internal/lsp/fuzzy: add fuzzy matching library
This change uses a fuzzy matching library to score completion results.
Updates golang/go#32754
Change-Id: Ia7771b33534de393a865443e05c0fcbf1e9a969b
Reviewed-on: https://go-review.googlesource.com/c/tools/+/184441
Run-TryBot: Rebecca Stambler <rstambler@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Cottrell <iancottrell@google.com>
diff --git a/internal/lsp/completion.go b/internal/lsp/completion.go
index affa8da..ecd3968 100644
--- a/internal/lsp/completion.go
+++ b/internal/lsp/completion.go
@@ -30,7 +30,7 @@
if err != nil {
return nil, err
}
- items, surrounding, err := source.Completion(ctx, view, f, rng.Start, source.CompletionOptions{
+ candidates, surrounding, err := source.Completion(ctx, view, f, rng.Start, source.CompletionOptions{
DeepComplete: s.useDeepCompletions,
})
if err != nil {
@@ -58,7 +58,7 @@
}
return &protocol.CompletionList{
IsIncomplete: false,
- Items: toProtocolCompletionItems(items, prefix, insertionRng, s.insertTextFormat, s.usePlaceholders, s.useDeepCompletions),
+ Items: toProtocolCompletionItems(candidates, prefix, insertionRng, s.insertTextFormat, s.usePlaceholders, s.useDeepCompletions),
}, nil
}
diff --git a/internal/lsp/fuzzy/input.go b/internal/lsp/fuzzy/input.go
new file mode 100644
index 0000000..b0e0ab2
--- /dev/null
+++ b/internal/lsp/fuzzy/input.go
@@ -0,0 +1,185 @@
+// Copyright 2019 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package fuzzy
+
+import (
+ "unicode"
+)
+
+// Input specifies the type of the input. This influences how the runes are interpreted wrt to
+// segmenting the input.
+type Input int
+
+const (
+ // Text represents a text input type. Input is not segmented.
+ Text Input = iota
+ // Filename represents a filepath input type with '/' segment delimitors.
+ Filename
+ // Symbol represents a symbol input type with '.' and ':' segment delimitors.
+ Symbol
+)
+
+// RuneRole specifies the role of a rune in the context of an input.
+type RuneRole byte
+
+const (
+ // RNone specifies a rune without any role in the input (i.e., whitespace/non-ASCII).
+ RNone RuneRole = iota
+ // RSep specifies a rune with the role of segment separator.
+ RSep
+ // RTail specifies a rune which is a lower-case tail in a word in the input.
+ RTail
+ // RUCTail specifies a rune which is an upper-case tail in a word in the input.
+ RUCTail
+ // RHead specifies a rune which is the first character in a word in the input.
+ RHead
+)
+
+// RuneRoles detects the roles of each byte rune in an input string and stores it in the output
+// slice. The rune role depends on the input type. Stops when it parsed all the runes in the string
+// or when it filled the output. If output is nil, then it gets created.
+func RuneRoles(str string, input Input, reuse []RuneRole) []RuneRole {
+ var output []RuneRole
+ if cap(reuse) < len(str) {
+ output = make([]RuneRole, 0, len(str))
+ } else {
+ output = reuse[:0]
+ }
+
+ prev, prev2 := rtNone, rtNone
+ for i := 0; i < len(str); i++ {
+ r := rune(str[i])
+
+ role := RNone
+
+ curr := rtLower
+ if str[i] <= unicode.MaxASCII {
+ curr = runeType(rt[str[i]] - '0')
+ }
+
+ if curr == rtLower {
+ if prev == rtNone || prev == rtPunct {
+ role = RHead
+ } else {
+ role = RTail
+ }
+ } else if curr == rtUpper {
+ role = RHead
+
+ if prev == rtUpper {
+ // This and previous characters are both upper case.
+
+ if i+1 == len(str) {
+ // This is last character, previous was also uppercase -> this is UCTail
+ // i.e., (current char is C): aBC / BC / ABC
+ role = RUCTail
+ }
+ }
+ } else if curr == rtPunct {
+ switch {
+ case input == Filename && r == '/':
+ role = RSep
+ case input == Symbol && r == '.':
+ role = RSep
+ case input == Symbol && r == ':':
+ role = RSep
+ }
+ }
+ if curr != rtLower {
+ if i > 1 && output[i-1] == RHead && prev2 == rtUpper && (output[i-2] == RHead || output[i-2] == RUCTail) {
+ // The previous two characters were uppercase. The current one is not a lower case, so the
+ // previous one can't be a HEAD. Make it a UCTail.
+ // i.e., (last char is current char - B must be a UCTail): ABC / ZABC / AB.
+ output[i-1] = RUCTail
+ }
+ }
+
+ output = append(output, role)
+ prev2 = prev
+ prev = curr
+ }
+ return output
+}
+
+type runeType byte
+
+const (
+ rtNone runeType = iota
+ rtPunct
+ rtLower
+ rtUpper
+)
+
+const rt = "00000000000000000000000000000000000000000000001122222222221000000333333333333333333333333330000002222222222222222222222222200000"
+
+// LastSegment returns the substring representing the last segment from the input, where each
+// byte has an associated RuneRole in the roles slice. This makes sense only for inputs of Symbol
+// or Filename type.
+func LastSegment(input string, roles []RuneRole) string {
+ // Exclude ending separators.
