fork of https://github.com/sourcegraph/zoekt
ranking: favor short paths and multiple matches (#417)
This changes our scoring logic slightly to put more emphasis on shorter filenames and to recognize multiple matches.
author
Stefan Hengl
committer
GitHub
date
(Aug 31, 2022, 9:28 AM +0200)
commit
543ddf98
543ddf9885606d80d38d013455e539081e867cee
parent
6a91f0b7
6a91f0b753c92b3b9cee9cbc90071ef0957d7f6d
+34
-5
+10
-5
build/builder.go
+10
-5
build/builder.go
···
38
38
"time"
39
39
40
40
"github.com/bmatcuk/doublestar"
41
-
"github.com/sourcegraph/zoekt"
42
-
"github.com/sourcegraph/zoekt/ctags"
43
41
"github.com/grafana/regexp"
44
42
"github.com/rs/xid"
45
43
"gopkg.in/natefinch/lumberjack.v2"
44
+
45
+
"github.com/sourcegraph/zoekt"
46
+
"github.com/sourcegraph/zoekt/ctags"
46
47
)
47
48
48
49
var DefaultDir = filepath.Join(os.Getenv("HOME"), ".zoekt")
···
870
871
rank []float64
871
872
}
872
873
874
+
// rank returns a vector of scores which is used at index-time to sort documents
875
+
// before writing them to disk. The order of documents in the shard is important
876
+
// at query time, because earlier documents receive a boost at query time and
877
+
// have a higher chance of being searched before limits kick in.
873
878
func rank(d *zoekt.Document, origIdx int) []float64 {
874
879
generated := 0.0
875
880
if strings.HasSuffix(d.Name, "min.js") || strings.HasSuffix(d.Name, "js.map") {
···
897
902
// Prefer docs that are not tests
898
903
test,
899
904
905
+
// With short names
906
+
squashRange(len(d.Name)),
907
+
900
908
// With many symbols
901
909
1.0 - squashRange(len(d.Symbols)),
902
910
903
911
// With short content
904
912
squashRange(len(d.Content)),
905
-
906
-
// With short names
907
-
squashRange(len(d.Name)),
908
913
909
914
// That is present is as many branches as possible
910
915
1.0 - squashRange(len(d.Branches)),
+15
build/e2e_test.go
+15
build/e2e_test.go
···
844
844
// 500 (word) + 400 (atom) + 10 (file order)
845
845
wantScore: 910,
846
846
},
847
+
{
848
+
fileName: "src/net/http/client.go",
849
+
content: []byte(`
850
+
package http
851
+
func Get() {
852
+
panic("")
853
+
}
854
+
`),
855
+
query: &query.And{Children: []query.Q{
856
+
&query.Symbol{Expr: &query.Substring{Pattern: "http", Content: true}},
857
+
&query.Symbol{Expr: &query.Substring{Pattern: "Get", Content: true}}}},
858
+
wantLanguage: "Go",
859
+
// 7000 (full base match) + 500 (word) + 400 (atom) + 10 (file order) + 1 (repetition-boost)
860
+
wantScore: 7911,
861
+
},
847
862
}
848
863
849
864
for _, c := range cases {
+1
contentprovider.go
+1
contentprovider.go
+8
eval.go
+8
eval.go
···
343
343
}
344
344
345
345
maxFileScore := 0.0
346
+
repetitions := 0
346
347
for i := range fileMatch.LineMatches {
347
348
if maxFileScore < fileMatch.LineMatches[i].Score {
348
349
maxFileScore = fileMatch.LineMatches[i].Score
350
+
repetitions = 0
351
+
} else if maxFileScore == fileMatch.LineMatches[i].Score {
352
+
repetitions += 1
349
353
}
350
354
351
355
// Order by ordering in file.
···
365
369
// strictly dominates the in-file ordering of
366
370
// the matches.
367
371
fileMatch.addScore("fragment", maxFileScore, opts.DebugScore)
372
+
373
+
// Prefer docs with several top-scored matches.
374
+
fileMatch.addScore("repetition-boost", scoreRepetitionFactor*float64(repetitions), opts.DebugScore)
375
+
368
376
fileMatch.addScore("atom", float64(atomMatchCount)/float64(totalAtomCount)*scoreFactorAtomMatch, opts.DebugScore)
369
377
370
378
// Prefer earlier docs.