mirror of https://github.com/apache/lucene.git
LUCENE-2002: Fix HighlighterTest to consistently use the same version compat
git-svn-id: https://svn.apache.org/repos/asf/lucene/java/trunk@829208 13f79535-47bb-0310-9956-ffa450edef68
This commit is contained in:
parent
aaddac8992
commit
831f665ba8
|
@ -84,7 +84,7 @@ import org.w3c.dom.NodeList;
|
|||
*/
|
||||
public class HighlighterTest extends BaseTokenStreamTestCase implements Formatter {
|
||||
// TODO: change to CURRENT, does not work because posIncr:
|
||||
static final Version TEST_VERSION = Version.LUCENE_29;
|
||||
static final Version TEST_VERSION = Version.LUCENE_CURRENT;
|
||||
|
||||
private IndexReader reader;
|
||||
static final String FIELD_NAME = "contents";
|
||||
|
@ -113,7 +113,7 @@ public class HighlighterTest extends BaseTokenStreamTestCase implements Formatte
|
|||
|
||||
public void testQueryScorerHits() throws Exception {
|
||||
Analyzer analyzer = new SimpleAnalyzer();
|
||||
QueryParser qp = new QueryParser(Version.LUCENE_CURRENT, FIELD_NAME, analyzer);
|
||||
QueryParser qp = new QueryParser(TEST_VERSION, FIELD_NAME, analyzer);
|
||||
query = qp.parse("\"very long\"");
|
||||
searcher = new IndexSearcher(ramDir, true);
|
||||
TopDocs hits = searcher.search(query, 10);
|
||||
|
@ -143,7 +143,7 @@ public class HighlighterTest extends BaseTokenStreamTestCase implements Formatte
|
|||
|
||||
String s1 = "I call our world Flatland, not because we call it so,";
|
||||
|
||||
QueryParser parser = new QueryParser(Version.LUCENE_CURRENT, FIELD_NAME, new StandardAnalyzer(TEST_VERSION));
|
||||
QueryParser parser = new QueryParser(TEST_VERSION, FIELD_NAME, new StandardAnalyzer(TEST_VERSION));
|
||||
|
||||
// Verify that a query against the default field results in text being
|
||||
// highlighted
|
||||
|
@ -221,7 +221,7 @@ public class HighlighterTest extends BaseTokenStreamTestCase implements Formatte
|
|||
String q = "(" + f1c + ph1 + " OR " + f2c + ph1 + ") AND (" + f1c + ph2
|
||||
+ " OR " + f2c + ph2 + ")";
|
||||
Analyzer analyzer = new WhitespaceAnalyzer();
|
||||
QueryParser qp = new QueryParser(Version.LUCENE_CURRENT, f1, analyzer);
|
||||
QueryParser qp = new QueryParser(TEST_VERSION, f1, analyzer);
|
||||
Query query = qp.parse(q);
|
||||
|
||||
QueryScorer scorer = new QueryScorer(query, f1);
|
||||
|
@ -590,7 +590,7 @@ public class HighlighterTest extends BaseTokenStreamTestCase implements Formatte
|
|||
// Need to explicitly set the QueryParser property to use TermRangeQuery
|
||||
// rather
|
||||
// than RangeFilters
|
||||
QueryParser parser = new QueryParser(Version.LUCENE_CURRENT, FIELD_NAME, analyzer);
|
||||
QueryParser parser = new QueryParser(TEST_VERSION, FIELD_NAME, analyzer);
|
||||
parser.setMultiTermRewriteMethod(MultiTermQuery.SCORING_BOOLEAN_QUERY_REWRITE);
|
||||
query = parser.parse(queryString);
|
||||
doSearching(query);
|
||||
|
@ -930,7 +930,7 @@ public class HighlighterTest extends BaseTokenStreamTestCase implements Formatte
|
|||
String srchkey = "football";
|
||||
|
||||
String s = "football-soccer in the euro 2004 footie competition";
|
||||
QueryParser parser = new QueryParser(Version.LUCENE_CURRENT, "bookid", analyzer);
|
||||
QueryParser parser = new QueryParser(TEST_VERSION, "bookid", analyzer);
|
||||
Query query = parser.parse(srchkey);
|
||||
|
||||
TokenStream tokenStream = analyzer.tokenStream(null, new StringReader(s));
|
||||
|
@ -1111,7 +1111,7 @@ public class HighlighterTest extends BaseTokenStreamTestCase implements Formatte
|
|||
searcher = new IndexSearcher(ramDir, true);
|
||||
Analyzer analyzer = new StandardAnalyzer(TEST_VERSION);
|
||||
|
||||
QueryParser parser = new QueryParser(Version.LUCENE_CURRENT, FIELD_NAME, analyzer);
|
||||
QueryParser parser = new QueryParser(TEST_VERSION, FIELD_NAME, analyzer);
|
||||
Query query = parser.parse("JF? or Kenned*");
|
||||
System.out.println("Searching with primitive query");
|
||||
// forget to set this and...
