RE: Problems within epicsTime.cpp

["Denison, PN (Peter)" <peter.denison@diamond.ac.uk>]

> From: Andrew Johnson [mailto:anj@aps.anl.gov] 
> Sent: 31 January 2008 14:44
> 
>  > On Jan 31, 2008, at 05:22 , Denison, PN (Peter) asked:
>  >>
>  >> Can I replace this code in epicsTimeLoadTimeInit with a constant?
> 
> Kay-Uwe Kasemir wrote:
> > 
> > Yes, make it a constant.
> 
> No, don't.  Apply the attached patch instead, which is a diff between 
> R3.14.9 and the current head of the R3.14 branch and should 
> work in all timezones.
> 
> The Posix spec apparently does not require that a time_t be 
> expressed in seconds, although Unix historically has used that.
> The existing (broken) code does not make that assumption, and
> neither does my patched version (which I did tell Peter about
> several months ago), but a using constant probably does.

I did look at CVS HEAD and this patch before I sent my e-mail.
Unfortunately the patch won't work, ONLY IN THE UK, with the strictly
valid "Europe/London" timezone, because the DST offsets don't cancel
out. This is because on 1/1/1970, the UK was on permanent daylight
savings time, and on 1/1/1990 it was not.

The problem is that mktime converts a local time, and we want the
distance between 2 epochs referenced from UTC.

Can I do it by leaving the existing code intact, but setting TZ to UTC,
calling tzset() and then restoring the previous state of TZ?

-- 
Peter Denison, Senior Software Engineer, Diamond Light Source Ltd.
Tel: +44 1235 778511
(apologies in advance for the lines below. Some bits are a legal
requirement and I have no control over them)