get-internal-real-time using times()

get-internal-real-time using times()

[Kevin Ryde]

[-- Attachment #1: Type: text/plain, Size: 398 bytes --]

        * stime.c (scm_get_internal_real_time, scm_your_base, scm_init_stime):
        Use times() when available, for a genuine real-time (unaffected by
        stime system time changes).

The name "real-time" suggests to me something independent of the
system date/time.  And since the return is clock ticks, times() would
seem natural.  Unless there's a particular reason it isn't already so.


[-- Attachment #2: stime.c.real-time.diff --]
[-- Type: text/plain, Size: 1148 bytes --]

--- stime.c.~1.79.~	1970-01-01 10:00:00.000000000 +1000
+++ stime.c	2003-08-31 15:19:06.000000000 +1000
@@ -87,7 +87,13 @@
 # endif
 #endif
 
-#ifdef HAVE_FTIME
+
+/* times() returns a genuine real-time, meaning it's unaffected by system
+   time changes made by stime().  If times() isn't available (mingw for
+   instance) then we have to fall back on ftime() or time().  */
+#if HAVE_TIMES
+clock_t scm_your_base;
+#elif HAVE_FTIME
 struct timeb scm_your_base = {0};
 #else
 timet scm_your_base = 0;
@@ -99,7 +105,10 @@
 	    "started.")
 #define FUNC_NAME s_scm_get_internal_real_time
 {
-#ifdef HAVE_FTIME
+#if HAVE_TIMES
+  struct tms dummy;
+  return scm_long2num (times (&dummy) - scm_your_base);
+#elif HAVE_FTIME
   struct timeb time_buffer;
 
   SCM tmp;
@@ -664,7 +673,13 @@
   scm_c_define ("internal-time-units-per-second",
 		scm_long2num((long) SCM_TIME_UNITS_PER_SECOND));
 
-#ifdef HAVE_FTIME
+#if HAVE_TIMES
+  if (!scm_your_base)
+    {
+      struct tms dummy;
+      scm_your_base = times(&dummy);
+    }
+#elif HAVE_FTIME
   if (!scm_your_base.time) ftime(&scm_your_base);
 #else
   if (!scm_your_base) time(&scm_your_base);

[-- Attachment #3: Type: text/plain, Size: 142 bytes --]

_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://mail.gnu.org/mailman/listinfo/guile-devel

Re: get-internal-real-time using times()

[Rob Browning]

Kevin Ryde <user42@zip.com.au> writes:

>         * stime.c (scm_get_internal_real_time, scm_your_base,
>         scm_init_stime): Use times() when available, for a genuine
>         real-time (unaffected by stime system time changes).
>
> The name "real-time" suggests to me something independent of the
> system date/time.  And since the return is clock ticks, times() would
> seem natural.  Unless there's a particular reason it isn't already so.

Presuming I understand what you're doing here (from a brief look), we
might want to be careful.  If the old semantics of
get-internal-real-time relied on ftime, then people may have written
code that depends on the fact that get-internal-real-time has been
returning wall-clock time, and in fact, when contrasted to
get-internal-run-time's description (and given familiarity with the
"times" distinctions), that might be a reasonable presumption.  Or did
I misunderstand?

-- 
Rob Browning
rlb @defaultvalue.org and @debian.org; previously @cs.utexas.edu
GPG starting 2002-11-03 = 14DD 432F AE39 534D B592  F9A0 25C8 D377 8C7E 73A4


_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://mail.gnu.org/mailman/listinfo/guile-devel

Re: get-internal-real-time using times()

[Kevin Ryde]

Rob Browning <rlb@defaultvalue.org> writes:
>
> If the old semantics of
> get-internal-real-time relied on ftime, then people may have written
> code that depends on the fact that get-internal-real-time has been
> returning wall-clock time,

I guess that's possible, though I might think it unlikely anyone who
understands the distinction would assume a function called `real-time'
in fact gives wall time.

> and in fact, when contrasted to
> get-internal-run-time's description (and given familiarity with the
> "times" distinctions), that might be a reasonable presumption.

Or one could assume that since -run-time is unaffected then -real-time
is also unaffected.

If the system time doesn't change too often then I guess there's not
much difference.  But I can't help thinking that it's not good for a
function called real-time to go backwards or suddenly jump forwards.


_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://mail.gnu.org/mailman/listinfo/guile-devel

Re: get-internal-real-time using times()

[Rob Browning]

Kevin Ryde <user42@zip.com.au> writes:

> I guess that's possible, though I might think it unlikely anyone who
> understands the distinction would assume a function called `real-time'
> in fact gives wall time.

Hmm.  I don't know.  Maybe I'm daft, but I could see myself at least
wondering, -- i.e. "perhaps real-time == time in the real-world",
though as you mention...

> If the system time doesn't change too often then I guess there's not
> much difference.  But I can't help thinking that it's not good for a
> function called real-time to go backwards or suddenly jump forwards.

Hmm.  You mean because of timezones/leap-foo, etc.?  If so, good
point.  I'd say we need better docs, better functions, or both, and
either way, a good NEWS entry.

-- 
Rob Browning
rlb @defaultvalue.org and @debian.org; previously @cs.utexas.edu
GPG starting 2002-11-03 = 14DD 432F AE39 534D B592  F9A0 25C8 D377 8C7E 73A4


_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://mail.gnu.org/mailman/listinfo/guile-devel

Re: get-internal-real-time using times()

[Kevin Ryde]

Rob Browning <rlb@defaultvalue.org> writes:
>
> Hmm.  You mean because of timezones

Timezones and daylight saving won't affect ftime on a unix system, but
might on DOS where the system time is local time.

> /leap-foo,

Not sure what normally happens for leap seconds.  If the system time
is properly tied to UTC like srfi-19.scm assumes then I think for a
leap second 23:59:59 to 00:00:00 will take 2 seconds of real time.  I
imagine only serious users would attempt to enforce that though.

> etc.?

Sudden changes with stime would be the main concern, gradual changes
with adjtime would be less of a worry but still not ideal.

> If so, good
> point.  I'd say we need better docs, better functions, or both, and
> either way, a good NEWS entry.

Oh well, these sort of things always depend on the intended usage,
which is why I ask about the current theory.  I'd certainly like to
tighten up the docs though.


_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://mail.gnu.org/mailman/listinfo/guile-devel