Re: a bug in date concerning daylight saving time

[John Cowan]

Lao DanTong scripsit:

> I found what is likely to be a bug in date. It does not recognise as valid 
> the date when clock jumps one hour due to DST. here in Brazil we use this 
> rule:
> 
> Rule    Brazil  2007    max     -       Oct     Sun>=8   0:00   1:00    S
> Rule    Brazil  2008    max     -       Feb     Sun>=15  0:00   0       -
> 
> so DST will start next October, 14. date says that --date=2007-10-14 is 
> not a valid date:
> 
> $ ./date --version
> date (GNU coreutils) 6.9
> ...
> $ ./date --date=2007-10-14
> ./date: invalid date `2007-10-14'

That looks exactly right to me.  2007-10-14 means 2007-10-14T00:00:00, but
that second does not exist in any timezone observing the "Brazil" rules:
time jumps directly from 2007-10-13T23:59:59 to 2007-10-14T01:00:00.
Internal evidence shows that the timezone in your examples is
America/Sao_Paulo, but this rule set also applies to America/Campo_Grande
and America/Cuiaba, the only other two Brazilian time zones currently
observing DST.

Maybe you want to argue that dates without times should mean the first
second of the day rather than 00:00:00.

> but no problem if time is in UTC:
> 
> $ ./date --date=2007-10-14 -u
> Sun Oct 14 00:00:00 UTC 2007

Naturally, since 2007-10-14T00:00:00 of course exists in UTC.

-- 
On the Semantic Web, it's too hard to prove     John Cowan    address@hidden
you're not a dog.  --Bill de hOra               http://www.ccil.org/~cowan