navidrome/persistence/sql_search_fts_test.go
Deluan Quintão ae9b8a5fe6
feat(search): rank exact matches above prefix matches (#5704)
* feat(search): boost exact token matches over prefix matches

buildFTS5Query now emits (word OR word*) instead of word* for plain tokens. The match set is unchanged (exact is a subset of prefix), but bm25 gives the rare exact token a high-IDF contribution, so rows containing the literal query word rank above prefix-only matches. The degraded-query check keeps evaluating the plain prefix form, preserving the LIKE fallback for queries like "1+" and "C++".

* feat(search): weight artist search_normalized equal to name in bm25

For the artist table, search_normalized holds only the artist's name in alternate spelling (transliterated/punctuation-stripped), so a hit there is as meaningful as a name hit. Combined with exact-token boosting, artists like MØ now rank in the top results for the query "MO" instead of dead last. media_file and album keep weight 1.0 because their search_normalized mixes title, album, and artist variants.

* test(persistence): add exact-match ranking regression test

Seeds MØ, Morrissey, and Modest Mouse and asserts MØ ranks first for the queries "MO" and "MØ": the exact transliterated hit in search_normalized must outrank name-prefix matches. The corpus deliberately has no competing exact-word names, since exact-vs-exact ordering depends on corpus statistics rather than the guaranteed exact-over-prefix property. Rows are inserted per-test (with their library_artist associations) and cleaned up to avoid disturbing the shared seed fixtures and their count assertions.

* docs(search): document exact-token OR emission in buildFTS5Query

* test(persistence): harden exact-match ranking test fixtures

Register the corpus cleanup before the insert loop so a mid-loop assertion failure cannot leak fts-rank-% rows into the shared integration DB, and reuse the existing createArtistWithLibrary helper instead of hand-rolling Put+AddArtist (which also replaces the ad-hoc context.TODO with the helper's GinkgoT().Context).

* fix(search): flag multi-word degraded queries for the LIKE fallback

The degradation probe was joined with explicit " AND " like the real query, so ftsQueryDegraded counted the literal AND as a long token and never flagged queries where every term degrades to a short token (e.g. "1+ 2+"). This predates this branch (the old code passed the same AND-joined string), but the probe now exists separately, so join it with spaces — it only feeds ftsQueryDegraded, which needs no explicit operators.
2026-07-02 15:51:03 -04:00

478 lines
20 KiB
Go

package persistence
import (
"context"
"github.com/navidrome/navidrome/log"
"github.com/navidrome/navidrome/model"
"github.com/navidrome/navidrome/model/request"
. "github.com/onsi/ginkgo/v2"
. "github.com/onsi/gomega"
)
var _ = DescribeTable("buildFTS5Query",
func(input, expected string) {
q, _ := buildFTS5Query(input)
Expect(q).To(Equal(expected))
},
Entry("returns empty string for empty input", "", ""),
Entry("returns empty string for whitespace-only input", " ", ""),
Entry("wraps a single word as exact OR prefix", "beatles", "(beatles OR beatles*)"),
Entry("wraps each word as exact OR prefix", "abbey road", "(abbey OR abbey*) AND (road OR road*)"),
Entry("preserves quoted phrases without wrapping", `"the beatles"`, `"the beatles"`),
Entry("does not wrap user-supplied prefix wildcard", "beat*", "beat*"),
Entry("strips FTS5 operators and wraps lowercased words", "AND OR NOT NEAR", "(and OR and*) AND (or OR or*) AND (not OR not*) AND (near OR near*)"),
Entry("strips special FTS5 syntax characters and wraps", "test^col:val", "(test OR test*) AND (col OR col*) AND (val OR val*)"),
Entry("handles mixed phrases and words", `"the beatles" abbey`, `"the beatles" AND (abbey OR abbey*)`),
Entry("handles prefix with multiple words", "beat* abbey", "beat* AND (abbey OR abbey*)"),
Entry("collapses multiple spaces", "abbey road", "(abbey OR abbey*) AND (road OR road*)"),
Entry("strips leading * from tokens and wraps", "*livia", "(livia OR livia*)"),
Entry("strips leading * and preserves existing trailing *", "*livia oliv*", "(livia OR livia*) AND oliv*"),
Entry("strips standalone *", "*", ""),
Entry("strips apostrophe from input", "Guns N' Roses", "(Guns OR Guns*) AND (N OR N*) AND (Roses OR Roses*)"),
Entry("converts slashed word to phrase+concat OR", "AC/DC", `("AC DC" OR ACDC*)`),
Entry("converts hyphenated word to phrase+concat OR", "a-ha", `("a ha" OR aha*)`),
Entry("converts partial hyphenated word to phrase+concat OR", "a-h", `("a h" OR ah*)`),
Entry("converts hyphenated name to phrase+concat OR", "Jay-Z", `("Jay Z" OR JayZ*)`),
Entry("converts contraction to phrase+concat OR", "it's", `("it s" OR its*)`),
Entry("handles punctuated word mixed with plain words", "best of a-ha", `(best OR best*) AND (of OR of*) AND ("a ha" OR aha*)`),
Entry("handles contraction followed by plain words", "you've got", `("you ve" OR youve*) AND (got OR got*)`),
Entry("strips miscellaneous punctuation", "rock & roll, vol. 2", "(rock OR rock*) AND (roll OR roll*) AND (vol OR vol*) AND (2 OR 2*)"),
Entry("transliterates NFKD-decomposable diacritics", "Björk début", "(Bjork OR Bjork*) AND (debut OR debut*)"),
Entry("transliterates ø to o", "Øystein", "(Oystein OR Oystein*)"),
Entry("transliterates œ ligature to oe", "œuvre", "(oeuvre OR oeuvre*)"),
Entry("transliterates æ ligature to ae", "Brennæ", "(Brennae OR Brennae*)"),
Entry("transliterates mixed unicode words", "Mø Sigur Rós", "(Mo OR Mo*) AND (Sigur OR Sigur*) AND (Ros OR Ros*)"),
Entry("transliterates ß to ss", "Straße", "(Strasse OR Strasse*)"),
Entry("preserves quoted unicode phrase verbatim", `"Björk"`, `"Björk"`),
Entry("collapses dotted abbreviation into phrase", "R.E.M.", `"R E M"`),
Entry("collapses abbreviation without trailing dot", "R.E.M", `"R E M"`),
Entry("collapses abbreviation mixed with words", "best of R.E.M.", `(best OR best*) AND (of OR of*) AND "R E M"`),
Entry("collapses two-letter abbreviation", "U.K.", `"U K"`),
Entry("does not collapse single letter surrounded by words", "I am fine", "(I OR I*) AND (am OR am*) AND (fine OR fine*)"),
Entry("does not collapse single standalone letter", "A test", "(A OR A*) AND (test OR test*)"),
Entry("preserves quoted phrase with punctuation verbatim", `"ac/dc"`, `"ac/dc"`),
Entry("preserves quoted abbreviation verbatim", `"R.E.M."`, `"R.E.M."`),
