Proposed 64-bit changes

Garrett Wollman wollman at csail.mit.edu
Wed May 4 00:05:28 UTC 2005


<<On Wed, 04 May 2005 06:34:22 +0700, Robert Elz <kre at munnari.oz.au> said:

> And that's even if putting system type information into environment
> variables (as distinct from user overrides of system information)
> wasn't fundamentally the wrong approach anyway.

This is an important point.  Where the System V TZ string, as extended
by POSIX, really went off the rails is when implementations decided
that they didn't need any way to represent the *installation*
timezone.  A portable program written for the only-TZ-in-environment
paradigm must use UTC for all security-sensitive times; with
TZ-in-the-filesystem, it can be certain of at least one secure local
timezone in addition to UTC, if the host is correctly configured.

(Perhaps it would be good if the next POSIX revision included some
words in the rationale to this effect, suggesting that implementations
not rely exclusively on the TZ environment variable.)

-GAWollman




More information about the tz mailing list