+ end := len(input) - 1
+ for end >= 0 && roles[end] == RSep {
+ end--
+ }
+ if end < 0 {
+ return ""
+ }
+
+ start := end - 1
+ for start >= 0 && roles[start] != RSep {
+ start--
+ }
+
+ return input[start+1 : end+1]
+}
+
+// ToLower transforms the input string to lower case, which is stored in the output byte slice.
+// The lower casing considers only ASCII values - non ASCII values are left unmodified.
+// Stops when parsed all input or when it filled the output slice. If output is nil, then it gets
+// created.
+func ToLower(input string, reuse []byte) []byte {
+ output := reuse
+ if cap(reuse) < len(input) {
+ output = make([]byte, len(input))
+ }
+
+ for i := 0; i < len(input); i++ {
+ r := rune(input[i])
+ if r <= unicode.MaxASCII {
+ if 'A' <= r && r <= 'Z' {
+ r += 'a' - 'A'
+ }
+ }
+ output[i] = byte(r)
+ }
+ return output[:len(input)]
+}
+
+// WordConsumer defines a consumer for a word delimited by the [start,end) byte offsets in an input
+// (start is inclusive, end is exclusive).
+type WordConsumer func(start, end int)
+
+// Words find word delimiters in an input based on its bytes' mappings to rune roles. The offset
+// delimiters for each word are fed to the provided consumer function.
+func Words(roles []RuneRole, consume WordConsumer) {
+ var wordStart int
+ for i, r := range roles {
+ switch r {
+ case RUCTail, RTail:
+ case RHead, RNone, RSep:
+ if i != wordStart {
+ consume(wordStart, i)
+ }
+ wordStart = i
+ if r != RHead {
+ // Skip this character.
+ wordStart = i + 1
+ }
+ }
+ }
+ if wordStart != len(roles) {
+ consume(wordStart, len(roles))
+ }
+}
diff --git a/internal/lsp/fuzzy/input_test.go b/internal/lsp/fuzzy/input_test.go
new file mode 100644
index 0000000..351ac9b
--- /dev/null
+++ b/internal/lsp/fuzzy/input_test.go
@@ -0,0 +1,186 @@
+// Copyright 2019 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package fuzzy_test
+
+import (
+ "bytes"
+ "sort"
+ "testing"
+
+ "golang.org/x/tools/internal/lsp/fuzzy"
+)
+
+var rolesTests = []struct {
+ str string
+ input fuzzy.Input
+ want string
+}{
+ {str: "abc", want: "Ccc", input: fuzzy.Text},
+ {str: ".abc", want: " Ccc", input: fuzzy.Text},
+ {str: "abc def", want: "Ccc Ccc", input: fuzzy.Text},
+ {str: "SWT MyID", want: "Cuu CcCu", input: fuzzy.Text},
+ {str: "ID", want: "Cu", input: fuzzy.Text},
+ {str: "IDD", want: "Cuu", input: fuzzy.Text},
+ {str: " ID ", want: " Cu ", input: fuzzy.Text},
+ {str: "IDSome", want: "CuCccc", input: fuzzy.Text},
+ {str: "0123456789", want: "Cccccccccc", input: fuzzy.Text},
+ {str: "abcdefghigklmnopqrstuvwxyz", want: "Cccccccccccccccccccccccccc", input: fuzzy.Text},
+ {str: "ABCDEFGHIGKLMNOPQRSTUVWXYZ", want: "Cuuuuuuuuuuuuuuuuuuuuuuuuu", input: fuzzy.Text},
+ {str: "こんにちは", want: "Ccccccccccccccc", input: fuzzy.Text}, // We don't parse unicode
+ {str: ":/.", want: " ", input: fuzzy.Text},
+
+ // Filenames
+ {str: "abc/def", want: "Ccc/Ccc", input: fuzzy.Filename},
+ {str: " abc_def", want: " Ccc Ccc", input: fuzzy.Filename},
+ {str: " abc_DDf", want: " Ccc CCc", input: fuzzy.Filename},
+ {str: ":.", want: " ", input: fuzzy.Filename},
+
+ // Symbols
+ {str: "abc::def::goo", want: "Ccc//Ccc//Ccc", input: fuzzy.Symbol},
+ {str: "proto::Message", want: "Ccccc//Ccccccc", input: fuzzy.Symbol},
+ {str: "AbstractSWTFactory", want: "CcccccccCuuCcccccc", input: fuzzy.Symbol},
+ {str: "Abs012", want: "Cccccc", input: fuzzy.Symbol},
+ {str: "/", want: " ", input: fuzzy.Symbol},
+ {str: "fOO", want: "CCu", input: fuzzy.Symbol},
+ {str: "fo_oo.o_oo", want: "Cc Cc/C Cc", input: fuzzy.Symbol},
+}
+
+func rolesString(roles []fuzzy.RuneRole) string {
+ var buf bytes.Buffer
+ for _, r := range roles {
+ buf.WriteByte(" /cuC"[int(r)])
+ }
+ return buf.String()
+}
+
+func TestRoles(t *testing.T) {
+ for _, tc := range rolesTests {
+ gotRoles := make([]fuzzy.RuneRole, len(tc.str))
+ fuzzy.RuneRoles(tc.str, tc.input, gotRoles)
+ got := rolesString(gotRoles)
+ if got != tc.want {
+ t.Errorf("roles(%s) = %v; want %v", tc.str, got, tc.want)
+ }
+ }
+}
+
+func words(strWords ...string) [][]byte {
+ var ret [][]byte
+ for _, w := range strWords {
+ ret = append(ret, []byte(w))
+ }
+ return ret
+}
+
