mirror of
https://github.com/overcuriousity/autopsy-flatpak.git
synced 2025-07-06 21:00:22 +00:00
remove unused code, update comments
This commit is contained in:
parent
5bcbe9a56f
commit
8bf50478af
@ -160,25 +160,13 @@ public class TermComponentQuery implements KeywordSearchQuery {
|
||||
|
||||
//get unique match result files
|
||||
|
||||
//execute per term Solr query to get files
|
||||
//more costly
|
||||
/*
|
||||
Set<FsContent> uniqueMatches = new HashSet<FsContent>();
|
||||
for (Term term : terms) {
|
||||
String word = term.getTerm();
|
||||
LuceneQuery filesQuery = new LuceneQuery(word);
|
||||
filesQuery.escape();
|
||||
List<FsContent> matches = filesQuery.performQuery();
|
||||
uniqueMatches.addAll(matches);
|
||||
}*/
|
||||
|
||||
|
||||
//combine the terms into single Solr query to get files
|
||||
//it's much more efficient and should yield the same file IDs as separate queries
|
||||
//TODO limited by GET length limit, try POST ?
|
||||
//it's much more efficient and should yield the same file IDs as per match queries
|
||||
//requires http POST query method due to potentially large query size
|
||||
StringBuilder filesQueryB = new StringBuilder();
|
||||
for (Term term : terms) {
|
||||
final String termS = KeywordSearchUtil.escapeLuceneQuery(term.getTerm());
|
||||
final String termS = term.getTerm();
|
||||
filesQueryB.append(termS);
|
||||
filesQueryB.append(" ");
|
||||
}
|
||||
|
Loading…
x
Reference in New Issue
Block a user