krickert commented on PR #1138:
URL: https://github.com/apache/opennlp/pull/1138#issuecomment-4939269880

   Thanks, all six points were on target. Addressed in 345d3c40:
   
   - Distribution attribution: CaseFolding.txt added to the opennlp-distr 
readme LICENSE enumeration and NOTICE bullets. The root files already had it; 
the distr copies were missed.
   - caseFold + fullCaseFold in TextNormalizer.Builder: documented rather than 
enforced. That builder composes rungs freely and polices no other combination 
(nfc().nfkc() is equally accepted), so the guard stays in the opinionated 
TermAnalyzer layer. The javadoc now states the combination is redundant and 
intentionally unenforced there, the rung coverage test no longer models it, and 
a new test pins the combination as byte-identical to fullCaseFold() alone.
   - NFC precondition in buildAligned(): stated in both the rung javadoc and 
fullCaseFold(). Feed the aligned pipeline composed text or decomposed sequences 
pass through unfolded; a preceding nfc() rung is rejected by design since NFC 
reports no offsets.
   - readResolve(): agreed, deferred. Every sibling normalizer singleton has 
the same gap, so this goes into the package-wide post-merge hardening batch 
instead of diverging one class here.
   - Redundant requireNotBothCaseFolds in caseFold(Locale): dropped; 
transform() enforces it. The existing fullCaseFold-then-locale-caseFold test 
still passes, pinning no behavior change.
   - transform() guard coverage: tests added for both directions through the 
transform(...) entry point.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to