Re: Weird time-zone oscillation

[David Kubicek <foceni@gmail.com>]

On 12/13/2009 06:47 PM, Kumar Appaiah wrote:
> > When we're sure relevant libraries are OK, we can explore other
> > possibilities if the issue persists. I doubt it will after this action,
> > unless, of course, you have some very peculiar configuration I cannot
> > imagine somewhere. :)
>
> Once I reinstall the above packages, my own program seems to give
> 1260517800 as output consistently. Mysterious.

The POSIX variant result would be important - it would tell us it's not 
Glib related. There would be less stuff to check with just libc/tzdata.

BUT, I made a typo in posix.c, which *could* in effect randomize the 
"zone" part of the input iso date. So, first of all - use this new one, 
recompile and try again. If it'll work fine, then the issue has been 
solved by refreshing those packages. It would mean that some of the 
files form your previous tzdata were corrupted.

However, if the issue remains - read on...

I want to see if there's any difference between Linux tzdata and pure 
POSIX TZ specification and between our tool and the standard date(1) 
command. I need you to run this:

# export TZ=Europe/Prague
# for i in `seq 1 20`; do ./posix; date; done
1260496200
Sun Dec 13 19:04:24 CET 2009
...
# export TZ=CET-1CES,M3.5.0,M10.5.0
# for i in `seq 1 20`; do ./posix; date; done
1260496200
Sun Dec 13 19:04:36 CET 2009
...
# unset TZ

"Europe/Prague" and "CET-1CES,M3.5.0,M10.5.0" is the same description of 
my time zone (inc. daylight savings). One points to a tzdata file, the 
other has all specs. *in itself*.

> I'm dumbfounded. But I'd REALLY like to get to the bottom of this!
> Could you please help me explore the other possibilities which could
> have lead to this?

Me too, I've never seen this.

From your description the issue seems random. 10 commands executed 
almost at the same time return different times. There must be something 
screwing your time zone data, resp. data files. That would fit with your 
chroot tests - my new test with two different TZ setting methods would 
reveal it (one is using tzdata's files, the other does not).

When I look at the results, I'll know where to move next. I'll probably 
let you run it again in a single-user mode (e.g. adding kernel param 
"init=/bin/sh" via lilo/grub boot menu), to eliminate stuff like ntpd 
and other possibly mischievous processes.

> Thank you very much for the pointers.

Wait until we'll have fixed it, if ever. :)

--
David Kubicek

#include <stdio.h>
#include <time.h>

void main(void) {
	char *iso_date = "2009-12-11T2:50:00";
	struct tm tm;
	time_t time;

	strptime(iso_date, "%FT%T", &tm);
	time = mktime(&tm);
	printf("%ld\n", time);
}