+var wordSplitTests = []struct {
+ input string
+ want []string
+}{
+ {
+ input: "foo bar baz",
+ want: []string{"foo", "bar", "baz"},
+ },
+ {
+ input: "fooBarBaz",
+ want: []string{"foo", "Bar", "Baz"},
+ },
+ {
+ input: "FOOBarBAZ",
+ want: []string{"FOO", "Bar", "BAZ"},
+ },
+ {
+ input: "foo123_bar2Baz3",
+ want: []string{"foo123", "bar2", "Baz3"},
+ },
+}
+
+func TestWordSplit(t *testing.T) {
+ for _, tc := range wordSplitTests {
+ roles := fuzzy.RuneRoles(tc.input, fuzzy.Symbol, nil)
+
+ var got []string
+ consumer := func(i, j int) {
+ got = append(got, tc.input[i:j])
+ }
+ fuzzy.Words(roles, consumer)
+
+ if eq := diffStringLists(tc.want, got); !eq {
+ t.Errorf("input %v: (want %v -> got %v)", tc.input, tc.want, got)
+ }
+ }
+}
+
+func diffStringLists(a, b []string) bool {
+ if len(a) != len(b) {
+ return false
+ }
+ sort.Strings(a)
+ sort.Strings(b)
+ for i := range a {
+ if a[i] != b[i] {
+ return false
+ }
+ }
+ return true
+}
+
+var lastSegmentSplitTests = []struct {
+ str string
+ input fuzzy.Input
+ want string
+}{
+ {
+ str: "identifier",
+ input: fuzzy.Symbol,
+ want: "identifier",
+ },
+ {
+ str: "two_words",
+ input: fuzzy.Symbol,
+ want: "two_words",
+ },
+ {
+ str: "first::second",
+ input: fuzzy.Symbol,
+ want: "second",
+ },
+ {
+ str: "foo.bar.FOOBar_buz123_test",
+ input: fuzzy.Symbol,
+ want: "FOOBar_buz123_test",
+ },
+ {
+ str: "golang.org/x/tools/internal/lsp/fuzzy_matcher.go",
+ input: fuzzy.Filename,
+ want: "fuzzy_matcher.go",
+ },
+ {
+ str: "golang.org/x/tools/internal/lsp/fuzzy_matcher.go",
+ input: fuzzy.Text,
+ want: "golang.org/x/tools/internal/lsp/fuzzy_matcher.go",
+ },
+}
+
+func TestLastSegment(t *testing.T) {
+ for _, tc := range lastSegmentSplitTests {
+ roles := fuzzy.RuneRoles(tc.str, tc.input, nil)
+
+ got := fuzzy.LastSegment(tc.str, roles)
+
+ if got != tc.want {
+ t.Errorf("str %v: want %v; got %v", tc.str, tc.want, got)
+ }
+ }
+}
+
+func BenchmarkRoles(b *testing.B) {
+ str := "AbstractSWTFactory"
+ out := make([]fuzzy.RuneRole, len(str))
+
+ for i := 0; i < b.N; i++ {
+ fuzzy.RuneRoles(str, fuzzy.Symbol, out)
+ }
+ b.SetBytes(int64(len(str)))
+}
diff --git a/internal/lsp/fuzzy/matcher.go b/internal/lsp/fuzzy/matcher.go
new file mode 100644
index 0000000..a38a6b4
--- /dev/null
+++ b/internal/lsp/fuzzy/matcher.go
@@ -0,0 +1,437 @@
+// Copyright 2019 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Package fuzzy implements a fuzzy matching algorithm.
+package fuzzy
+
+import (
+ "bytes"
+ "fmt"
+)
+
+const (
+ // MaxInputSize is the maximum size of the input scored against the fuzzy matcher. Longer inputs
+ // will be truncated to this size.
+ MaxInputSize = 127
+ // MaxPatternSize is the maximum size of the pattern used to construct the fuzzy matcher. Longer
+ // inputs are truncated to this size.
+ MaxPatternSize = 63
+)
+
+type scoreVal int
+
+func (s scoreVal) val() int {
+ return int(s) >> 1
+}
+
+func (s scoreVal) prevK() int {
+ return int(s) & 1
+}
+
+func score(val int, prevK int /*0 or 1*/) scoreVal {
+ return scoreVal(val<<1 + prevK)
+}
+
+// Matcher implements a fuzzy matching algorithm for scoring candidates against a pattern.
+// The matcher does not support parallel usage.
+type Matcher struct {
+ input Input
+
+ pattern string
+ patternLower []byte // lower-case version of the pattern
+ patternShort []byte // first characters of the pattern
+ caseSensitive bool // set if the pattern is mix-cased
+
+ patternRoles []RuneRole // the role of each character in the pattern
+ roles []RuneRole // the role of each character in the tested string
+
+ scores [MaxInputSize + 1][MaxPatternSize + 1][2]scoreVal
+
+ scoreScale float32
+
+ lastCandidateLen int // in bytes
+ lastCandidateMatched bool
+
+ // Here we save the last candidate in lower-case. This is basically a byte slice we reuse for
+ // performance reasons, so the slice is not reallocated for every candidate.
+ lowerBuf [MaxInputSize]byte
+ rolesBuf [MaxInputSize]RuneRole
+}
+
+func (m *Matcher) bestK(i, j int) int {
+ if m.scores[i][j][0].val() < m.scores[i][j][1].val() {
+ return 1
+ }
+ return 0
+}
+
+// NewMatcher returns a new fuzzy matcher for scoring candidates against the provided pattern.
