Re: Cygwin with clock_gettime and CLOCK_MONOTONIC - gives always 0

[Tobias Burnus <burnus at net-b dot de>]

Corinna Vinschen wrote:
> On Apr 11 21:13, Tobias Burnus wrote:
> > Where is actually the source code of the clock_gettime, which Cygwin
> > uses? I thought that it was newlib. But looking at
> > http://sourceware.org/cgi-bin/cvsweb.cgi/src/newlib/libc/sys/linux/clock_gettime.c?cvsroot=src
> > the code should return -1 for "clock_gettime (CLOCK_MONOTONIC" as
> > CLOCK_MONOTONIC (= 4) is not handled, only CLOCK_REALTIME (= 1,
> > according to src/newlib/libc/include/time.h).
> >
> > However, as both you and Angelo get a zero return value, I must look
> > at the wrong file.
> Yes, for Cygwin it's
> http://cygwin.com/cgi-bin/cvsweb.cgi/src/winsup/cygwin/timer.cc?cvsroot=src

Well, it's "times.cc" not "timer.cc". I think the problem with Cygwin 
was a red herring. But there are a bunch of problems:


a) Angelo's code to initialize the random seed assumes that Fortran's 
"system_clock" and the "clock_gettime(CLOCK_MONOTONIC" it calls give a 
result where 0 is the beginning of the epoch (like on Linux/glibc) or 
the system start up (as Windows' GetTickCount/GetTickCount64). However, 
QueryPerformanceCounter (probably intrinsically, but at least as used by 
Cygwin) returns some time based on the startup time - and as the 
frequency seems to be slow such that the result is 0 or (nearly) the 
same number for the startup. Thus, the random_seed initialization 
doesn't work as expected on Cygwin/Windows.

b) Newlib is broken. For clock_gettime, CLOCK_MONOTONIC is not 
implemented. But both CLOCK_MONOTONIC and even _POSIX_MONOTONIC_CLOCK 
(with value 200112L) are defined. However, POSIX states: "If the 
Monotonic Clock option is supported, all implementations shall support a 
clock_id of CLOCK_MONOTONIC defined in <time.h>."


Thus:

* Newlib should be fixed to not define CLOCK_MONOTONIC and 
_POSIX_MONOTONIC_CLOCK - or it should implement it, otherwise it doesn't 
follow POSIX.

* Angelo shouldn't use system_clock to set the random seed.

* libgfortran should fall back to gf_gettime (= gettimeofday or 
clock_gettime CLOCK_REALTIME or time) if clock_gettime returns nonzero 
to work around the broken netlib (and possibly other implementations of 
clock_gettime)

* gfortran's example for random_see should be change to not use 
system_clock for the random seed.


Sorry for misinterpreting the results of Cygwin. And thanks for the 
reply and testing.

Tobias

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple