick sketch of something that we could add; note
that the implemented monotonicity is really crude.

diff 03d870e0283299404b0eb46689d13f8538e83a2f uncommitted
--- a//sys/include/ape/time.h
+++ b//sys/include/ape/time.h
@@ -53,15 +53,20 @@
 #ifdef _POSIX_SOURCE
 extern void tzset(void);
 
+#define        CLOCK_REALTIME  1
+#define        CLOCK_MONOTONIC 2
+
+typedef int    clockid_t;
+
 struct timespec {
        time_t tv_sec;
        long tv_nsec;
 };
 extern int nanosleep(const struct timespec *req, struct timespec *rem);
-#endif
 
-#ifdef __cplusplus
-}
+extern int     clock_getres(clockid_t, struct timespec*);
+extern int     clock_gettime(clockid_t, struct timespec*);
+extern int     clock_settime(clockid_t, struct timespec*);
 #endif
 
 #ifdef _POSIX_SOURCE
@@ -69,6 +74,10 @@
 extern long timezone;
 extern long altzone;
 extern int daylight;
+#endif
+
+#ifdef __cplusplus
+}
 #endif
 
 #endif /* __TIME_H */
--- /tmp/diff100001791256
+++ b/sys/src/ape/lib/bsd/clock.c
@@ -1,0 +1,60 @@
+#include <sys/types.h>
+#include <time.h>
+#include <errno.h>
+#include <lock.h>
+
+/* ap/plan9/9nsec.c */
+extern long long _NSEC(void);
+
+#define Ns2sec (1000*1000*1000)
+
+int
+clock_getres(clockid_t id, struct timespec *ts)
+{
+       if(id != CLOCK_REALTIME && id != CLOCK_MONOTONIC){
+               errno = EINVAL;
+               return -1;
+       }
+       ts->tv_sec = 0;
+       ts->tv_nsec = 1;
+       return 0;
+}
+
+int
+clock_settime(clockid_t, struct timespec *)
+{
+       errno = EPERM;
+       return -1;
+}
+
+int
+clock_gettime(clockid_t clk, struct timespec *ts)
+{
+       static Lock maxlk;
+       static long long max;
+       long long t;
+
+       t = _NSEC();
+       switch(clk){
+       case CLOCK_REALTIME:
+               /* nothing */
+               break;
+       case CLOCK_MONOTONIC:
+               /* Bug: we should support a real monotonic clock */
+               lock(&maxlk);
+               if(t < max)
+                       t = max;
+               else
+                       max = t;
+               unlock(&maxlk);
+               break;
+       default:
+               errno = EINVAL;
+               return -1;
+       }
+
+       ts->tv_sec = t/Ns2sec;
+       ts->tv_nsec = t%Ns2sec;
+
+       return 0;
+}



------------------------------------------
9fans: 9fans
Permalink: https://9fans.topicbox.com/groups/9fans/T1992d89fd8b382c7-Mf08f6aaf225fc08a5763ecb2
Delivery options: https://9fans.topicbox.com/groups/9fans/subscription

Re: [9fans] Is POSIX:clock_gettime(2) in APE somewhere/anywhere?

[tlaronde]

Le Thu, Oct 14, 2021 at 05:54:44PM -0400, ori@eigenstate.org a écrit :
> Quoth tlaronde@polynum.com:
> > But is clock_gettime(2) available in APE in any instance of Plan9 or is
> > there a routine achieving the same purpose?
> 
> Here's a quick sketch of something that we could add; note
> that the implemented monotonicity is really crude.
> 
> diff 03d870e0283299404b0eb46689d13f8538e83a2f uncommitted
> --- a//sys/include/ape/time.h
> +++ b//sys/include/ape/time.h
> @@ -53,15 +53,20 @@
>  #ifdef _POSIX_SOURCE
>  extern void tzset(void);
>  
> +#define        CLOCK_REALTIME  1
> +#define        CLOCK_MONOTONIC 2
> +
> +typedef int    clockid_t;
> +
>  struct timespec {
>         time_t tv_sec;
>         long tv_nsec;
>  };
>  extern int nanosleep(const struct timespec *req, struct timespec *rem);
> -#endif
>  
> -#ifdef __cplusplus
> -}
> +extern int     clock_getres(clockid_t, struct timespec*);
> +extern int     clock_gettime(clockid_t, struct timespec*);
> +extern int     clock_settime(clockid_t, struct timespec*);
>  #endif
>  
>  #ifdef _POSIX_SOURCE
> @@ -69,6 +74,10 @@
>  extern long timezone;
>  extern long altzone;
>  extern int daylight;
> +#endif
> +
> +#ifdef __cplusplus
> +}
>  #endif
>  
>  #endif /* __TIME_H */
> --- /tmp/diff100001791256
> +++ b/sys/src/ape/lib/bsd/clock.c
> @@ -1,0 +1,60 @@
> +#include <sys/types.h>
> +#include <time.h>
> +#include <errno.h>
> +#include <lock.h>
> +
> +/* ap/plan9/9nsec.c */
> +extern long long _NSEC(void);
> +
> +#define Ns2sec (1000*1000*1000)
> +
> +int
> +clock_getres(clockid_t id, struct timespec *ts)
> +{
> +       if(id != CLOCK_REALTIME && id != CLOCK_MONOTONIC){
> +               errno = EINVAL;
> +               return -1;
> +       }
> +       ts->tv_sec = 0;
> +       ts->tv_nsec = 1;
> +       return 0;
> +}
> +
> +int
> +clock_settime(clockid_t, struct timespec *)
> +{
> +       errno = EPERM;
> +       return -1;
> +}
> +
> +int
> +clock_gettime(clockid_t clk, struct timespec *ts)
> +{
> +       static Lock maxlk;
> +       static long long max;
> +       long long t;
> +
> +       t = _NSEC();
> +       switch(clk){
> +       case CLOCK_REALTIME:
> +               /* nothing */
> +               break;
> +       case CLOCK_MONOTONIC:
> +               /* Bug: we should support a real monotonic clock */
> +               lock(&maxlk);
> +               if(t < max)
> +                       t = max;
> +               else
> +                       max = t;
> +               unlock(&maxlk);
> +               break;
> +       default:
> +               errno = EINVAL;
> +               return -1;
> +       }
> +
> +       ts->tv_sec = t/Ns2sec;
> +       ts->tv_nsec = t%Ns2sec;
> +
> +       return 0;
> +}

Since only CLOCK_REALTIME is mandatory (up to current POSIX 7), you could drop the MONOTONIC
variant. And nsec(2) is good enough for REALTIME.

Thanks for working on this!

Best,
-- 
        Thierry Laronde <tlaronde +AT+ polynum +dot+ com>
                     http://www.kergis.com/
                    http://kertex.kergis.com/
                       http://www.sbfa.fr/
Key fingerprint = 0FF7 E906 FBAF FE95 FD89  250D 52B1 AE95 6006 F40C

------------------------------------------
9fans: 9fans
Permalink: https://9fans.topicbox.com/groups/9fans/T1992d89fd8b382c7-Me0d4b6affdd5977cbe6d2c8b
Delivery options: https://9fans.topicbox.com/groups/9fans/subscription