changes

source of perl for linux application,

Revision history for the Perl extension Time::HiRes.

1.9711	[2007-11-29]
	- lost VMS test skippage from Craig Berry
	- reformat the test code a little

1.9710	[2007-11-29]
	- I got the sense of the QNX test the wrong way in an attempt
	  to generalize it for future

1.9709	[2007-11-28]
	- casting fixes from Robin Barker for g++ and 64bitint
	- in QNX skip the itimer tests because though the API
	  is there, the implementation isn't, from Matt Kraai
	- raise the dead man timer to 180 seconds for really
	  slow/busy systems
	- elaborate the UTF-8 locale warning from Makefile.PL

1.9708	[2007-10-05]
	- [rt.cpan.org #29747]: Build failure with perl 5.005_05
	  Fixed by regenerating the ppport.h using Devel::PPPort 3.13.

1.9707	[2007-02-27]
	- t/HiRes.t failed in Perl 5.6.2,
	  "action is not of type POSIX::SigAction at t/HiRes.t line 318",
	  reported and fixed by Anton Berezin, the reason was faulty
	  use of sigaction() when restoring the old action.

1.9706	[2007-02-25]
	- with bleadperl in VMS the HiRes.t overrun the maximum number
	  of deferred signals because the libc SIGALRM was not strong
	  enough to interrupt select(), and select() got restarted every
	  time, solution is to use POSIX::SigAction if available.
	  A fix from Craig Berry (not 100% there, but helps).
	- allow for more measuring noise for ualarm() tests 35..37

1.9705	[2007-02-06]
	- nanosleep() and clock_nanosleep() detection and use were
	  quite broken; in Linux -lrt needed; fixes from Zefram
	- [internal] slightly cleaner building of $DEFINE in Makefile.PL,
	  should avoid double/conflicting -D flags

1.9704	[2007-01-01]
	- allow 10% of slop in test #14 (testing difference between
	  CORE::time() and Time::HiRes::time()), there seem to be often
	  transient failures from Perl smoke builds on this test
	- small pod tweaks

1.9703	[2006-12-08]
	- use int main(int argc, char **argv) consistently in Makefile.PL,
	  should help with
	  [rt.cpan.org #23868] nanosleep not detected under Mac OS 10.3.9 starting with Time::HiRes 1.96
	- if someone still has the locale-broken Perl 5.8.0,
	  suggest that they upgrade their Perl

1.9702	[2006-12-06]
	- restore the -DATLEASTFIVEOHOHFIVE, Win32 needed it still

1.9701	[2006-12-04]
	- upgrade to ppport.h 3.10_02
	- remove the -DATLEASTFIVEOHOHFIVE
	- use the ppport.h PL_ppaddr, PL_statcache, PL_laststatval
	- use the ppport.h aTHXR for calling Perl stat()
	- switch into four-digit version since 2.0 is coming up
	  awfully fast but not feeling like a major rewrite

1.97	[2006-11-30]
	- 1.95 broke building in Win32 (since pp_stat is not exported),
	  figured out how to call an op directly in 5.005 (use Perl_ppaddr
	  instead of PL_ppaddr)
	- backport to Perl 5.004_05  (requires using statcache
	  and laststatval instead of PL_statcache and PL_laststatval)
	  (also checked to work in 5.005_04, 5.6.1, and 5.8.8 with threads)

1.96	[2006-11-30]
	- 1.95 broke builds for threaded Perls, rt.cpan.org tickets:
	  [rt.cpan.org #23694] Time::HiRes fails tests on Solaris and Perl 5.6.1
	  [rt.cpan.org #23712] Time-HiRes 1.95 Fails make on AIX 5.2 with Perl 5.8.8
	  [rt.cpan.org #23730] Time::HiRes 1.95 fails make on MacOS X 10.3.9/perl 5.8.8
	- use main() prototype consistently in Makefile.PL

