VictoriaMetrics/vendor/github.com/rivo/uniseg/gen_breaktest.go

214 lines
5.8 KiB
Go
Raw Normal View History

2022-08-02 08:19:38 +02:00
//go:build generate
// This program generates a Go containing a slice of test cases based on the
// Unicode Character Database auxiliary data files. The command line arguments
// are as follows:
//
// 1. The name of the Unicode data file (just the filename, without extension).
// 2. The name of the locally generated Go file.
// 3. The name of the slice containing the test cases.
// 4. The name of the generator, for logging purposes.
//
//go:generate go run gen_breaktest.go GraphemeBreakTest graphemebreak_test.go graphemeBreakTestCases graphemes
//go:generate go run gen_breaktest.go WordBreakTest wordbreak_test.go wordBreakTestCases words
//go:generate go run gen_breaktest.go SentenceBreakTest sentencebreak_test.go sentenceBreakTestCases sentences
//go:generate go run gen_breaktest.go LineBreakTest linebreak_test.go lineBreakTestCases lines
package main
import (
"bufio"
"bytes"
"errors"
"fmt"
"go/format"
"io/ioutil"
"log"
"net/http"
"os"
"time"
)
// We want to test against a specific version rather than the latest. When the
// package is upgraded to a new version, change these to generate new tests.
const (
testCaseURL = `https://www.unicode.org/Public/14.0.0/ucd/auxiliary/%s.txt`
)
func main() {
if len(os.Args) < 5 {
fmt.Println("Not enough arguments, see code for details")
os.Exit(1)
}
log.SetPrefix("gen_breaktest (" + os.Args[4] + "): ")
log.SetFlags(0)
// Read text of testcases and parse into Go source code.
src, err := parse(fmt.Sprintf(testCaseURL, os.Args[1]))
if err != nil {
log.Fatal(err)
}
// Format the Go code.
formatted, err := format.Source(src)
if err != nil {
log.Fatalln("gofmt:", err)
}
// Write it out.
log.Print("Writing to ", os.Args[2])
if err := ioutil.WriteFile(os.Args[2], formatted, 0644); err != nil {
log.Fatal(err)
}
}
// parse reads a break text file, either from a local file or from a URL. It
// parses the file data into Go source code representing the test cases.
func parse(url string) ([]byte, error) {
log.Printf("Parsing %s", url)
res, err := http.Get(url)
if err != nil {
return nil, err
}
body := res.Body
defer body.Close()
buf := new(bytes.Buffer)
buf.Grow(120 << 10)
buf.WriteString(`package uniseg
// Code generated via go generate from gen_breaktest.go. DO NOT EDIT.
// ` + os.Args[3] + ` are Grapheme testcases taken from
// ` + url + `
// on ` + time.Now().Format("January 2, 2006") + `. See
// https://www.unicode.org/license.html for the Unicode license agreement.
var ` + os.Args[3] + ` = []testCase {
`)
sc := bufio.NewScanner(body)
num := 1
var line []byte
original := make([]byte, 0, 64)
expected := make([]byte, 0, 64)
for sc.Scan() {
num++
line = sc.Bytes()
if len(line) == 0 || line[0] == '#' {
continue
}
var comment []byte
if i := bytes.IndexByte(line, '#'); i >= 0 {
comment = bytes.TrimSpace(line[i+1:])
line = bytes.TrimSpace(line[:i])
}
original, expected, err := parseRuneSequence(line, original[:0], expected[:0])
if err != nil {
return nil, fmt.Errorf(`line %d: %v: %q`, num, err, line)
}
fmt.Fprintf(buf, "\t{original: \"%s\", expected: %s}, // %s\n", original, expected, comment)
}
if err := sc.Err(); err != nil {
return nil, err
}
// Check for final "# EOF", useful check if we're streaming via HTTP
if !bytes.Equal(line, []byte("# EOF")) {
return nil, fmt.Errorf(`line %d: exected "# EOF" as final line, got %q`, num, line)
}
buf.WriteString("}\n")
return buf.Bytes(), nil
}
// Used by parseRuneSequence to match input via bytes.HasPrefix.
var (
prefixBreak = []byte("÷ ")
prefixDontBreak = []byte("× ")
breakOk = []byte("÷")
breakNo = []byte("×")
)
// parseRuneSequence parses a rune + breaking opportunity sequence from b
// and appends the Go code for testcase.original to orig
// and appends the Go code for testcase.expected to exp.
// It retuns the new orig and exp slices.
//
// E.g. for the input b="÷ 0020 × 0308 ÷ 1F1E6 ÷"
// it will append
// "\u0020\u0308\U0001F1E6"
// and "[][]rune{{0x0020,0x0308},{0x1F1E6},}"
// to orig and exp respectively.
//
// The formatting of exp is expected to be cleaned up by gofmt or format.Source.
// Note we explicitly require the sequence to start with ÷ and we implicitly
// require it to end with ÷.
func parseRuneSequence(b, orig, exp []byte) ([]byte, []byte, error) {
// Check for and remove first ÷ or ×.
if !bytes.HasPrefix(b, prefixBreak) && !bytes.HasPrefix(b, prefixDontBreak) {
return nil, nil, errors.New("expected ÷ or × as first character")
}
if bytes.HasPrefix(b, prefixBreak) {
b = b[len(prefixBreak):]
} else {
b = b[len(prefixDontBreak):]
}
boundary := true
exp = append(exp, "[][]rune{"...)
for len(b) > 0 {
if boundary {
exp = append(exp, '{')
}
exp = append(exp, "0x"...)
// Find end of hex digits.
var i int
for i = 0; i < len(b) && b[i] != ' '; i++ {
if d := b[i]; ('0' <= d || d <= '9') ||
('A' <= d || d <= 'F') ||
('a' <= d || d <= 'f') {
continue
}
return nil, nil, errors.New("bad hex digit")
}
switch i {
case 4:
orig = append(orig, "\\u"...)
case 5:
orig = append(orig, "\\U000"...)
default:
return nil, nil, errors.New("unsupport code point hex length")
}
orig = append(orig, b[:i]...)
exp = append(exp, b[:i]...)
b = b[i:]
// Check for space between hex and ÷ or ×.
if len(b) < 1 || b[0] != ' ' {
return nil, nil, errors.New("bad input")
}
b = b[1:]
// Check for next boundary.
switch {
case bytes.HasPrefix(b, breakOk):
boundary = true
b = b[len(breakOk):]
case bytes.HasPrefix(b, breakNo):
boundary = false
b = b[len(breakNo):]
default:
return nil, nil, errors.New("missing ÷ or ×")
}
if boundary {
exp = append(exp, '}')
}
exp = append(exp, ',')
if len(b) > 0 && b[0] == ' ' {
b = b[1:]
}
}
exp = append(exp, '}')
return orig, exp, nil
}