Entry("returns empty string for punctuation-only input", "!!!!!!!", ""),
Entry("returns empty string for mixed punctuation", "!@#$%^&", ""),
Entry("returns empty string for empty quoted phrase", `""`, ""),
)
var _ = DescribeTable("buildFTS5Query degraded flag",
func(input string, expected bool) {
_, degraded := buildFTS5Query(input)
Expect(degraded).To(Equal(expected))
},
Entry("plain words are not degraded", "beatles", false),
Entry("special chars stripped leaving short token is degraded", "1+", true),
Entry("multiple short tokens are degraded", "1+ 2+", true),
Entry("short tokens mixed with a long word are not degraded", "1+ beatles", false),
Entry("quoted short-token phrase is degraded", `"1+"`, true),
Entry("punctuated-name group is not degraded", "AC/DC", false),
Entry("empty input is not degraded", "", false),
)
var _ = DescribeTable("ftsQueryDegraded",
func(original, ftsQuery string, expected bool) {
Expect(ftsQueryDegraded(original, ftsQuery)).To(Equal(expected))
},
Entry("not degraded for empty original", "", "1*", false),
Entry("not degraded for empty ftsQuery", "1+", "", false),
Entry("not degraded for purely alphanumeric query", "beatles", "beatles*", false),
Entry("not degraded when long tokens remain", "test^val", "test* val*", false),
Entry("not degraded for quoted phrase with long tokens", `"the beatles"`, `"the beatles"`, false),
Entry("degraded for quoted phrase with only short tokens after tokenizer strips special chars", `"1+"`, `"1+"`, true),
Entry("not degraded for quoted phrase with meaningful content", `"C++ programming"`, `"C++ programming"`, false),
Entry("degraded when special chars stripped leaving short token", "1+", "1*", true),
Entry("degraded when special chars stripped leaving two short tokens", "C# 1", "C* 1*", true),
Entry("not degraded when at least one long token remains", "1+ beatles", "1* beatles*", false),
Entry("not degraded for OR groups from processPunctuatedWords", "AC/DC", `("AC DC" OR ACDC*)`, false),
)
var _ = DescribeTable("containsCJK",
func(input string, expected bool) {
Expect(containsCJK(input)).To(Equal(expected))
},
Entry("returns false for empty string", "", false),
Entry("returns false for ASCII text", "hello world", false),
Entry("returns false for Latin with diacritics", "Björk début", false),
Entry("detects Chinese characters (Han)", "周杰伦", true),
Entry("detects Japanese Hiragana", "こんにちは", true),
Entry("detects Japanese Katakana", "カタカナ", true),
Entry("detects Korean Hangul", "한국어", true),
Entry("detects CJK mixed with Latin", "best of 周杰伦", true),
Entry("detects single CJK character", "a曲b", true),
)
var _ = DescribeTable("qualifyOrderBy",
func(tableName, orderBy, expected string) {
Expect(qualifyOrderBy(tableName, orderBy)).To(Equal(expected))
},
Entry("returns empty string for empty input", "artist", "", ""),
Entry("qualifies simple column with table name", "artist", "name", "artist.name"),
Entry("qualifies column with direction", "artist", "name desc", "artist.name desc"),
Entry("preserves already-qualified column", "artist", "artist.name", "artist.name"),
Entry("preserves already-qualified column with direction", "artist", "artist.name desc", "artist.name desc"),
Entry("returns empty for function call expression", "artist", "sum(json_extract(stats, '$.total.m')) desc", ""),
Entry("returns empty for expression with comma", "artist", "a, b", ""),
Entry("qualifies media_file column", "media_file", "title", "media_file.title"),
)
var _ = Describe("ftsColumnDefs helpers", func() {
Describe("ftsColumnFilters", func() {
It("returns column filter for media_file", func() {
Expect(ftsColumnFilters).To(HaveKeyWithValue("media_file",
"{title album artist album_artist sort_title sort_album_name sort_artist_name sort_album_artist_name disc_subtitle search_participants search_normalized}",
))
})
