package glob import ( "fmt" "github.com/gobwas/glob/match" "strings" ) const ( any = '*' single = '?' escape = '\\' range_open = '[' range_close = ']' ) const ( inside_range_not = '!' inside_range_minus = '-' ) var syntaxPhrases = string([]byte{any, single, escape, range_open, range_close}) // Glob represents compiled glob pattern. type Glob interface { Match(string) bool } // New creates Glob for given pattern and uses other given (if any) strings as separators. // The pattern syntax is: // // pattern: // { term } // term: // `*` matches any sequence of non-separator characters // `**` matches any sequence of characters // `?` matches any single non-separator character // c matches character c (c != `*`, `**`, `?`, `\`) // `\` c matches character c func New(pattern string, separators ...string) (Glob, error) { chunks, err := parse(pattern, strings.Join(separators, ""), state{}) if err != nil { return nil, err } switch len(chunks) { case 1: return chunks[0].matcher, nil case 2: if chunks[0].matcher.Kind() == match.KindRaw && chunks[1].matcher.Kind() == match.KindMultipleSuper { return &match.Prefix{chunks[0].str}, nil } if chunks[1].matcher.Kind() == match.KindRaw && chunks[0].matcher.Kind() == match.KindMultipleSuper { return &match.Suffix{chunks[1].str}, nil } case 3: if chunks[0].matcher.Kind() == match.KindRaw && chunks[1].matcher.Kind() == match.KindMultipleSuper && chunks[2].matcher.Kind() == match.KindRaw { return &match.PrefixSuffix{chunks[0].str, chunks[2].str}, nil } } var c []match.Matcher for _, chunk := range chunks { c = append(c, chunk.matcher) } return &match.Composite{c}, nil } // parse parsed given pattern into list of tokens func parse(str string, sep string, st state) ([]token, error) { if len(str) == 0 { return st.tokens, nil } // if there are no syntax symbols - pattern is simple string i := strings.IndexAny(str, syntaxPhrases) if i == -1 { return append(st.tokens, token{match.Raw{str}, str}), nil } c := string(str[i]) // if syntax symbol is not at the start of pattern - add raw part before it if i > 0 { st.tokens = append(st.tokens, token{match.Raw{str[0:i]}, str[0:i]}) } // if we are in escape state if st.escape { st.tokens = append(st.tokens, token{match.Raw{c}, c}) st.escape = false } else { switch str[i] { case range_open: closed := indexByteNonEscaped(str, range_close, escape, 0) if closed == -1 { return nil, fmt.Errorf("'%s' should be closed with '%s'", string(range_open), string(range_close)) } r := str[i+1 : closed] g, err := parseRange(r) if err != nil { return nil, err } st.tokens = append(st.tokens, token{g, r}) if closed == len(str)-1 { return st.tokens, nil } return parse(str[closed+1:], sep, st) case escape: st.escape = true case any: if len(str) > i+1 && str[i+1] == any { st.tokens = append(st.tokens, token{match.Any{}, c}) return parse(str[i+len(c)+1:], sep, st) } st.tokens = append(st.tokens, token{match.Any{sep}, c}) case single: st.tokens = append(st.tokens, token{match.Single{sep}, c}) } } return parse(str[i+len(c):], sep, st) } func parseRange(def string) (match.Matcher, error) { var ( not bool esc bool minus bool minusIndex int b []byte ) for i, c := range []byte(def) { if esc { b = append(b, c) esc = false continue } switch c { case inside_range_not: if i == 0 { not = true } case escape: if i == len(def)-1 { return nil, fmt.Errorf("there should be any character after '%s'", string(escape)) } esc = true case inside_range_minus: minus = true minusIndex = len(b) default: b = append(b, c) } } if len(b) == 0 { return nil, fmt.Errorf("range could not be empty") } def = string(b) if minus { r := []rune(def) if len(r) != 2 || minusIndex != 1 { return nil, fmt.Errorf("invalid range syntax: '%s' should be between two characters", string(inside_range_minus)) } return &match.Between{r[0], r[1], not}, nil } return &match.RangeList{def, not}, nil } type token struct { matcher match.Matcher str string } type state struct { escape bool tokens []token }