+func NewMatcher(pattern string, input Input) *Matcher {
+ if len(pattern) > MaxPatternSize {
+ pattern = pattern[:MaxPatternSize]
+ }
+
+ m := &Matcher{
+ input: input,
+ pattern: pattern,
+ patternLower: ToLower(pattern, nil),
+ }
+
+ for i, c := range m.patternLower {
+ if pattern[i] != c {
+ m.caseSensitive = true
+ break
+ }
+ }
+
+ if len(pattern) > 3 {
+ m.patternShort = m.patternLower[:3]
+ } else {
+ m.patternShort = m.patternLower
+ }
+
+ m.patternRoles = RuneRoles(pattern, input, nil)
+
+ if len(pattern) > 0 {
+ maxCharScore := 4
+ if input == Text {
+ maxCharScore = 6
+ }
+ m.scoreScale = 1 / float32(maxCharScore*len(pattern))
+ }
+
+ return m
+}
+
+// SetInput updates the input type for subsequent scoring attempts.
+func (m *Matcher) SetInput(input Input) {
+ if m.input == input {
+ return
+ }
+ m.input = input
+ m.patternRoles = RuneRoles(m.pattern, input, m.patternRoles)
+}
+
+// Score returns the score returned by matching the candidate to the pattern.
+// This is not designed for parallel use. Multiple candidates must be scored sequentally.
+// Returns a score between 0 and 1 (0 - no match, 1 - perfect match).
+func (m *Matcher) Score(candidate string) float32 {
+ if len(candidate) > MaxInputSize {
+ candidate = candidate[:MaxInputSize]
+ }
+ lower := ToLower(candidate, m.lowerBuf[:])
+ m.lastCandidateLen = len(candidate)
+
+ if len(m.pattern) == 0 {
+ // Empty patterns perfectly match candidates.
+ return 1
+ }
+
+ if m.match(candidate, lower) {
+ sc := m.computeScore(candidate, lower)
+ if sc > minScore/2 && !m.poorMatch() {
+ m.lastCandidateMatched = true
+ if len(m.pattern) == len(candidate) {
+ // Perfect match.
+ return 1
+ }
+
+ if sc < 0 {
+ sc = 0
+ }
+ normalizedScore := float32(sc) * m.scoreScale
+ if normalizedScore > 1 {
+ normalizedScore = 1
+ }
+
+ return normalizedScore
+ }
+ }
+
+ m.lastCandidateMatched = false
+ return -1
+}
+
+const minScore = -10000
+
+// MatchedRanges returns matches ranges for the last scored string as a flattened array of
+// [begin, end) byte offset pairs.
+func (m *Matcher) MatchedRanges() []int {
+ if len(m.pattern) == 0 || !m.lastCandidateMatched {
+ return nil
+ }
+ i, j := m.lastCandidateLen, len(m.pattern)
+ if m.scores[i][j][0].val() < minScore/2 && m.scores[i][j][1].val() < minScore/2 {
+ return nil
+ }
+
+ var ret []int
+ k := m.bestK(i, j)
+ for i > 0 {
+ take := (k == 1)
+ k = m.scores[i][j][k].prevK()
+ if take {
+ if len(ret) == 0 || ret[len(ret)-1] != i {
+ ret = append(ret, i)
+ ret = append(ret, i-1)
+ } else {
+ ret[len(ret)-1] = i - 1
+ }
+ j--
+ }
+ i--
+ }
+ // Reverse slice.
+ for i := 0; i < len(ret)/2; i++ {
+ ret[i], ret[len(ret)-1-i] = ret[len(ret)-1-i], ret[i]
+ }
+ return ret
+}
+
+func (m *Matcher) match(candidate string, candidateLower []byte) bool {
+ i, j := 0, 0
+ for ; i < len(candidateLower) && j < len(m.patternLower); i++ {
+ if candidateLower[i] == m.patternLower[j] {
+ j++
+ }
+ }
+ if j != len(m.patternLower) {
+ return false
+ }
+
+ // The input passes the simple test against pattern, so it is time to classify its characters.
+ // Character roles are used below to find the last segment.
+ m.roles = RuneRoles(candidate, m.input, m.rolesBuf[:])
+ if m.input != Text {
+ sep := len(candidateLower) - 1
+ for sep >= i && m.roles[sep] != RSep {
+ sep--
+ }
+ if sep >= i {
+ // We are not in the last segment, check that we have at least one character match in the last
+ // segment of the candidate.
+ return bytes.IndexByte(candidateLower[sep:], m.patternLower[len(m.pattern)-1]) != -1
+ }
+ }
+ return true
+}
+
+func (m *Matcher) computeScore(candidate string, candidateLower []byte) int {
+ pattLen, candLen := len(m.pattern), len(candidate)
+
+ for j := 0; j <= len(m.pattern); j++ {
+ m.scores[0][j][0] = minScore << 1
+ m.scores[0][j][1] = minScore << 1
+ }
+ m.scores[0][0][0] = score(0, 0) // Start with 0.
+
+ segmentsLeft, lastSegStart := 1, 0
+ for i := 0; i < candLen; i++ {
+ if m.roles[i] == RSep {
+ segmentsLeft++
+ lastSegStart = i + 1
+ }
+ }
+
+ // A per-character bonus for a consecutive match.
+ consecutiveBonus := 2
+ if m.input == Text {
+ // Consecutive matches for text are more important.
+ consecutiveBonus = 4
+ }
+ wordIdx := 0 // Word count within segment.
