From: Anders Kaseorg <andersk@mit.edu>
Date: Mon, 30 May 2011 21:35:55 GMT
Subject: Re: [PATCH 3/3] gitk: Allow displaying time zones from author and commit timestamps
Message-ID: <alpine.DEB.2.02.1105301717520.23145@dr-wily.mit.edu>
In-Reply-To: <201105300829.52619.jnareb@gmail.com>

On Mon, 30 May 2011, Jakub Narebski wrote:
> > > Why not use UTC+N timezone […]?
> 
> I think full specification is UTC+HH:MM or UTC-HH:MM.  GNU date
> understands 'TZ=UTC+09:30 date'.

No, that’s not what the letter code in TZ means.  TZ=EDT+04:00 means that
  • The current time zone is called EDT.
  • The current time zone is at offset -04:00 from UTC.  (Note the sign 
    reversal.)
EDT is displayed as the name of the zone and is not used in any 
calculations; it does not mean we’re taking an offset _from_ EDT.

So it doesn’t make sense to write TZ=UTC±NN:NN for any NN:NN other than 
00:00.  Otherwise libc will happily give you the nonsense that you asked 
for:
  $ TZ=UTC+00:00 date; TZ=UTC+09:30 date
  Mon May 30 21:21:32 UTC 2011
  Mon May 30 11:51:32 UTC 2011

This is why I used “Unknown” instead of “UTC”.

(Oh, my comment should have said “Unknown±NN:NN” instead of 
“Unknown±NNNN”; the code gets it right.)

If I wanted to do better, I’d need a way to translate UTC offsets to zone 
names, but I don’t know of a canonical way to do that (there’s not even a 
unique answer in general).

In Tcl ≥ 8.5, none of this matters because ‘clock format -timezone’ just 
uses the numerical UTC offset as the zone name.  For the fallback code, I 
can’t do that by manipulating TZ, because the zone name needs to be made 
of letters.

Anders