Commit cadbd3ea authored by Marcel van Lohuizen's avatar Marcel van Lohuizen

exp/norm: fixed two unrelated bugs in normalization library.

1) incorrect length given for out buffer in String.
2) patchTail bug that could cause characters to be lost
   when crossing into the out-buffer boundary.

Added tests to expose these bugs.  Also slightly improved
performance of Bytes() and String() by sharing the reorderBuffer
across operations.

Fixes #2567.

R=r
CC=golang-dev
https://golang.org/cl/5502069
parent 335c5db7
...@@ -8,7 +8,7 @@ import "unicode/utf8" ...@@ -8,7 +8,7 @@ import "unicode/utf8"
type input interface { type input interface {
skipASCII(p int) int skipASCII(p int) int
skipNonStarter() int skipNonStarter(p int) int
appendSlice(buf []byte, s, e int) []byte appendSlice(buf []byte, s, e int) []byte
copySlice(buf []byte, s, e int) copySlice(buf []byte, s, e int)
charinfo(p int) (uint16, int) charinfo(p int) (uint16, int)
...@@ -25,8 +25,7 @@ func (s inputString) skipASCII(p int) int { ...@@ -25,8 +25,7 @@ func (s inputString) skipASCII(p int) int {
return p return p
} }
func (s inputString) skipNonStarter() int { func (s inputString) skipNonStarter(p int) int {
p := 0
for ; p < len(s) && !utf8.RuneStart(s[p]); p++ { for ; p < len(s) && !utf8.RuneStart(s[p]); p++ {
} }
return p return p
...@@ -71,8 +70,7 @@ func (s inputBytes) skipASCII(p int) int { ...@@ -71,8 +70,7 @@ func (s inputBytes) skipASCII(p int) int {
return p return p
} }
func (s inputBytes) skipNonStarter() int { func (s inputBytes) skipNonStarter(p int) int {
p := 0
for ; p < len(s) && !utf8.RuneStart(s[p]); p++ { for ; p < len(s) && !utf8.RuneStart(s[p]); p++ {
} }
return p return p
......
...@@ -34,24 +34,28 @@ const ( ...@@ -34,24 +34,28 @@ const (
// Bytes returns f(b). May return b if f(b) = b. // Bytes returns f(b). May return b if f(b) = b.
func (f Form) Bytes(b []byte) []byte { func (f Form) Bytes(b []byte) []byte {
n := f.QuickSpan(b) rb := reorderBuffer{}
rb.init(f, b)
n := quickSpan(&rb, 0)
if n == len(b) { if n == len(b) {
return b return b
} }
out := make([]byte, n, len(b)) out := make([]byte, n, len(b))
copy(out, b[0:n]) copy(out, b[0:n])
return f.Append(out, b[n:]...) return doAppend(&rb, out, n)
} }
// String returns f(s). // String returns f(s).
func (f Form) String(s string) string { func (f Form) String(s string) string {
n := f.QuickSpanString(s) rb := reorderBuffer{}
rb.initString(f, s)
n := quickSpan(&rb, 0)
if n == len(s) { if n == len(s) {
return s return s
} }
out := make([]byte, 0, len(s)) out := make([]byte, n, len(s))
copy(out, s[0:n]) copy(out, s[0:n])
return string(f.AppendString(out, s[n:])) return string(doAppend(&rb, out, n))
} }
// IsNormal returns true if b == f(b). // IsNormal returns true if b == f(b).
...@@ -122,23 +126,27 @@ func (f Form) IsNormalString(s string) bool { ...@@ -122,23 +126,27 @@ func (f Form) IsNormalString(s string) bool {
// patchTail fixes a case where a rune may be incorrectly normalized // patchTail fixes a case where a rune may be incorrectly normalized
// if it is followed by illegal continuation bytes. It returns the // if it is followed by illegal continuation bytes. It returns the
// patched buffer and the number of trailing continuation bytes that // patched buffer and whether there were trailing continuation bytes.
// have been dropped. func patchTail(rb *reorderBuffer, buf []byte) ([]byte, bool) {
