CollationKey for AlphanumericCollator always does codepoint collation
At some point the alphanumeric collator (which, for example, sorts "Chap2" before "Chap10") was enhanced so that the non-numeric parts could be compared using some base collation, rather than always being compared using codepoint collation. However, AlphanumericCollator.getCollationKey() returns a collation key that assumes codepoint collation for the non-numeric parts.
Updated by Michael Kay about 3 years ago
I have created a new test in xslt30extra, collation-007, to test this; there was indeed a bug and it is now fixed.
Along the way, I found that collation URIs of the form
are partially implemented but not actually working; it's a simple bug, but since the feature is undocumented and untested and is in any case superseded by proper UCA collations, I'm going to fix the bug on the 9.9 branch without documenting the feature, and on 10.0 I shall withdraw it.
The collation used in the test is
which should give the same collation key for "chap001" and "CHAP1"
Please register to edit this issue