+ for i := 1; i <= candLen; i++ {
+
+ role := m.roles[i-1]
+ isHead := role == RHead
+
+ if isHead {
+ wordIdx++
+ } else if role == RSep && segmentsLeft > 1 {
+ wordIdx = 0
+ segmentsLeft--
+ }
+
+ var skipPenalty int
+ if segmentsLeft == 1 && isHead && m.input != Text {
+ // Skipping a word.
+ skipPenalty++
+ }
+ if i-1 == lastSegStart {
+ // Skipping the start of the last segment.
+ skipPenalty += 3
+ }
+
+ for j := 0; j <= pattLen; j++ {
+ // By default, we don't have a match. Fill in the skip data.
+ m.scores[i][j][1] = minScore << 1
+
+ if segmentsLeft > 1 && j == pattLen {
+ // The very last pattern character can only be matched in the last segment.
+ m.scores[i][j][0] = minScore << 1
+ continue
+ }
+
+ // Compute the skip score.
+ k := 0
+ if m.scores[i-1][j][0].val() < m.scores[i-1][j][1].val() {
+ k = 1
+ }
+
+ skipScore := m.scores[i-1][j][k].val()
+ // Do not penalize missing characters after the last matched segment.
+ if j != pattLen {
+ skipScore -= skipPenalty
+ }
+ m.scores[i][j][0] = score(skipScore, k)
+
+ if j == 0 || candidateLower[i-1] != m.patternLower[j-1] {
+ // Not a match.
+ continue
+ }
+ pRole := m.patternRoles[j-1]
+
+ if role == RTail && pRole == RHead {
+ if j > 1 {
+ // Not a match: a head in the pattern matches a tail character in the candidate.
+ continue
+ }
+ // Special treatment for the first character of the pattern. We allow
+ // matches in the middle of a word if they are long enough, at least
+ // min(3, pattern.length) characters.
+ if !bytes.HasPrefix(candidateLower[i-1:], m.patternShort) {
+ continue
+ }
+ }
+
+ // Compute the char score.
+ var charScore int
+ // Bonus 1: the char is in the candidate's last segment.
+ if segmentsLeft <= 1 {
+ charScore++
+ }
+ // Bonus 2: Case match or a Head in the pattern aligns with one in the word.
+ // Single-case patterns lack segmentation signals and we assume any character
+ // can be a head of a segment.
+ if candidate[i-1] == m.pattern[j-1] || role == RHead && (!m.caseSensitive || pRole == RHead) {
+ charScore++
+ }
+
+ // Penalty 1: pattern char is Head, candidate char is Tail.
+ if role == RTail && pRole == RHead {
+ charScore--
+ }
+ // Penalty 2: first pattern character matched in the middle of a word.
+ if j == 1 && role == RTail {
+ charScore -= 4
+ }
+
+ // Third dimension encodes whether there is a gap between the previous match and the current
+ // one.
+ for k := 0; k < 2; k++ {
+ sc := m.scores[i-1][j-1][k].val() + charScore
+
+ isConsecutive := k == 1 || i-1 == 0 || i-1 == lastSegStart
+ if isConsecutive || (m.input == Text && j-1 == 0) {
+ // Bonus 3: a consecutive match. First character match also gets a bonus to
+ // ensure prefix final match score normalizes to 1.0.
+ // Logically, this is a part of charScore, but we have to compute it here because it
+ // only applies for consecutive matches (k == 1).
+ sc += consecutiveBonus
+ }
+ if k == 0 {
+ // Penalty 3: Matching inside a segment (and previous char wasn't matched). Penalize for the lack
+ // of alignment.
+ if role == RTail || role == RUCTail {
+ sc -= 3
+ }
+ }
+
+ if sc > m.scores[i][j][1].val() {
+ m.scores[i][j][1] = score(sc, k)
+ }
+ }
+ }
+ }
+
+ result := m.scores[len(candidate)][len(m.pattern)][m.bestK(len(candidate), len(m.pattern))].val()
+
+ return result
+}
+
+// ScoreTable returns the score table computed for the provided candidate. Used only for debugging.
