[tz] [PATCH] Phase out invented abbrs for ex-Soviet Union

[Paul.Koning at dell.com]

> On Sep 28, 2016, at 8:41 PM, Paul Eggert <eggert at CS.UCLA.EDU> wrote:
> 
> On 09/28/2016 04:25 PM, Derick Rethans wrote:
>> this change is now breaking builds of PHP, which tested for the correct
>> representation of timezone abbreviations
> 
> I'm not surprised to see changes in the tz data affect regression tests like that. That's OK. Regression tests are designed to find changes, and when they find changes they're doing their job. The same thing will happen for sufficiently-extensive regression tests on Turkish time stamps in the next release, since we'll be changing Turkish time stamps in the next release too. Perhaps this will affect the PHP code as well; if so, the PHP developers will know how to deal with it.
> 
>> stability is important
> 
> Sure, but it's not more important than accuracy.

This brings to mind a question.  Is there a specification of what the tz data is, and what aspects of it should be considered stable vs. what aspects are subject to change?

It seems to me that there may have been an implied assumption on the part of the PHP team that abbreviations are stable, and a (possibly unstated) assumption i the TZ team that they are not.  If this were documented as part of a specification, it would be clear that abbreviations are not stable, and a regression test that assumes so has a bug.  It would also allow people to see clearly what can change, and perhaps have discussions about whether certain things should be stable instead of changeable.

	paul