It("returns column filter for album", func() {
Expect(ftsColumnFilters).To(HaveKeyWithValue("album",
"{name sort_album_name album_artist search_participants discs catalog_num album_version search_normalized}",
))
})
It("returns column filter for artist", func() {
Expect(ftsColumnFilters).To(HaveKeyWithValue("artist",
"{name sort_artist_name search_normalized}",
))
})
It("has no entry for unknown table", func() {
Expect(ftsColumnFilters).ToNot(HaveKey("unknown"))
})
})
Describe("ftsBM25Weights", func() {
It("returns weight CSV for media_file", func() {
Expect(ftsBM25Weights).To(HaveKeyWithValue("media_file",
"10.0, 5.0, 3.0, 3.0, 1.0, 1.0, 1.0, 1.0, 1.0, 2.0, 1.0",
))
})
It("returns weight CSV for album", func() {
Expect(ftsBM25Weights).To(HaveKeyWithValue("album",
"10.0, 1.0, 3.0, 2.0, 1.0, 1.0, 1.0, 1.0",
))
})
It("returns weight CSV for artist", func() {
Expect(ftsBM25Weights).To(HaveKeyWithValue("artist",
"10.0, 1.0, 10.0",
))
})
It("has no entry for unknown table", func() {
Expect(ftsBM25Weights).ToNot(HaveKey("unknown"))
})
})
It("has definitions for all known tables", func() {
for _, table := range []string{"media_file", "album", "artist"} {
Expect(ftsColumnDefs).To(HaveKey(table))
Expect(ftsColumnDefs[table]).ToNot(BeEmpty())
}
})
It("has matching column count between filter and weights", func() {
for table, cols := range ftsColumnDefs {
// Column filter only includes Weight > 0 columns
filterCount := 0
for _, c := range cols {
if c.Weight > 0 {
filterCount++
}
}
// For now, all columns have Weight > 0, so filter count == total count
Expect(filterCount).To(Equal(len(cols)), "table %s: all columns should have positive weights", table)
}
})
})
var _ = Describe("newFTSSearch", func() {
It("returns nil for empty query", func() {
Expect(newFTSSearch("media_file", "")).To(BeNil())
})
It("returns non-nil for single-character query", func() {
strategy := newFTSSearch("media_file", "a")
Expect(strategy).ToNot(BeNil(), "single-char queries must not be rejected; min-length is enforced in doSearch, not here")
sql, _, err := strategy.ToSql()
Expect(err).ToNot(HaveOccurred())
Expect(sql).To(ContainSubstring("MATCH"))
})
It("returns ftsSearch with correct table names and MATCH expression", func() {
strategy := newFTSSearch("media_file", "beatles")
fts, ok := strategy.(*ftsSearch)
Expect(ok).To(BeTrue())
Expect(fts.tableName).To(Equal("media_file"))
Expect(fts.ftsTable).To(Equal("media_file_fts"))
Expect(fts.matchExpr).To(HavePrefix("{title album artist album_artist"))
Expect(fts.matchExpr).To(ContainSubstring("beatles*"))
})
It("ToSql generates rowid IN subquery with MATCH (fallback path)", func() {
strategy := newFTSSearch("media_file", "beatles")
sql, args, err := strategy.ToSql()
Expect(err).ToNot(HaveOccurred())
Expect(sql).To(ContainSubstring("media_file.rowid IN"))
Expect(sql).To(ContainSubstring("media_file_fts"))
Expect(sql).To(ContainSubstring("MATCH"))
Expect(args).To(HaveLen(1))
})
It("generates correct FTS table name per entity", func() {
for _, table := range []string{"media_file", "album", "artist"} {
strategy := newFTSSearch(table, "test")
fts, ok := strategy.(*ftsSearch)
Expect(ok).To(BeTrue())
Expect(fts.tableName).To(Equal(table))
Expect(fts.ftsTable).To(Equal(table + "_fts"))
}
})
It("builds bm25() rank expression with column weights", func() {
strategy := newFTSSearch("media_file", "beatles")
fts, ok := strategy.(*ftsSearch)
Expect(ok).To(BeTrue())
Expect(fts.rankExpr).To(HavePrefix("bm25(media_file_fts,"))
Expect(fts.rankExpr).To(ContainSubstring("10.0"))
strategy = newFTSSearch("artist", "beatles")
fts, ok = strategy.(*ftsSearch)
Expect(ok).To(BeTrue())
Expect(fts.rankExpr).To(HavePrefix("bm25(artist_fts,"))
})
It("falls back to ftsTable.rank for unknown tables", func() {
strategy := newFTSSearch("unknown_table", "test")
fts, ok := strategy.(*ftsSearch)