|
||||
|
@ -1245,7 +1245,7 @@ public class HighlighterTest extends BaseTokenStreamTestCase implements Formatte
|
|||
searchers[0] = new IndexSearcher(ramDir1, true);
|
||||
searchers[1] = new IndexSearcher(ramDir2, true);
|
||||
MultiSearcher multiSearcher = new MultiSearcher(searchers);
|
||||
QueryParser parser = new QueryParser(Version.LUCENE_CURRENT, FIELD_NAME, new StandardAnalyzer(TEST_VERSION));
|
||||
QueryParser parser = new QueryParser(TEST_VERSION, FIELD_NAME, new StandardAnalyzer(TEST_VERSION));
|
||||
parser.setMultiTermRewriteMethod(MultiTermQuery.SCORING_BOOLEAN_QUERY_REWRITE);
|
||||
query = parser.parse("multi*");
|
||||
System.out.println("Searching for: " + query.toString(FIELD_NAME));
|
||||
|
@ -1278,7 +1278,7 @@ public class HighlighterTest extends BaseTokenStreamTestCase implements Formatte
|
|||
|
||||
public void run() throws Exception {
|
||||
String docMainText = "fred is one of the people";
|
||||
QueryParser parser = new QueryParser(Version.LUCENE_CURRENT, FIELD_NAME, analyzer);
|
||||
QueryParser parser = new QueryParser(TEST_VERSION, FIELD_NAME, analyzer);
|
||||
Query query = parser.parse("fred category:people");
|
||||
|
||||
// highlighting respects fieldnames used in query
|
||||
|
@ -1419,64 +1419,64 @@ public class HighlighterTest extends BaseTokenStreamTestCase implements Formatte
|
|||
Highlighter highlighter;
|
||||
String result;
|
||||
|
||||
query = new QueryParser(Version.LUCENE_CURRENT, "text", new WhitespaceAnalyzer()).parse("foo");
|
||||
query = new QueryParser(TEST_VERSION, "text", new WhitespaceAnalyzer()).parse("foo");
|
||||
highlighter = getHighlighter(query, "text", getTS2(), HighlighterTest.this);
|
||||
result = highlighter.getBestFragments(getTS2(), s, 3, "...");
|
||||
assertEquals("Hi-Speed10 <B>foo</B>", result);
|
||||
|
||||
query = new QueryParser(Version.LUCENE_CURRENT, "text", new WhitespaceAnalyzer()).parse("10");
|
||||
query = new QueryParser(TEST_VERSION, "text", new WhitespaceAnalyzer()).parse("10");
|
||||
highlighter = getHighlighter(query, "text", getTS2(), HighlighterTest.this);
|
||||
result = highlighter.getBestFragments(getTS2(), s, 3, "...");
|
||||
assertEquals("Hi-Speed<B>10</B> foo", result);
|
||||
|
||||
query = new QueryParser(Version.LUCENE_CURRENT, "text", new WhitespaceAnalyzer()).parse("hi");
|
||||
query = new QueryParser(TEST_VERSION, "text", new WhitespaceAnalyzer()).parse("hi");
|
||||
highlighter = getHighlighter(query, "text", getTS2(), HighlighterTest.this);
|
||||
result = highlighter.getBestFragments(getTS2(), s, 3, "...");
|
||||
assertEquals("<B>Hi</B>-Speed10 foo", result);
|
||||
|
||||
query = new QueryParser(Version.LUCENE_CURRENT, "text", new WhitespaceAnalyzer()).parse("speed");
|
||||
query = new QueryParser(TEST_VERSION, "text", new WhitespaceAnalyzer()).parse("speed");
|
||||
highlighter = getHighlighter(query, "text", getTS2(), HighlighterTest.this);
|
||||
result = highlighter.getBestFragments(getTS2(), s, 3, "...");
|
||||
assertEquals("Hi-<B>Speed</B>10 foo", result);
|
||||
|
||||
query = new QueryParser(Version.LUCENE_CURRENT, "text", new WhitespaceAnalyzer()).parse("hispeed");
|
||||
query = new QueryParser(TEST_VERSION, "text", new WhitespaceAnalyzer()).parse("hispeed");
|
||||
highlighter = getHighlighter(query, "text", getTS2(), HighlighterTest.this);
|
||||
result = highlighter.getBestFragments(getTS2(), s, 3, "...");
|
||||
assertEquals("<B>Hi-Speed</B>10 foo", result);
|
||||
|
||||
query = new QueryParser(Version.LUCENE_CURRENT, "text", new WhitespaceAnalyzer()).parse("hi speed");
|
||||
query = new QueryParser(TEST_VERSION, "text", new WhitespaceAnalyzer()).parse("hi speed");
|
||||
highlighter = getHighlighter(query, "text", getTS2(), HighlighterTest.this);
|
||||
result = highlighter.getBestFragments(getTS2(), s, 3, "...");
