The man page, the SVID, and the behaviour of libc5 all
had the timezone global containing the offset from GMT
of the standard timezone. (e.g. EST).

I have found that with RH5.1 (libc 2.0.7) that the value
of timezone depends, e.g. on the last call to localtime, and
the value is the offset from GMT of the time last reported.
If daylight savings time is in effect for this converted
time, then timezone is the DST timezone, not the standard
timezone.

For my locale (GB), calling localtime with a time in
December, gives a timezone value of 0, which is correct.
But calling localtime with a time in April, gives a timezone
value of -3600, which is not correct.