Expect(ok).To(BeTrue())
Expect(fts.rankExpr).To(Equal("unknown_table_fts.rank"))
})
It("wraps query with column filter", func() {
strategy := newFTSSearch("artist", "Beatles")
fts, ok := strategy.(*ftsSearch)
Expect(ok).To(BeTrue())
Expect(fts.matchExpr).To(Equal("{name sort_artist_name search_normalized} : ((Beatles OR Beatles*))"))
})
It("passes query without column filter for unknown tables", func() {
strategy := newFTSSearch("unknown_table", "test")
fts, ok := strategy.(*ftsSearch)
Expect(ok).To(BeTrue())
Expect(fts.matchExpr).To(Equal("(test OR test*)"))
})
It("preserves phrase queries inside column filter", func() {
strategy := newFTSSearch("media_file", `"the beatles"`)
fts, ok := strategy.(*ftsSearch)
Expect(ok).To(BeTrue())
Expect(fts.matchExpr).To(ContainSubstring(`"the beatles"`))
})
It("preserves prefix queries inside column filter", func() {
strategy := newFTSSearch("media_file", "beat*")
fts, ok := strategy.(*ftsSearch)
Expect(ok).To(BeTrue())
Expect(fts.matchExpr).To(ContainSubstring("beat*"))
})
It("falls back to LIKE search for punctuation-only query", func() {
strategy := newFTSSearch("media_file", "!!!!!!!")
Expect(strategy).ToNot(BeNil())
_, ok := strategy.(*ftsSearch)
Expect(ok).To(BeFalse(), "punctuation-only should fall back to LIKE, not FTS")
sql, args, err := strategy.ToSql()
Expect(err).ToNot(HaveOccurred())
Expect(sql).To(ContainSubstring("LIKE"))
Expect(args).To(ContainElement("%!!!!!!!%"))
})
It("falls back to LIKE search for degraded query (special chars stripped leaving short tokens)", func() {
strategy := newFTSSearch("album", "1+")
Expect(strategy).ToNot(BeNil())
_, ok := strategy.(*ftsSearch)
Expect(ok).To(BeFalse(), "degraded query should fall back to LIKE, not FTS")
sql, args, err := strategy.ToSql()
Expect(err).ToNot(HaveOccurred())
Expect(sql).To(ContainSubstring("LIKE"))
Expect(args).To(ContainElement("%1+%"))
})
It("returns nil for empty string even with LIKE fallback", func() {
Expect(newFTSSearch("media_file", "")).To(BeNil())
Expect(newFTSSearch("media_file", " ")).To(BeNil())
})
It("returns nil for empty quoted phrase", func() {
Expect(newFTSSearch("media_file", `""`)).To(BeNil())
})
})
var _ = Describe("FTS5 Integration Search", func() {
var (
mr model.MediaFileRepository
alr model.AlbumRepository
arr model.ArtistRepository
)
BeforeEach(func() {
ctx := log.NewContext(context.TODO())
ctx = request.WithUser(ctx, adminUser)
conn := GetDBXBuilder()
mr = NewMediaFileRepository(ctx, conn)
alr = NewAlbumRepository(ctx, conn)
arr = NewArtistRepository(ctx, conn)
})
Describe("MediaFile search", func() {
It("finds media files by title", func() {
results, err := mr.Search("Radioactivity", model.QueryOptions{Max: 10})
Expect(err).ToNot(HaveOccurred())
Expect(results).To(HaveLen(1))
Expect(results[0].Title).To(Equal("Radioactivity"))
Expect(results[0].ID).To(Equal(songRadioactivity.ID))
})
It("finds media files by artist name", func() {
results, err := mr.Search("Beatles", model.QueryOptions{Max: 10})
Expect(err).ToNot(HaveOccurred())
Expect(results).To(HaveLen(3))
for _, r := range results {
Expect(r.Artist).To(Equal("The Beatles"))
}
})
})
Describe("Album search", func() {
It("finds albums by name", func() {
results, err := alr.Search("Sgt Peppers", model.QueryOptions{Max: 10})
Expect(err).ToNot(HaveOccurred())
Expect(results).To(HaveLen(1))
Expect(results[0].Name).To(Equal("Sgt Peppers"))
Expect(results[0].ID).To(Equal(albumSgtPeppers.ID))
})
It("finds albums with multi-word search", func() {
results, err := alr.Search("Abbey Road", model.QueryOptions{Max: 10})
Expect(err).ToNot(HaveOccurred())
Expect(results).To(HaveLen(2))
})
})
Describe("Artist search", func() {
It("finds artists by name", func() {
results, err := arr.Search("Kraftwerk", model.QueryOptions{Max: 10})
Expect(err).ToNot(HaveOccurred())