+func (m *Matcher) ScoreTable(candidate string) string {
+ var buf bytes.Buffer
+
+ var line1, line2, separator bytes.Buffer
+ line1.WriteString("\t")
+ line2.WriteString("\t")
+ for j := 0; j < len(m.pattern); j++ {
+ line1.WriteString(fmt.Sprintf("%c\t\t", m.pattern[j]))
+ separator.WriteString("----------------")
+ }
+
+ buf.WriteString(line1.String())
+ buf.WriteString("\n")
+ buf.WriteString(separator.String())
+ buf.WriteString("\n")
+
+ for i := 1; i <= len(candidate); i++ {
+ line1.Reset()
+ line2.Reset()
+
+ line1.WriteString(fmt.Sprintf("%c\t", candidate[i-1]))
+ line2.WriteString("\t")
+
+ for j := 1; j <= len(m.pattern); j++ {
+ line1.WriteString(fmt.Sprintf("M%6d(%c)\t", m.scores[i][j][0].val(), dir(m.scores[i][j][0].prevK())))
+ line2.WriteString(fmt.Sprintf("H%6d(%c)\t", m.scores[i][j][1].val(), dir(m.scores[i][j][1].prevK())))
+ }
+ buf.WriteString(line1.String())
+ buf.WriteString("\n")
+ buf.WriteString(line2.String())
+ buf.WriteString("\n")
+ buf.WriteString(separator.String())
+ buf.WriteString("\n")
+ }
+
+ return buf.String()
+}
+
+func dir(prevK int) rune {
+ if prevK == 0 {
+ return 'M'
+ }
+ return 'H'
+}
+
+func (m *Matcher) poorMatch() bool {
+ if len(m.pattern) < 2 {
+ return false
+ }
+
+ i, j := m.lastCandidateLen, len(m.pattern)
+ k := m.bestK(i, j)
+
+ var counter, len int
+ for i > 0 {
+ take := (k == 1)
+ k = m.scores[i][j][k].prevK()
+ if take {
+ len++
+ if k == 0 && len < 3 && m.roles[i-1] == RTail {
+ // Short match in the middle of a word
+ counter++
+ if counter > 1 {
+ return true
+ }
+ }
+ j--
+ } else {
+ len = 0
+ }
+ i--
+ }
+ return false
+}
diff --git a/internal/lsp/fuzzy/matcher_test.go b/internal/lsp/fuzzy/matcher_test.go
new file mode 100644
index 0000000..49f1caa
--- /dev/null
+++ b/internal/lsp/fuzzy/matcher_test.go
@@ -0,0 +1,352 @@
+// Copyright 2019 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Benchmark results:
+//
+// BenchmarkMatcher-12 1000000 1615 ns/op 30.95 MB/s 0 B/op 0 allocs/op
+//
+package fuzzy_test
+
+import (
+ "bytes"
+ "fmt"
+ "math"
+ "testing"
+
+ "golang.org/x/tools/internal/lsp/fuzzy"
+)
+
+func ExampleFuzzyMatcher() {
+ pattern := "TEdit"
+ candidates := []string{"fuzzy.TextEdit", "ArtEdit", "TED talks about IT"}
+
+ // Create a fuzzy matcher for the pattern.
+ matcher := fuzzy.NewMatcher(pattern, fuzzy.Text)
+
+ for _, candidate := range candidates {
+ // Compute candidate's score against the matcher.
+ score := matcher.Score(candidate)
+
+ if score > -1 {
+ // Get the substrings in the candidate matching the pattern.
+ ranges := matcher.MatchedRanges()
+
+ fmt.Println(ranges) // Do something with the ranges.
+ }
+ }
+}
+
+type comparator struct {
+ f func(val, ref float32) bool
+ descr string
+}
+
+var (
+ eq = comparator{
+ f: func(val, ref float32) bool {
+ return val == ref
+ },
+ descr: "==",
+ }
+ ge = comparator{
+ f: func(val, ref float32) bool {
+ return val >= ref
+ },
+ descr: ">=",
+ }
+)
+
+func (c comparator) eval(val, ref float32) bool {
+ return c.f(val, ref)
+}
+
+func (c comparator) String() string {
+ return c.descr
+}
+
+type scoreTest struct {
+ candidate string
+ comparator
+ ref float32
+}
+
+var matcherTests = []struct {
+ pattern string
+ input fuzzy.Input
+ tests []scoreTest
+}{
+ {
+ pattern: "",
+ input: fuzzy.Text,
+ tests: []scoreTest{
+ {"def", eq, 1},
+ {"Ab stuff c", eq, 1},
+ },
+ },
+ {
+ pattern: "abc",
+ input: fuzzy.Text,
+ tests: []scoreTest{
+ {"def", eq, -1},
+ {"abd", eq, -1},
+ {"abc", ge, 0},
+ {"Abc", ge, 0},
+ {"Ab stuff c", ge, 0},
+ },
+ },
+ {
+ pattern: "Abc",
+ input: fuzzy.Text,
+ tests: []scoreTest{
+ {"def", eq, -1},
+ {"abd", eq, -1},
+ {"abc", ge, 0},
+ {"Abc", ge, 0},
+ {"Ab stuff c", ge, 0},
+ },
+ },
+ {
+ pattern: "subs",
+ input: fuzzy.Filename,
+ tests: []scoreTest{
+ {"sub/seq", ge, 0},
+ {"sub/seq/end", eq, -1},
+ {"sub/seq/base", ge, 0},
+ },
+ },
+ {
+ pattern: "subs",
+ input: fuzzy.Filename,
+ tests: []scoreTest{
+ {"//sub/seq", ge, 0},
+ {"//sub/seq/end", eq, -1},
+ {"//sub/seq/base", ge, 0},
+ },
+ },
+}
+
+func TestScore(t *testing.T) {
+ for _, tc := range matcherTests {
+ m := fuzzy.NewMatcher(tc.pattern, tc.input)
+ for _, sct := range tc.tests {
+ score := m.Score(sct.candidate)
+ if !sct.comparator.eval(score, sct.ref) {
+ t.Errorf("not true that m.Score(%s)[=%v] %s %v", sct.candidate, score, sct.comparator, sct.ref)
+ }
+ }
+ }
+}
+
+type candidateCompTest struct {
+ c1 string
+ comparator comparator
+ c2 string
+}
+
+var compareCandidatesTestCases = []struct {
+ pattern string
+ input fuzzy.Input
+ orderedCandidates []string
+}{
+ {
+ pattern: "aa",
+ input: fuzzy.Filename,
+ orderedCandidates: []string{
+ "baab",
