Re: [Last-Call] Artart last call review of draft-murchison-rfc8536bis-12

[Paul Eggert <[e-mail]>]

On 2024-03-24 03:02, Eliot Lear wrote:
>>
> For version 4 entries, it's possible to make new rules.  So, you could 
> instead require a POSIX string for this purpose, but only for version 
> 4.  Otherwise, I'd leave the text alone.

For context, in the default build, TZDB currently generates 345 version 
2 files, 7 version 3 files, and no version 4 files.

Version 4 is unpopular because of the recommendation in section 4 bullet 
3, which in the latest draft says "Other than version 1, writers SHOULD 
generate the lowest version number needed by a file's data." This 
supports historical practice in which new TZif versions have never taken 
away old functionality, and where 'zic' outputs the lowest 
version-numbered format that's good enough to represent the data, 
because this practice is more compatible for older TZif readers.

It'd be a bit of a stretch to depart from this historical practice, and 
to make a newer TZif version take away functionality of older versions, 
without a lot of discussion and compatibility testing. So it sounds like 
we should leave that part of the text alone.

That being said, it might make sense to change the reference TZDB 
implementation to sanitize poorly-encoded TZif strings, and thus follow 
the practice that I suggested. Perhaps we could use experience with that 
effort to improve a later iteration of this RFC.