• Regex using abstract text access APIs (UText), roll in work by Jordan Rose: #4521 (ensure perf OK, ensure UTF8 support)
• Ensure that there are APIs providing access to all CLDR data: e.g. #4836, #5478, etc. (Google also interested) (Peter has CLDR task to enumerate the data that is missing; based on that we can file additional bugs and divide up the work)
• Improved search capabilities (Peter to generate design doc) - mainly asymmetric search, i.e. type e, match e,é,è; type é, match é but probably not e and certainly not è (#7093) (Google also interested). Other possibilities (lower priority) include:
• Position-dependent matching? (e.g. Arabic HEH and TEH MARBUTA  should match for a search when both are at the end of a word)
• Use of search object distinct from collator? (Possible optimization, may not be necessary, not of interest to others)
• Reduce ICU4C dynamically-allocated memory, especially for time zone data (more compact data formats may help with this): #6873, #6879 (Google also interested) (Peter will look at porting Yoshito's ICU4J work to C; requires interpreting const in a "logical" way - can do lazy loading, just make sure thread safe. Should document this interpretation. Peter to coordinate with Andy on this)

• Number spellout format & parse support for CJK numbers, including in dates. Note: CLDR 1.7 added relevant capability per cldrbug:1927; is there anything else that needs to be done in ICU (may work if appropriate patterns are used, Peter will do some experiments)

• Support >2GB text length for search, regex, text break, encoding conversion, perhaps transliteration. Use of UText will provide appropriate interfaces for regex and RBBI with additional internal changes. #5451 is for the RBBI changes.

• Encoding detection for a wider range of encodings, with some finer distinctions. For example pure ShiftJIS text should return both ShiftJIS and cp932 with 100% confidence; text including cp932 extensions should also return both but with lower confidence for ShiftJIS.
• Additional conversion tables (not necessarily in default build). Don't need a ticket for this yet.

Already implemented on branch

Google

Tier 1

• General
• Spoof detection in ICU4J (#7094) (IBM has also expressed interest)
  
• Complete BCP47 Support (6177, 6916, 6791, 6860)
• Switch data formats to use UTrie2— requires UTrie2 in ICU4J as well (#7077)
• (apple) Provide APIs to access all CLDR data  #4836, #5478
• Fully customizable data for objects, formatters, and others; there are holes
• Normalization/IDNA
• Support custom normalization data (#7273)
• Support UTS #46 (#7144) [via custom normalization data]
  
• Formatting
• Able to select numbering systems (#7160)
  
• BigDecimal format/parse in ICU4C  (#5193)
• Footprint
• Split locale display names, time zone names, currency names out of locale data (into separate data) (#7163)
  
• Introduce Locale base class — only id, no display names (#7164)
• Introduce MessageFormat base class — only string substitution (#7165)
• Smaller "core" set (without formatting data, only manipulation/algorithms) (#7161)
  
• Generalized cache management for ICU4C (#2863, #3035, #3118, #6029, #6030, #6031, #6099, #6708)
• More compact collation (rules/binary) data
• Compact collation tailoring syntax for lists of characters with same level difference (#7015)
• Add import rule to collation tailoring syntax (#7023, CldrBug:2268)
• Locale data filtering: display names for fewer codes
• (apple) Reduce C heap memory usage, especially for time zone #6873, #6879

Tier 2

• General
• Collation script reordering (#3984, CldrBug:2267)
• Best match for locale IDs: #4712
• Better C++ implementation (scoped_ptr, byte string class; see design/C++ page) (#7162)
• Cast reduction — move methods into base
• Formatting
• Day divisions — more than just am/pm, e.g. Chinese 'noon', 'midnight' (#7150, CLDR#713)(Apple also interested in this; Peter has a CLDR bug)
• More kinds of duration formatting (Apple also interested in this; Peter has a CLDR bug)
• Footprint
• Filter out locales with insufficient data

IBM

• General focus: Usability, Maintainability and Performance
• Code and Data Maintainability Improvements, e.g. Separating timezone data from code.
• Overriding/updating locale information in an ICU installation: 4597 6633
• Collation and string search service code clean up: 4562
• Misc layout bug fixes: 5589 6625 5431 6113 6182
• Improved ICU performance and regression for selected service areas only, e.g. Collation
• Extended IETF BCP47 support: language/locale specification for HTTP/XML/OpenJDK
• Lenient parsing, e.g. DateFormat. (Already implemented by Apple on branch)
• Locale service SPI
• JSR-310 Date and Time APIs
• @provider multiple version support, Calling old ICU service code through new ICU API
• Java 5 migration (ICU4J)
• Supporting generics to match JDK APIs
• ICU 4.4 will no longer support Java 1.4 or older versions
• Java Logging support (ICU4J)
• ICU Resource Bundle footprint optimization