> Why not fix it completely with this stuff: > ftp://elsie.nci.nih.gov/pub/ > Just an idea.
Ah, the real zic implementation. afaik this is public domain or BSD or at least compatible with our BSD license wrt distribution. Great idea. We may end up doing this! Though I hate for the project to pick up the task of maintaining sync with that distro. We already have a NO_MKTIME_BEFORE_1970 #define'd for AIX and IRIX (always paragons of standard behavior :/ Funny enough it doesn't actually guard the mktime() code, since I think that there is a good signature from the exit from mktime() on those systems (independent of the return value) to detect that there is a problem. glibc is sinking to new depths in lack of support for this feature by brute force exiting early on. It looks like we might (easily?) get good behavior beyond y2038, since we might be able to redefine time_t within our code. At the moment zic looks for it from sys/types.h, but maybe we could isolate it and force it to be a 64-bit number on systems which support it. Hmm, need to find how to translate current system time to that representation... - Thomas ---------------------------(end of broadcast)--------------------------- TIP 4: Don't 'kill -9' the postmaster