The tests in misc-UCACollation carry no dependency information other than a dependency on XP31/XQ31. The test results assume that UCA collation is available.

However:

(a) if fallback=no is specified, then the implementation has the option of failing with error FOCH0002

(b) if fallback is omitted or has the value yes, then the implementation must accept the collation URI but can (in effect) sort in any way it likes (this is the case that applies to most of the tests in this test set)

(c) if fallback is present with a value other than yes or no, the outcome is entirely implementation-defined.

I propose that

(a) we change most of the tests to use fallback=no and to have an alternative result of FOCH0002

(b) where we use fallback=yes, we either (i) allow the result to be in any order, or (ii) add a dependency "uca-close-fallback" with the meaning "when fallback=yes is specified in the UCA collation URI, the implementation falls back to a collation that gives the same result as the UCA collation in the common case where: the UCA collation uses only the parameters lang, strength=1|2|3, and caseFirst; and the strings to be sorted consist entirely of alphabetic characters from the character set normally used for writing the selected language".

(c) for fallback=unknown (one test) we allow either a failure result, or a success result consisting of any permutation of the input.