1.95	[2006-11-29]
	- integrate core change #29180: Silence VC++ compiler warnings
	  from Steve Hay
	- do not use PL_ppaddr in stat() because that is not available
	  in Perl 5.005_04
	- regenerate fallback/*.inc for older Perls without
          ExtUtils::Constant because of d_hires_stat, resolves
	  [rt.cpan.org #23694] Time::HiRes fails tests on Solaris and Perl 5.6.1
	- Make Makefile.PL more defensive against false PERL_CORE

1.94	[2006-10-16]
	- file timestamps oddities seen: the atime and mtime
	  can be out of sync (modify first and read second can leave
	  atime < mtime) and mtime can be subsecond while atime is not.
	  So make the test more forgiving.

1.93	[2006-10-15]
	- the ualarm() tests (34-37) assumed that ualarm(N)
	  could never alarm in less than N seconds, widened
	  the acceptable relative range to 0.9..1.5.  Addresses
	  [rt.cpan.org #22090] and [rt.cpan.org #22091].

	- skip the stat() tests in cygwin and win32, because
	  if run on FAT the timestamp granularity is only 2 seconds.
	  Any good way to detect (cygwin or win32) whether we are
	  being run on NTFS or anywhere with better timestamps?
	  Addresses [rt.cpan.org #22089] and [rt.cpan.org #22098].

1.92	[2006-10-13]
	- scan for subsecond resolution timestamps in struct stat,
	  some known possibilities:

	  (1)  struct  timespec st_atimespec;
                                st_atimespec.tv_nsec;
	  (2)  time_t  st_atime;
	       long    st_atimensec;
          (3)  time_t  st_atime;
               int     st_atime_n;
	  (4)  timestruc_t st_atim;
                           st_atim.tv_nsec
	  (5)  time_t  st_atime;
               int     st_uatime;

          If something like this is found, one can do

	    use Time::HiRes;
	    my @stat = Time::HiRes::stat();

	  or even override the standard stat():
      
	    use Time::HiRes qw(stat);

          to get the stat() timestamps

	    my ($atime, $mtime, $ctime) = @stat[8, 9, 10];

	  with subsecond resolution (assuming both the operating
	  system and the filesystem support that kind of thing).

	  Contributions for more systems (especially non-UNIX,
	  e.g. but not limited to: Win32, VMS, OS/2) gladly accepted.
	  (also more UNIX variants welcome: HP-UX? IRIX?)

	  Thanks to H.Merijn Brand, John Peacock, and Craig
	  Berry for brave beta testing.

1.91	[2006-09-29]
	- ualarm() in SuSE 10.1 was overflowing after ~4.2 seconds,
	  possibly due to a glibc bug/feature (suspected overflow at
	  2**32 microseconds?), workaround by using the setitimer()
	  implementation of ualarm() if either useconds or
	  interval > 999_999 (this case seems to vary between systems:
	  are useconds more than 999_999 for ualarm() defined or not)
	  Added more ualarm() tests to catch various overflow points,
	  hopefully no problems in various platforms.
	  (The problem report by Mark Seger and Jon Paul Sullivan of HP.)

1.90	[2006-08-22]
	- tweak still needed for Const64(), from Jerry Hedden
	- get a freshly generated ppport.h
	- update Copyright years

1.89	[2006-08-22]
	- Const64() already appends an 'LL' (or i64), so provide LL and i64
	  forms for the IV_1E[679] (effects Win32 and Cygwin), reported by
	  Jerry Hedden.
	- the Changes entry for 1.88 talked about [IN]V_1[679],
	  missing the 'E'.

