forked from flaxsearch/luwak
-
Notifications
You must be signed in to change notification settings - Fork 0
/
CHANGES.txt
113 lines (86 loc) · 4.11 KB
/
CHANGES.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
Luwak 1.5.0
===========
This release of luwak requires Java 8
API changes:
* Upgrade to lucene 6.5
* Monitor.update() is declared as throwing an UpdateException (#106)
Optimizations:
* Disable bulk scoring, reducing fixed overhead for large boolean disjunctions
* Query decomposer will recurse into DisjunctionMaxQuery (#112)
New features:
* HighlightingMatcher can handle exact phrase queries
* HighlightingMatcher reports matches even if it can't actually highlight a
query
* Several code cleanups contributed by Bloomberg (#100, #102, #107, #118, #119)
* Expose hits map in HighlightsMatch (#92)
* All lucene core queries are now handled by TermFilteredPresearcher
* SpanRewriter can handle ConstantScoreQuery (#114)
* SpanRewriter will attempt to rewrite unknown queries. This allows luwak to
highlight queries produced by, for example, the ComplexPhraseQueryParser
(#136)
Bug fixes:
* MonitorQuery metadata checks for null values
* InputDocument propagates position and offset gaps for multivalued fields
(#117)
* SpanRewriter.rewriteBoolean() propagates minShouldMatch (#123)
* HighlightingMatcher was only returning hits for the last doc in a batch (#134)
* Unpositioned span terms were being incorrectly highlighted due to two-phase
iteration (fixed by LUCENE-7628)
* InputDocument was not respecting omitNorms values on fields (fixed by
LUCENE-7696) (#135)
Luwak 1.4.0
===========
API changes:
* Upgrade to lucene 5.4.1
Optimizations:
* Refactored the internal query index code into its own class, for easier
testing
* Avoid re-analyzing queries that have already been added
New features:
* Added better toString() representations of QueryMatch objects
* Several code cleanups contributed by DevFactory
Bug fixes:
* Several fixes to the HighlightingMatcher span-rewrite code
* Luwak-specific queries correctly implement equals() and hashCode(), fixing
some query-cache interaction bugs.
* HighlightsMatch.Hit now implements hashCode()
* QueryTreeAnalyzer uses a LinkedHashSet to enforce deterministic iteration
order over its children
Luwak 1.3.0
===========
The major change here is the introduction of the batching API, allowing users to
match several documents at once. In general this will improve throughput at the
expense of latency, although as always you should measure timings carefully.
API changes:
* Added Monitor.match(DocumentBatch, MatcherFactory) and
Monitor.debug(DocumentBatch, MatcherFactory) in addition to the existing
InputDocument methods.
* QueryMatch now takes a document ID as well as a query ID
* Matches<T> exposes DocumentMatches<T> objects for document in an input batch
* Presearcher.buildQuery() now takes a QueryTermFilter rather than an
IndexReaderContext for filtering out terms, and a LeafReader over a batch's
document index rather than an InputDocument
* Various queryindex configuration options (commit batch size, cache purge
frequency, whether or not to store queries, which QueryDecomposer to use) are
now encapsulated in a QueryIndexConfiguration object, passed to the Monitor
constructor.
* Monitor state updates can now be watched by registering
QueryIndexUpdateListeners. The beforeCommit() and afterCommit() methods are
removed.
* Monitor.getStats() is renamed to Monitor.getQueryCacheStats()
New features:
* You can now configure Similarity implementations on a DocumentBatch
* Query metadata is passed to CandidateMatcher.doMatchQuery()
* DisjunctionMaxQuery can be highlighted
* DisjunctionMaxQuery can be decomposed and indexed separately
* Performance improvements: use TermsQuery for the document disjunction, and
filter out non-existent query terms before they're added to the query
* ConcurrentQueryLoader allows more efficient query loading on startup
* WildcardNGramPresearcherComponent makes it easier to adjust its max token
length
* You can choose not to store MonitorQueries in the Monitor for some memory
savings, by setting QueryIndexConfiguration.storeQueries(false).
Bug fixes:
* The slowlog was reporting nanosecond values as milliseconds
* FieldFilterPresearcherComponent didn't work for binary terms
* FieldFilterPresearcherComponent was breaking debug()