|
||||
assertEquals("<B>Hi-Speed</B>10 foo", result);
|
||||
|
||||
// ///////////////// same tests, just put the bigger overlapping token
|
||||
// first
|
||||
query = new QueryParser(Version.LUCENE_CURRENT, "text", new WhitespaceAnalyzer()).parse("foo");
|
||||
query = new QueryParser(TEST_VERSION, "text", new WhitespaceAnalyzer()).parse("foo");
|
||||
highlighter = getHighlighter(query, "text", getTS2a(), HighlighterTest.this);
|
||||
result = highlighter.getBestFragments(getTS2a(), s, 3, "...");
|
||||
assertEquals("Hi-Speed10 <B>foo</B>", result);
|
||||
|
||||
query = new QueryParser(Version.LUCENE_CURRENT, "text", new WhitespaceAnalyzer()).parse("10");
|
||||
query = new QueryParser(TEST_VERSION, "text", new WhitespaceAnalyzer()).parse("10");
|
||||
highlighter = getHighlighter(query, "text", getTS2a(), HighlighterTest.this);
|
||||
result = highlighter.getBestFragments(getTS2a(), s, 3, "...");
|
||||
assertEquals("Hi-Speed<B>10</B> foo", result);
|
||||
|
||||
query = new QueryParser(Version.LUCENE_CURRENT, "text", new WhitespaceAnalyzer()).parse("hi");
|
||||
query = new QueryParser(TEST_VERSION, "text", new WhitespaceAnalyzer()).parse("hi");
|
||||
highlighter = getHighlighter(query, "text", getTS2a(), HighlighterTest.this);
|
||||
result = highlighter.getBestFragments(getTS2a(), s, 3, "...");
|
||||
assertEquals("<B>Hi</B>-Speed10 foo", result);
|
||||
|
||||
query = new QueryParser(Version.LUCENE_CURRENT, "text", new WhitespaceAnalyzer()).parse("speed");
|
||||
query = new QueryParser(TEST_VERSION, "text", new WhitespaceAnalyzer()).parse("speed");
|
||||
highlighter = getHighlighter(query, "text", getTS2a(), HighlighterTest.this);
|
||||
result = highlighter.getBestFragments(getTS2a(), s, 3, "...");
|
||||
assertEquals("Hi-<B>Speed</B>10 foo", result);
|
||||
|
||||
query = new QueryParser(Version.LUCENE_CURRENT, "text", new WhitespaceAnalyzer()).parse("hispeed");
|
||||
query = new QueryParser(TEST_VERSION, "text", new WhitespaceAnalyzer()).parse("hispeed");
|
||||
highlighter = getHighlighter(query, "text", getTS2a(), HighlighterTest.this);
|
||||
result = highlighter.getBestFragments(getTS2a(), s, 3, "...");
|
||||
assertEquals("<B>Hi-Speed</B>10 foo", result);
|
||||
|
||||
query = new QueryParser(Version.LUCENE_CURRENT, "text", new WhitespaceAnalyzer()).parse("hi speed");
|
||||
query = new QueryParser(TEST_VERSION, "text", new WhitespaceAnalyzer()).parse("hi speed");
|
||||
highlighter = getHighlighter(query, "text", getTS2a(), HighlighterTest.this);
|
||||
result = highlighter.getBestFragments(getTS2a(), s, 3, "...");
|
||||
assertEquals("<B>Hi-Speed</B>10 foo", result);
|
||||
|
@ -1521,7 +1521,7 @@ public class HighlighterTest extends BaseTokenStreamTestCase implements Formatte
|
|||
|
||||
private void searchIndex() throws IOException, ParseException, InvalidTokenOffsetsException {
|
||||
String q = "t_text1:random";
|
||||
QueryParser parser = new QueryParser(Version.LUCENE_CURRENT, "t_text1", a );
|
||||
QueryParser parser = new QueryParser(TEST_VERSION, "t_text1", a );
|
||||
Query query = parser.parse( q );
|
||||
IndexSearcher searcher = new IndexSearcher( dir, true );
|
||||
// This scorer can return negative idf -> null fragment
|
||||
|
@ -1575,7 +1575,7 @@ public class HighlighterTest extends BaseTokenStreamTestCase implements Formatte
|
|||
}
|
||||
|
||||
public void doSearching(String queryString) throws Exception {
|
||||
QueryParser parser = new QueryParser(Version.LUCENE_CURRENT, FIELD_NAME, analyzer);
|
||||
QueryParser parser = new QueryParser(TEST_VERSION, FIELD_NAME, analyzer);
|
||||
parser.setEnablePositionIncrements(true);
|
||||
parser.setMultiTermRewriteMethod(MultiTermQuery.SCORING_BOOLEAN_QUERY_REWRITE);
|
||||
query = parser.parse(queryString);
|
||||
|
|
Loading…
Reference in New Issue