func patchTail(rb *reorderBuffer, buf []byte) ([]byte, int) {
info, p := lastRuneStart(&rb.f, buf) info, p := lastRuneStart(&rb.f, buf)
if p == -1 || info.size == 0 { if p == -1 || info.size == 0 {
return buf, 0 return buf, false
} }
end := p + int(info.size) end := p + int(info.size)
extra := len(buf) - end extra := len(buf) - end
if extra > 0 { if extra > 0 {
// Potentially allocating memory. However, this only
// happens with ill-formed UTF-8.
x := make([]byte, 0)
x = append(x, buf[len(buf)-extra:]...)
buf = decomposeToLastBoundary(rb, buf[:end]) buf = decomposeToLastBoundary(rb, buf[:end])
if rb.f.composing { if rb.f.composing {
rb.compose() rb.compose()
} }
return rb.flush(buf), extra buf = rb.flush(buf)
return append(buf, x...), true
} }
return buf, 0 return buf, false
} }
func appendQuick(rb *reorderBuffer, dst []byte, i int) ([]byte, int) { func appendQuick(rb *reorderBuffer, dst []byte, i int) ([]byte, int) {
...@@ -157,23 +165,23 @@ func (f Form) Append(out []byte, src ...byte) []byte { ...@@ -157,23 +165,23 @@ func (f Form) Append(out []byte, src ...byte) []byte {
} }
rb := reorderBuffer{} rb := reorderBuffer{}
rb.init(f, src) rb.init(f, src)
return doAppend(&rb, out) return doAppend(&rb, out, 0)
} }
func doAppend(rb *reorderBuffer, out []byte) []byte { func doAppend(rb *reorderBuffer, out []byte, p int) []byte {
src, n := rb.src, rb.nsrc src, n := rb.src, rb.nsrc
doMerge := len(out) > 0 doMerge := len(out) > 0
p := 0 if q := src.skipNonStarter(p); q > p {
if p = src.skipNonStarter(); p > 0 {
// Move leading non-starters to destination. // Move leading non-starters to destination.
out = src.appendSlice(out, 0, p) out = src.appendSlice(out, p, q)
buf, ndropped := patchTail(rb, out) buf, endsInError := patchTail(rb, out)
if ndropped > 0 { if endsInError {
out = src.appendSlice(buf, p-ndropped, p) out = buf
doMerge = false // no need to merge, ends with illegal UTF-8 doMerge = false // no need to merge, ends with illegal UTF-8
} else { } else {
out = decomposeToLastBoundary(rb, buf) // force decomposition out = decomposeToLastBoundary(rb, buf) // force decomposition
} }
p = q
} }
fd := &rb.f fd := &rb.f
if doMerge { if doMerge {
...@@ -217,7 +225,7 @@ func (f Form) AppendString(out []byte, src string) []byte { ...@@ -217,7 +225,7 @@ func (f Form) AppendString(out []byte, src string) []byte {
} }
rb := reorderBuffer{} rb := reorderBuffer{}
rb.initString(f, src) rb.initString(f, src)
return doAppend(&rb, out) return doAppend(&rb, out, 0)
} }
// QuickSpan returns a boundary n such that b[0:n] == f(b[0:n]). // QuickSpan returns a boundary n such that b[0:n] == f(b[0:n]).
...@@ -225,7 +233,8 @@ func (f Form) AppendString(out []byte, src string) []byte { ...@@ -225,7 +233,8 @@ func (f Form) AppendString(out []byte, src string) []byte {
func (f Form) QuickSpan(b []byte) int { func (f Form) QuickSpan(b []byte) int {
rb := reorderBuffer{} rb := reorderBuffer{}
rb.init(f, b) rb.init(f, b)
return quickSpan(&rb, 0) n := quickSpan(&rb, 0)
return n
} }
func quickSpan(rb *reorderBuffer, i int) int { func quickSpan(rb *reorderBuffer, i int) int {
...@@ -301,7 +310,7 @@ func (f Form) FirstBoundary(b []byte) int { ...@@ -301,7 +310,7 @@ func (f Form) FirstBoundary(b []byte) int {
func firstBoundary(rb *reorderBuffer) int { func firstBoundary(rb *reorderBuffer) int {
src, nsrc := rb.src, rb.nsrc src, nsrc := rb.src, rb.nsrc
