mirror of
https://github.com/VictoriaMetrics/VictoriaMetrics.git
synced 2024-11-21 14:44:00 +00:00
87 lines
2.5 KiB
Go
87 lines
2.5 KiB
Go
package uniseg
|
|
|
|
import "unicode/utf8"
|
|
|
|
// FirstWord returns the first word found in the given byte slice according to
|
|
// the rules of Unicode Standard Annex #29, Word Boundaries. This function can
|
|
// be called continuously to extract all words from a byte slice, as illustrated
|
|
// in the example below.
|
|
//
|
|
// If you don't know the current state, for example when calling the function
|
|
// for the first time, you must pass -1. For consecutive calls, pass the state
|
|
// and rest slice returned by the previous call.
|
|
//
|
|
// The "rest" slice is the sub-slice of the original byte slice "b" starting
|
|
// after the last byte of the identified word. If the length of the "rest" slice
|
|
// is 0, the entire byte slice "b" has been processed. The "word" byte slice is
|
|
// the sub-slice of the input slice containing the identified word.
|
|
//
|
|
// Given an empty byte slice "b", the function returns nil values.
|
|
func FirstWord(b []byte, state int) (word, rest []byte, newState int) {
|
|
// An empty byte slice returns nothing.
|
|
if len(b) == 0 {
|
|
return
|
|
}
|
|
|
|
// Extract the first rune.
|
|
r, length := utf8.DecodeRune(b)
|
|
if len(b) <= length { // If we're already past the end, there is nothing else to parse.
|
|
return b, nil, wbAny
|
|
}
|
|
|
|
// If we don't know the state, determine it now.
|
|
if state < 0 {
|
|
state, _ = transitionWordBreakState(state, r, b[length:], "")
|
|
}
|
|
|
|
// Transition until we find a boundary.
|
|
var boundary bool
|
|
for {
|
|
r, l := utf8.DecodeRune(b[length:])
|
|
state, boundary = transitionWordBreakState(state, r, b[length+l:], "")
|
|
|
|
if boundary {
|
|
return b[:length], b[length:], state
|
|
}
|
|
|
|
length += l
|
|
if len(b) <= length {
|
|
return b, nil, wbAny
|
|
}
|
|
}
|
|
}
|
|
|
|
// FirstWordInString is like [FirstWord] but its input and outputs are strings.
|
|
func FirstWordInString(str string, state int) (word, rest string, newState int) {
|
|
// An empty byte slice returns nothing.
|
|
if len(str) == 0 {
|
|
return
|
|
}
|
|
|
|
// Extract the first rune.
|
|
r, length := utf8.DecodeRuneInString(str)
|
|
if len(str) <= length { // If we're already past the end, there is nothing else to parse.
|
|
return str, "", wbAny
|
|
}
|
|
|
|
// If we don't know the state, determine it now.
|
|
if state < 0 {
|
|
state, _ = transitionWordBreakState(state, r, nil, str[length:])
|
|
}
|
|
|
|
// Transition until we find a boundary.
|
|
var boundary bool
|
|
for {
|
|
r, l := utf8.DecodeRuneInString(str[length:])
|
|
state, boundary = transitionWordBreakState(state, r, nil, str[length+l:])
|
|
|
|
if boundary {
|
|
return str[:length], str[length:], state
|
|
}
|
|
|
|
length += l
|
|
if len(str) <= length {
|
|
return str, "", wbAny
|
|
}
|
|
}
|
|
}
|