Nicolas Williams wrote at 02/24/09 11:20:
> On Tue, Feb 24, 2009 at 11:02:11AM -0800, Ienup Sung wrote:
>> Nicolas Williams wrote at 02/24/09 08:30:
>>> So everything Volatile. If the actual interfaces really are so
>>> volatile, might it not make sense to go for multiple versions being
>>> installed, each being Uncommitted?
>> Trouble is that with Uncommitted as you know incompatible change can happen
>> only at minor release including EOF. Furthermore, it is possible with ICU
>
> That's why I suggested multiple versions.
FWIW, we are already doing so, i.e., we already deliver ICU 2.1 and 3.2
together and now trying 4.0.1.
>> that their patch level release (i.e., ICU 4.0 vs. ICU 4.0.1 and not Sun's
>> Patch release binding) occupies the same name space in terms of versioning
>> and yet could contain incompatibility.
>
> That would be a problem, but does that actually happen? Do they really
> break compatbility on patch? I haven't followed the ICU community, but
> I have a hard time believing that.
>
> Can you point to specific instances where they've done this?
Yes, between ICU 3.8 and 3.8.1, date formatting and parsing has changed in
ways that may require re-coding of user programs. Please see Known Issues,
Bugs and Patches section of the following:
http://icu-project.org/download/3.8.html#ICU4C
Ienup