Expect(results).To(HaveLen(1))
Expect(results[0].Name).To(Equal("Kraftwerk"))
Expect(results[0].ID).To(Equal(artistKraftwerk.ID))
})
})
Describe("CJK search", func() {
It("finds media files by CJK title", func() {
results, err := mr.Search("プラチナ", model.QueryOptions{Max: 10})
Expect(err).ToNot(HaveOccurred())
Expect(results).To(HaveLen(1))
Expect(results[0].Title).To(Equal("プラチナ・ジェット"))
Expect(results[0].ID).To(Equal(songCJK.ID))
})
It("finds media files by CJK artist name", func() {
results, err := mr.Search("シートベルツ", model.QueryOptions{Max: 10})
Expect(err).ToNot(HaveOccurred())
Expect(results).To(HaveLen(1))
Expect(results[0].Artist).To(Equal("シートベルツ"))
})
It("finds albums by CJK artist name", func() {
results, err := alr.Search("シートベルツ", model.QueryOptions{Max: 10})
Expect(err).ToNot(HaveOccurred())
Expect(results).To(HaveLen(1))
Expect(results[0].Name).To(Equal("COWBOY BEBOP"))
Expect(results[0].ID).To(Equal(albumCJK.ID))
})
It("finds artists by CJK name", func() {
results, err := arr.Search("シートベルツ", model.QueryOptions{Max: 10})
Expect(err).ToNot(HaveOccurred())
Expect(results).To(HaveLen(1))
Expect(results[0].Name).To(Equal("シートベルツ"))
Expect(results[0].ID).To(Equal(artistCJK.ID))
})
})
Describe("Album version search", func() {
It("finds albums by version tag via FTS", func() {
results, err := alr.Search("Deluxe", model.QueryOptions{Max: 10})
Expect(err).ToNot(HaveOccurred())
Expect(results).To(HaveLen(1))
Expect(results[0].ID).To(Equal(albumWithVersion.ID))
})
})
Describe("Punctuation-only search", func() {
It("finds media files with punctuation-only title", func() {
results, err := mr.Search("!!!!!!!", model.QueryOptions{Max: 10})
Expect(err).ToNot(HaveOccurred())
Expect(results).To(HaveLen(1))
Expect(results[0].Title).To(Equal("!!!!!!!"))
Expect(results[0].ID).To(Equal(songPunctuation.ID))
})
})
Describe("Single-character search (doSearch min-length guard)", func() {
It("returns empty results for single-char query via Search", func() {
results, err := mr.Search("a", model.QueryOptions{Max: 10})
Expect(err).ToNot(HaveOccurred())
Expect(results).To(BeEmpty(), "doSearch should reject single-char queries")
})
})
Describe("Max=0 means no limit (regression: must not produce LIMIT 0)", func() {
It("returns results with Max=0", func() {
results, err := mr.Search("Beatles", model.QueryOptions{Max: 0})
Expect(err).ToNot(HaveOccurred())
Expect(results).ToNot(BeEmpty(), "Max=0 should mean no limit, not LIMIT 0")
})
})
Describe("Exact-match ranking", func() {
BeforeEach(func() {
// Registered before the inserts so a mid-loop failure cannot leak corpus rows.
DeferCleanup(func() {
// library_artist rows are removed by the artist_id ON DELETE CASCADE FK.
_, err := GetDBXBuilder().NewQuery("DELETE FROM artist WHERE id LIKE 'fts-rank-%'").Execute()
Expect(err).ToNot(HaveOccurred())
})
// Corpus has no competing exact-word names ("Mo X"): exact-vs-exact order depends
// on corpus statistics; the guaranteed property is exact > prefix.
for _, a := range []model.Artist{
{ID: "fts-rank-1", Name: "MØ", OrderArtistName: "mø"},
{ID: "fts-rank-2", Name: "Modest Mouse", OrderArtistName: "modest mouse"},
{ID: "fts-rank-3", Name: "Morrissey", OrderArtistName: "morrissey"},
} {
Expect(createArtistWithLibrary(arr, &a, 1)).To(Succeed())
}
})
It("ranks the exact transliterated match first for 'MO'", func() {
results, err := arr.Search("MO", model.QueryOptions{Max: 10})
Expect(err).ToNot(HaveOccurred())
Expect(results).To(HaveLen(3))
Expect(results[0].Name).To(Equal("MØ"), "exact match via search_normalized must outrank prefix matches")
})
It("ranks the exact match first for the accented query 'MØ'", func() {
results, err := arr.Search("MØ", model.QueryOptions{Max: 10})
Expect(err).ToNot(HaveOccurred())
Expect(results).ToNot(BeEmpty())
Expect(results[0].Name).To(Equal("MØ"))
})
})
})