DWDS/Dialing Concordance Code

+ v2.0.21 updated changes

+ updated to v2.0.21

+ fixed mantis bug #652 : context-sort not working for boolean queries (&&, ||)
- fix was to ensure occurrences are sorted in QueryNode.cpp CQueryBinaryOperationNode::hits_and_positions_union(), hits_and_positions_intersection()
- not quite kosher in #separate mode; e.g. for input "a b1 b2 c" querying (a && b* #separate) finds only "__a__ _b1_ b2 c" and
not "__a__ b1 _b2_ c"

v2.0.21-pre1 Fri, 21 Mar 2014 11:08:02 +0100 moocow
+ fixed mantis bug #651 : bogus highlighting for near/2
- problem was overlapping matches some of which crossed sentence boundaries
- the bogus cross-boundary matches were getting mis-interpreted in CQueryNode::ConvertOccurrencesToHitsForPatterns()
- workaround tracks current break start as well as end, but may potentially miss some overlapping occurrences if
(as in the case of near()) m_Occurrences[] aren't sorted in strictly break-increasing order
and (as is hopefully NOT the case for near()) if the valid matches in m_Occurrences[] also arent't sorted in
strictly break-increasing order

+ tagged v2.0.20-0

+ renamed #(asc|dsc)_by_context to #(asc|dsc)_by_middle to more intuitively reflect anchor-positioning tweaks

+ updated to v2.0.20 in preparation for release

+ re-worked near/2 in preparation for real generalization to near/3 (mantis #646)

+ documented DefaultQueryIndex

v2.0.20-pre4 Tue, 18 Mar 2014 12:32:54 +0100 moocow
+ added *.opt options LemmaQueryUsesMorphPattern, DefaultQueryIndex
+ added parsing of boolean values for *.opt options to CConcIndexator::LoadOptionsFromString()
- this means we can finally override pre-specified boolean flags
+ added CConcIndexator::m_OpDefaultIndexNames : operator-specific default index names
- potentially useful to discriminate between _ (bareword TokInfl queries) vs. @_ (literal match) and /_/ (regex, globs)
- idea: map _ to Lemma, set default expander Lemma to some (CAB-based) external best-lemmatizer,
and avoid the traffic, computational load, and loss of both precision and recall associated with
the current all-inflected-forms strategy of CQTokInfl
+ cleaned up some lingering bigrams-index crud