i := src.skipNonStarter() i := src.skipNonStarter(0)
if i >= nsrc { if i >= nsrc {
return -1 return -1
} }
......
...@@ -253,7 +253,7 @@ var quickSpanNFDTests = []PositionTest{ ...@@ -253,7 +253,7 @@ var quickSpanNFDTests = []PositionTest{
{"\u0316\u0300cd", 6, ""}, {"\u0316\u0300cd", 6, ""},
{"\u043E\u0308b", 5, ""}, {"\u043E\u0308b", 5, ""},
// incorrectly ordered combining characters // incorrectly ordered combining characters
{"ab\u0300\u0316", 1, ""}, // TODO(mpvl): we could skip 'b' as well. {"ab\u0300\u0316", 1, ""}, // TODO: we could skip 'b' as well.
{"ab\u0300\u0316cd", 1, ""}, {"ab\u0300\u0316cd", 1, ""},
// Hangul // Hangul
{"같은", 0, ""}, {"같은", 0, ""},
...@@ -465,6 +465,7 @@ var appendTests = []AppendTest{ ...@@ -465,6 +465,7 @@ var appendTests = []AppendTest{
{"\u0300", "\xFC\x80\x80\x80\x80\x80\u0300", "\u0300\xFC\x80\x80\x80\x80\x80\u0300"}, {"\u0300", "\xFC\x80\x80\x80\x80\x80\u0300", "\u0300\xFC\x80\x80\x80\x80\x80\u0300"},
{"\xF8\x80\x80\x80\x80\u0300", "\u0300", "\xF8\x80\x80\x80\x80\u0300\u0300"}, {"\xF8\x80\x80\x80\x80\u0300", "\u0300", "\xF8\x80\x80\x80\x80\u0300\u0300"},
{"\xFC\x80\x80\x80\x80\x80\u0300", "\u0300", "\xFC\x80\x80\x80\x80\x80\u0300\u0300"}, {"\xFC\x80\x80\x80\x80\x80\u0300", "\u0300", "\xFC\x80\x80\x80\x80\x80\u0300\u0300"},
{"\xF8\x80\x80\x80", "\x80\u0300\u0300", "\xF8\x80\x80\x80\x80\u0300\u0300"},
} }
func appendF(f Form, out []byte, s string) []byte { func appendF(f Form, out []byte, s string) []byte {
...@@ -475,9 +476,23 @@ func appendStringF(f Form, out []byte, s string) []byte { ...@@ -475,9 +476,23 @@ func appendStringF(f Form, out []byte, s string) []byte {
return f.AppendString(out, s) return f.AppendString(out, s)
} }
func bytesF(f Form, out []byte, s string) []byte {
buf := []byte{}
buf = append(buf, out...)
buf = append(buf, s...)
return f.Bytes(buf)
}
func stringF(f Form, out []byte, s string) []byte {
outs := string(out) + s
return []byte(f.String(outs))
}
func TestAppend(t *testing.T) { func TestAppend(t *testing.T) {
runAppendTests(t, "TestAppend", NFKC, appendF, appendTests) runAppendTests(t, "TestAppend", NFKC, appendF, appendTests)
runAppendTests(t, "TestAppendString", NFKC, appendStringF, appendTests) runAppendTests(t, "TestAppendString", NFKC, appendStringF, appendTests)
runAppendTests(t, "TestBytes", NFKC, bytesF, appendTests)
runAppendTests(t, "TestString", NFKC, stringF, appendTests)
} }
func doFormBenchmark(b *testing.B, f Form, s string) { func doFormBenchmark(b *testing.B, f Form, s string) {
......
...@@ -27,7 +27,7 @@ func (w *normWriter) Write(data []byte) (n int, err error) { ...@@ -27,7 +27,7 @@ func (w *normWriter) Write(data []byte) (n int, err error) {
} }
w.rb.src = inputBytes(data[:m]) w.rb.src = inputBytes(data[:m])
w.rb.nsrc = m w.rb.nsrc = m
w.buf = doAppend(&w.rb, w.buf) w.buf = doAppend(&w.rb, w.buf, 0)
data = data[m:] data = data[m:]
n += m n += m
...@@ -101,7 +101,7 @@ func (r *normReader) Read(p []byte) (int, error) { ...@@ -101,7 +101,7 @@ func (r *normReader) Read(p []byte) (int, error) {
r.rb.src = inputBytes(r.inbuf[0:n]) r.rb.src = inputBytes(r.inbuf[0:n])
r.rb.nsrc, r.err = n, err r.rb.nsrc, r.err = n, err
if n > 0 { if n > 0 {
r.outbuf = doAppend(&r.rb, r.outbuf) r.outbuf = doAppend(&r.rb, r.outbuf, 0)
} }
if err == io.EOF { if err == io.EOF {
r.lastBoundary = len(r.outbuf) r.lastBoundary = len(r.outbuf)
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment