[ILUG] Update on the Gnu date problem

[Niall O Broin]

On 10 Apr 2007, at 01:28, paul at clubi.ie wrote:

> On Mon, 9 Apr 2007, Niall O Broin wrote:
>
>> Inspired by the fact that I had one server where I did NOT see the  
>> log rotation problem, I've now tested this on a variety of systems  
>> using
>>
>> date --version|head -1;date -d "3 weeks ago" +%H; date +%H
>
>> IMO the two numbers returned after the version string should be  
>> the same.
>
> Why do you think that?

Because I feel that 3 weeks ago does NOT always mean 3*7*24 hours  
ago, the exception being where there has been an intervening DST  
transition.


> The hour number might work out the same for 2*7*24 hours ago, but  
> between then and 3*7*24 hours ago, an hour was 'squeezed' out..  
> Further, you're leaving stuff out. I.e. try:
>
> $ date --version|head -1;date -d "3 weeks ago" +"%H %Z" ; date "+%H  
> %Z"
> date (GNU coreutils) 5.97
> 00 GMT
> 01 IST
>
>> for TZ, or with TZ=CET, I see the anomalous behaviour with the  
>> 5.9x versions.
>
> You mean you saw the anamalous behaviour with versions /prior/ to  
> 5.9x.

That's your view (and presumably the Gnu date maintainers). I wonder  
is there some standards document which defines this behaviour?

The particularly hairy answer is where you request   date -d "3 weeks  
ago 08:00" where the date returned will vary by 24 hours, depending  
on what time of the day you run date.

Bottom line of course is that if you want to do reliable date  
arithmetic or manipulation you use UTC. This is something I'm well  
used to from working in an industry where that was standard, but when  
you're producing information for the great unwashed, they tend to  
like it to be related to their timezone IME.


Niall