Issues

1. ULocale#getFallback() never get to ULocale.ROOT (ticket#6673), instead, the final locale is empty locale (new ULocale("")), then null.
• Question from Markus: Should any of the fallbacks ever get to null? Should they not stop at the root locale?
2. ULocale.getFallback(String) never get to "root", the final fallback is "".
3. ULocale.getFallback(String) may return a locale string ending empty segment.  For example, "en__POSIX" -> "en_" -> "en" -> ""

Design Questions

1. What is the canonical representation of root locale?
1. Three possible options - "" (empty string), "root", or "und" (undetermined)
2. JDK 1.6 added Locale.ROOT using empty language - new Locale("", "", "").
3. {Yoshito} I prefer "" (empty string) for several reasons
1. Logical (no special handling)
2. Same with Java
3. However, backward compatibility concern - can we change ULocale.ROOT from ULocale("root") to ULocale("") now?
2. Normalization
1. What should be done in the locale constructor?
2. What is the expected behavior of ULocale.canonicalize(String) ?
3. {Yoshito} canonicalize should normalize casing and following mappings
   
1. Deprecated ICU locales
1. fr_FR_PREEURO -> fr_FR@currency=FRF
2. hi__DIRECT -> hr@collation=direct
3. other variants mapped to keywords
   
2. Grandfathered BCP 47 tags
1. art_LOJBAN -> jbo
2. zh_HAKKA / zh__HAKKA -> hak
3. other BCP47 grandfathered tag - preferred mapping
3. POSIX
1. C -> en_US_POSIX
4. .NET names
1. az_AZ_CYRL -> az_Cyrl_AZ
2. zh_CHS -> zh_Hans
5. Common mistakes
1. three-letter language codes (eng) that have two letter equivalents
2. three-letter region codes (xxx) that have two letter equivalents
3. three-digit codes (813) that have two letter equivalents
4. swapping script and region code (see also .NET names above)
   
6. Deprecated codes
1. iw -> he
2. some others
   

Proposed Changes

1. ULocale.ROOT
1. current: new ULocale("root");
2. proposed:new ULocale("");
2. ULocale#getFallback()
1. current: ULocale("en__POSIX") -> ULocale("en_") -> ULocale("en") -> ULocale("") -> null
2. proposed: ULocale("en__POSIX") -> ULocale("en") -> ULocale("") -> null
3. ULocale.getFallback(String)
1. current: "en__POSIX" -> "en_" -> "en" -> "" -> ""
2. proposed: "en_POSIX" -> "en" -> "" -> null?
   

Conclusions

• ULocale.ROOT.toString() == "", not "root"
• BCP47
  
• ULocale.ROOT to BCP47 "und"
• Locale.ROOT to BCP47 "und"
• BCP47 "und" to ULocale ""
• getFallback() chops off from nominal form not from canonical form, never leaves trailing underscore, just works on '_'-separated strings.
  
• class ULocale {
    ...
    static ULocale getCanonicalInstance(String);   // Factory
    ULocale getCanonicalEquivalent();  // Uses cached internal pointer
    ...
  };
• Canonicalize "und" to "".  ULocale("und-DE") will have
• lang = "und", region = "DE"
  
• canonical lang = "", canonical region = "DE"
• Resource bundles
• root.res
• de.res
• de-DE.res
• root-DE.res
• en -> "" -> null?
• Yes in Java ULocale#getFallback() - e.g. ULocale("en") -> ULocale("") -> null
  
• No in Java ULocale#getFallback(String) - e.g. "en" -> "" -> ""
  
• No in C++
• Document well.