1.88	[2006-08-21]
	- clean up the g++ warnings in HiRes.xs, all of them
	  about mixing integer and floating point, introduce
	  constants IV_1E[679] and NV_1E[679]

1.87	[2006-02-13]
	- [rt.cpan.org #17442] 'make test' frequently fails under
	  Cygwin Perl v5.8.8, reported and patched by J. R. Hedden
	  (two race condition bugs in the END block in the case the
	   main process dies before the timer process, unearthed
	   by a bug in Cygwin ualarm)

1.86	[2005-12-17]
	- HiRes.t:s/ok 32/ok 33/, from Dominic Dunlop
	- tighten up the clock() test marginally by requiring non-negative
	- clock_nanosleep() and clock() doc tweaks

1.85	[2005-12-16]
	- the interface to clock_nanosleep() is more natural
	  when it is like (hires) time() (instead of like nanosleep),
	  and the .xs implementation of clock_nanosleep() in 1.84
	  was broken anyway
	- the semantics of clock() are not quite so silly as I thought,
	  but still somewhat odd, documented as such
	- additional enhancements to the clock() documentation
	- add test for clock_nanosleep() (I cannot test this
	  since none of my systems have the function)
	- add test for clock()

1.84	[2005-12-16]
	- add clock() which returns the processor time in
	  (floating point) seconds since an arbitrary era
	- add clock_nanosleep() which suspends the current
	  thread until either absolute time or for relative time
	- [rt.cpan.org #16486] printf missing value in HiRes.t
	- add constants CLOCKS_PER_SEC, CLOCK_SOFTTIME, TIMER_ABSTIME
	- tiny typo fixes

1.83	[2005-11-19]
	- has_symbol() was wrong since e.g. ITIMER_VIRTUAL is exported
	  via @EXPORT_OK even when it is not available.  This is heinous.
	  @EXPORT_OK should be determined at Makefile.PL time.
	- be more lenient is testing clock_gettime(): allow more slop,
	  and retry up to three times, sleeping a random nap between
	  the retries
	- human months are one-based (noticed by Anton Berezin)

1.82	[2005-10-06]
	- CLOCK_REALTIME is an enum value (of the clockid_t enum)
	  in HP-UX (and might be so elsewhere, too), debugged by
	  H. Merijn Brand
	- include const-c.inc as late as possible (from Randy Kobes,
	  [rt.cpan.org #15552] to avoid undefined usleep() on Win32

1.81	[2005-11-05]
	- try to be more robust and consistent in the detection of
          CLOCK_REALTIME and ITIMER_VIRTUAL in HiRes.t: the proper
	  way is

		sub has_symbol {
		    my $symbol = shift;
		    eval 'import Time::HiRes qw($symbol)';
		    return 0 unless $@ eq '';
		    return exists ${"Time::HiRes::$symbol"};
		}

	  and then use

		&FOO_BAR

	  in the test.  All these moves are needed because

	  1) one cannot directly do eval 'Time::HiRes::FOO_BAR'
	     because FOO_BAR might have a true value of zero
	     (or in the general case an empty string or even undef)

	  2) In case FOO_BAR is not available in this platform,
	     &FOO_BAR avoids the bareword warning

	- wait more (1.5 seconds instead of 0.1) for the CLOCK_REALTIME test
	  but expect the 'customary' slop of 0.20 instead of 0.25
	- fixed inside a comment HAS_POLL -> TIME_HIRES_NANOSLEEP
	- at the end of HiRest.t tell how close we were to termination

1.80	[2005-11-04]
	- Gisle noticed a mistake (using HAS_NANOSLEEP) in 1.79

1.79	[2005-11-03]
	- try nanosleep for emulating usleep -- may help in some weird
	  embedded realtime places which have nanosleep but neither usleep
	  nor select nor poll (doesn't have to be weird embedded realtime
	  place, though -- in many places usleep is nanosleep anyway)
	- try poll for emulating usleep -- this may help some obscure/old
	  SVR4 places that have neither usleep nor select
	- a redundant test guard in HiRes.t

1.78	[2005-11-03]
	- ITIMER_VIRTUAL detection in HiRes.t had problems (that we cannot
	  in the general case fail already at 'use' phase is suboptimal)
	- fixes to the documentation of clock_gettime() and clock_getres()

1.77	[2005-11-03]
	- add support for the POSIX clock_gettime() and clock_getres(),
	  if available, either as library calls or as syscalls
	- be more defensive about missing functionality: break out
	  early (during 'use') if no e.g. clock_getres() is available,