On 4 October 2014 18:08, Paul Eggert <eggert@cs.ucla.edu> wrote:
I wasn't asking about Harare, where the issue was the name. I was asking about Jamaica, where the issue was the data. Did Joda-Time's regression tests fail for 2014h's changes to Jamaica time in early 1974? To some extent I'm asking whether these regression tests focus on the data or on the metadata.
As others have said, it's always been the intent in the tz database that users are unaffected by whether a name is defined via a Zone or a Link, which means a regression test sensitive to this issue could well be a false alarm.
In general, Joda-Time does not test the tzdb data, it just expects it to be right and exist. The test failed because something that used to exist (Harare) ceased to exist (because aliases are normalised). Jamaica made no difference, and in fact most tzdb changes would not. The problem is around the assumption that IDs would continue to exist and not be converted to aliases. As Mark noted, in effect there has been a policy change, from the occasional use of aliases for very minor ISO territories, to a use of aliases whenever and wherever it suits, with the other knock on effects we've discussed before. Stephen