+ "bb_aa",
+ "a/a/a",
+ "aa_bb",
+ "aa_b",
+ "aabb",
+ "aab",
+ "b/aa",
+ },
+ },
+ {
+ pattern: "Foo",
+ input: fuzzy.Text,
+ orderedCandidates: []string{
+ "Barfoo",
+ "F_o_o",
+ "Faoo",
+ "F__oo",
+ "F_oo",
+ "FaoFooa",
+ "BarFoo",
+ "FooA",
+ "FooBar",
+ "Foo",
+ },
+ },
+}
+
+func TestCompareCandidateScores(t *testing.T) {
+ for _, tc := range compareCandidatesTestCases {
+ m := fuzzy.NewMatcher(tc.pattern, tc.input)
+
+ var prevScore float32
+ prevCand := "MIN_SCORE"
+ for _, cand := range tc.orderedCandidates {
+ score := m.Score(cand)
+ if prevScore > score {
+ t.Errorf("%s[=%v] is scored lower than %s[=%v]", cand, score, prevCand, prevScore)
+ }
+ if score < -1 || score > 1 {
+ t.Errorf("%s score is %v; want value between [-1, 1]", cand, score)
+ }
+ prevScore = score
+ prevCand = cand
+ }
+ }
+}
+
+var fuzzyMatcherTestCases = []struct {
+ p string
+ str string
+ want string
+ input fuzzy.Input
+}{
+ // fuzzy.Filename
+ {p: "aa", str: "a_a/a_a", want: "[a]_a/[a]_a", input: fuzzy.Filename},
+ {p: "aaaa", str: "a_a/a_a", want: "[a]_[a]/[a]_[a]", input: fuzzy.Filename},
+ {p: "aaaa", str: "aaaa", want: "[aaaa]", input: fuzzy.Filename},
+ {p: "aaaa", str: "a_a/a_aaaa", want: "a_a/[a]_[aaa]a", input: fuzzy.Filename},
+ {p: "aaaa", str: "a_a/aaaaa", want: "a_a/[aaaa]a", input: fuzzy.Filename},
+ {p: "aaaa", str: "aabaaa", want: "[aa]b[aa]a", input: fuzzy.Filename},
+ {p: "aaaa", str: "a/baaa", want: "[a]/b[aaa]", input: fuzzy.Filename},
+ {p: "abcxz", str: "d/abc/abcd/oxz", want: "d/[abc]/abcd/o[xz]", input: fuzzy.Filename},
+ {p: "abcxz", str: "d/abcd/abc/oxz", want: "d/[abc]d/abc/o[xz]", input: fuzzy.Filename},
+
+ // fuzzy.Symbol
+ {p: "foo", str: "abc::foo", want: "abc::[foo]", input: fuzzy.Symbol},
+ {p: "foo", str: "foo.foo", want: "foo.[foo]", input: fuzzy.Symbol},
+ {p: "foo", str: "fo_oo.o_oo", want: "[fo]_oo.[o]_oo", input: fuzzy.Symbol},
+ {p: "foo", str: "fo_oo.fo_oo", want: "fo_oo.[fo]_[o]o", input: fuzzy.Symbol},
+ {p: "fo_o", str: "fo_oo.o_oo", want: "[f]o_oo.[o_o]o", input: fuzzy.Symbol},
+ {p: "fOO", str: "fo_oo.o_oo", want: "[f]o_oo.[o]_[o]o", input: fuzzy.Symbol},
+ {p: "tedit", str: "foo.TextEdit", want: "foo.[T]ext[Edit]", input: fuzzy.Symbol},
+ {p: "TEdit", str: "foo.TextEdit", want: "foo.[T]ext[Edit]", input: fuzzy.Symbol},
+ {p: "Tedit", str: "foo.TextEdit", want: "foo.[T]ext[Edit]", input: fuzzy.Symbol},
+ {p: "Tedit", str: "foo.Textedit", want: "foo.[Te]xte[dit]", input: fuzzy.Symbol},
+ {p: "TEdit", str: "foo.Textedit", want: "", input: fuzzy.Symbol},
+ {p: "te", str: "foo.Textedit", want: "foo.[Te]xtedit", input: fuzzy.Symbol},
+ {p: "ee", str: "foo.Textedit", want: "", input: fuzzy.Symbol}, // short middle of the word match
+ {p: "ex", str: "foo.Textedit", want: "foo.T[ex]tedit", input: fuzzy.Symbol},
+ {p: "exdi", str: "foo.Textedit", want: "", input: fuzzy.Symbol}, // short middle of the word match
+ {p: "exdit", str: "foo.Textedit", want: "", input: fuzzy.Symbol}, // short middle of the word match
+ {p: "extdit", str: "foo.Textedit", want: "foo.T[ext]e[dit]", input: fuzzy.Symbol},
+ {p: "e", str: "foo.Textedit", want: "foo.T[e]xtedit", input: fuzzy.Symbol},
+ {p: "E", str: "foo.Textedit", want: "foo.T[e]xtedit", input: fuzzy.Symbol},
+ {p: "ed", str: "foo.Textedit", want: "foo.Text[ed]it", input: fuzzy.Symbol},
+ {p: "edt", str: "foo.Textedit", want: "", input: fuzzy.Symbol}, // short middle of the word match
+ {p: "edit", str: "foo.Textedit", want: "foo.Text[edit]", input: fuzzy.Symbol},
+ {p: "edin", str: "foo.TexteditNum", want: "foo.Text[edi]t[N]um", input: fuzzy.Symbol},
+ {p: "n", str: "node.GoNodeMax", want: "node.Go[N]odeMax", input: fuzzy.Symbol},
+ {p: "N", str: "node.GoNodeMax", want: "node.Go[N]odeMax", input: fuzzy.Symbol},
+ {p: "completio", str: "completion", want: "[completio]n", input: fuzzy.Symbol},
+ {p: "completio", str: "completion.None", want: "[completi]on.N[o]ne", input: fuzzy.Symbol},
+}
+
+func TestFuzzyMatcherRanges(t *testing.T) {
+ for _, tc := range fuzzyMatcherTestCases {
+ matcher := fuzzy.NewMatcher(tc.p, tc.input)
+ score := matcher.Score(tc.str)
+ if tc.want == "" {
+ if score >= 0 {
+ t.Errorf("Score(%s, %s) = %v; want: <= 0", tc.p, tc.str, score)
+ }
+ continue
+ }
+ if score < 0 {
+ t.Errorf("Score(%s, %s) = %v, want: > 0", tc.p, tc.str, score)
+ continue
+ }
+ got := highlightMatches(tc.str, matcher)
+ if tc.want != got {
+ t.Errorf("highlightMatches(%s, %s) = %v, want: %v", tc.p, tc.str, got, tc.want)
+ }
+ }
+}
+
+var scoreTestCases = []struct {
+ p string
+ str string
+ want float64
+}{
+ // Score precision up to five digits. Modify if changing the score, but make sure the new values
+ // are reasonable.
