Re: [PATCH] Fix "cvs log" to use UTC timezone instead of local

From: Linus Torvalds <torvalds@linux-foundation.org>
Date: Sep 4, 2007, 13:22 UTC
Message-ID: <alpine.LFD.0.999.0709040612260.3088@evo.linux-foundation.org>
In-Reply-To: <11889090932256-git-send-email-xkr47@outerspace.dyndns.org>

On Tue, 4 Sep 2007, Jonas Berlin wrote:

Show 7 quoted lines

>
> The timestamp format used in "cvs log" output does not include a
> timezone, and must thus be in UTC timezone. The timestamps from git on
> the other hand contain timezone information for each commit timestamp,
> but git-cvsserver discarded this information and used the timestamps
> without adjusting the time accordingly. The patch adds code to apply
> the timezone offset to produce a UTC timestamp.

I think this is wrong.

Git *internally* stores things in UTC anyway, so if there are any local date format things, it's because git-cvsserver.perl has read the dates using some format where git has turned its internal date into a local date.

So instead of turning it back into UTC here, I think git-cvsserver should be changed to ask for the date in the native git format in the first place.

That can be done various ways:

 - use the "raw log format" which has dates as seconds-since-UTC (and with 
   an *informational* timezone thing that should then just be ignored).

   This is likely the best approach, since anything but this will 
   almost invariably result in some potentially broken TZ conversion
   back-and-forth..

 - if it really wants to use the pretty-printing support, git-cvsserver 
   should probably be changed to do something like

	TZ=UTC git rev-list --pretty --date=local

   which will pretty-print the date in local time format rather than in 
   the timezone that the commit was done in, and then the TZ=UTC obviously 
   says that the "local" zone is UTC.

Anything else *will* be broken, or will be converting back-and-forth.

For example, I think your patch may fix "cvs log", but I'm seeing some suspiciously similar code in the "cvs annotate" handling, so I suspect that would need it too.

If instead of trying to convert things to UTC on demand, git-cvsserver just asks for the git date stamps in UTC in the first place, none of the places should ever need any timezone conversion.

		Linus