Deborah Goldsmith <goldsmit@apple.com> writes:
We can't use the output files -- which *are* stable
No, actually they're not. Their format just got changed in a big way, to support 64-bit time_t transitions. It's an upward-compatible change, but it's a big one nonetheless, and all binary file readers should upgrade well before 2038.
The ICU project has had a difficult time keeping up with changes to the file and tools.
Hmmm, what changes caused these difficulties, exactly? Here are all the changes to the tz textual format made in the last 20 years that I know about. Perhaps I'm missing something, but if so I'd like to know. * Removal of double leap seconds in 1994 (these were an error). * Removal of the "uspres" feature in 1994 (a never-used feature in real tz data). * Support for the "u" suffix (for universal time) added in 1994. * Support for the "STD/DST" notation, added in 1995. * The clock-advance/clock-retreat fixup of 1996. * Support for "24:00", added in 1998. (Not yet used by tz data.) * Support for out-of-month-range transitions like "Sun<=1" and "Sun>=31", added in 2004. (Not yet used by tz data.) Given this list, I am skeptical whether switching to an XML-based format would improve stability. (If anything, I suspect the reverse would happen.) There are some practical advantages to supporting an XML-based format, but I don't see stability as being one of them.