ICU development team meeting
Time: 10am-11am PDT, 1pm-2pm EDT, 19:00-20:00 CET
Date: 2001-jun-27

Attendees:
    Yves Arrouye (RealNames)
    Helena Shih Chapman (IBM)
    Mark Davis (IBM)
    Andy Heninger (IBM)
    George Rhoten (IBM)
    Markus Scherer (IBM)
    Tex Texin (Progress)
    Raghuram Viswanadha (IBM)
    Syn Wee Quek (IBM)
    Vladimir Weinstein (IBM)

Minutes taker: Markus Scherer

Agenda items:
- NUL termination of output strings in the C API
  + Markus: want to repropose to keep current behavior of C wrappers,
    make all other APIs consistent with this
  + always attempt to append a NUL
  + no error code if only NUL does not fit
  + user needs to allocate length+1 if they need the NUL
  + may need to check length<capacity to make sure that it fit if NUL is necessary
  + one reason to change original proposal:
    many APIs involved do not have srcLength, so consistent behavior
    cannot depend on that

- versioning of collators
  + Mark: there is a problem with versioning:
    versioning tailoring may result in wrong version if fallback chain changed
    from last collator use
  + new proposal: get a BLOB from a collator that stores its entire state
    and allows to later instantiate a collator with the very same behavior
  + Tex: this is an issue for database indexing
    given the described problem, we will want to use this BLOB
  + Tex: databases are rarely re-indexed because it is an expensive operation
  + Andy: How far back do we need to keep old versions of collation data?
    Tex/Helena: users need some 5 to 10 years of support
    may or may not want to update a collation implementation during that time
  + It will be possible to get information about a BLOB by instantiating
    a collator from it and getting its attributes
- Discussion of sort key length
  + worst case may be some 10x of average for a given string length
  + Tex: databases may need maximum or may have variable-length keys
  + Mark: at least when you truncate sort keys
    you truncate information that is less important
  + Tex: better to provide APIs to work with sort keys than to document
    their structure