+ {p: "abc", str: "abc", want: 1},
+ {p: "abc", str: "Abc", want: 1},
+ {p: "abc", str: "Abcdef", want: 1},
+ {p: "strc", str: "StrCat", want: 1},
+ {p: "abc_def", str: "abc_def_xyz", want: 1},
+ {p: "abcdef", str: "abc_def_xyz", want: 0.91667},
+ {p: "abcxyz", str: "abc_def_xyz", want: 0.875},
+ {p: "sc", str: "StrCat", want: 0.75},
+ {p: "abc", str: "AbstrBasicCtor", want: 0.75},
+ {p: "foo", str: "abc::foo", want: 1},
+ {p: "afoo", str: "abc::foo", want: 0.9375},
+ {p: "abr", str: "abc::bar", want: 0.5},
+ {p: "br", str: "abc::bar", want: 0.375},
+ {p: "aar", str: "abc::bar", want: 0.16667},
+ {p: "edin", str: "foo.TexteditNum", want: 0},
+ {p: "ediu", str: "foo.TexteditNum", want: 0},
+ // We want the next two items to have roughly similar scores.
+ {p: "up", str: "unique_ptr", want: 0.75},
+ {p: "up", str: "upper_bound", want: 1},
+}
+
+func TestScores(t *testing.T) {
+ for _, tc := range scoreTestCases {
+ matcher := fuzzy.NewMatcher(tc.p, fuzzy.Symbol)
+ got := math.Round(float64(matcher.Score(tc.str))*1e5) / 1e5
+ if got != tc.want {
+ t.Errorf("Score(%s, %s) = %v, want: %v", tc.p, tc.str, got, tc.want)
+ }
+ }
+}
+
+func highlightMatches(str string, matcher *fuzzy.Matcher) string {
+ matches := matcher.MatchedRanges()
+
+ var buf bytes.Buffer
+ index := 0
+ for i := 0; i < len(matches)-1; i += 2 {
+ s, e := matches[i], matches[i+1]
+ fmt.Fprintf(&buf, "%s[%s]", str[index:s], str[s:e])
+ index = e
+ }
+ buf.WriteString(str[index:])
+ return buf.String()
+}
+
+func BenchmarkMatcher(b *testing.B) {
+ pattern := "Foo"
+ candidates := []string{
+ "F_o_o",
+ "Barfoo",
+ "Faoo",
+ "F__oo",
+ "F_oo",
+ "FaoFooa",
+ "BarFoo",
+ "FooA",
+ "FooBar",
+ "Foo",
+ }
+
+ matcher := fuzzy.NewMatcher(pattern, fuzzy.Text)
+
+ b.ResetTimer()
+ for i := 0; i < b.N; i++ {
+ for _, c := range candidates {
+ matcher.Score(c)
+ }
+ }
+ var numBytes int
+ for _, c := range candidates {
+ numBytes += len(c)
+ }
+ b.SetBytes(int64(numBytes))
+}
diff --git a/internal/lsp/source/completion.go b/internal/lsp/source/completion.go
index 8bf7e21..33980eb 100644
--- a/internal/lsp/source/completion.go
+++ b/internal/lsp/source/completion.go
@@ -12,6 +12,7 @@
"go/types"
"golang.org/x/tools/go/ast/astutil"
+ "golang.org/x/tools/internal/lsp/fuzzy"
"golang.org/x/tools/internal/lsp/snippet"
"golang.org/x/tools/internal/lsp/telemetry/trace"
"golang.org/x/tools/internal/span"
@@ -149,6 +150,9 @@
// deepState contains the current state of our deep completion search.
deepState deepCompletionState
+
+ // matcher does fuzzy matching of the candidates for the surrounding prefix.
+ matcher *fuzzy.Matcher
}
type compLitInfo struct {
@@ -187,16 +191,17 @@
if c.surrounding != nil {
return
}
-
if !(ident.Pos() <= c.pos && c.pos <= ident.End()) {
return
}
-
c.surrounding = &Selection{
Content: ident.Name,
Range: span.NewRange(c.view.Session().Cache().FileSet(), ident.Pos(), ident.End()),
Cursor: c.pos,
}
+ if c.surrounding.Prefix() != "" {
+ c.matcher = fuzzy.NewMatcher(c.surrounding.Prefix(), fuzzy.Symbol)
+ }
}
// found adds a candidate completion. We will also search through the object's