8000 Issue #14738: Speed-up UTF-8 decoding on non-ASCII data. Patch by Se… · python/cpython@ca5f91b · GitHub
[go: up one dir, main page]
Skip to content

Commit ca5f91b

Browse files
pitroupitrou
committed
Issue #14738: Speed-up UTF-8 decoding on non-ASCII data. Patch by Serhiy Storchaka.
1 parent fda08b0 commit ca5f91b

File tree

8 files changed

+316
-552
lines changed

8 files changed

+316
-552
lines changed

‎Misc/NEWS‎

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -10,6 +10,9 @@ What's New in Python 3.3.0 Alpha 4?
1010
Core and Builtins
1111
-----------------
1212

13+
- Issue #14738: Speed-up UTF-8 decoding on non-ASCII data. Patch by Serhiy
14+
Storchaka.
15+
1316
- Issue #14700: Fix two broken and undefined-behaviour-inducing overflow checks
1417
in old-style string formatting.
1518

‎Objects/stringlib/asciilib.h‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -7,6 +7,7 @@
77
#define STRINGLIB(F) asciilib_##F
88
#define STRINGLIB_OBJECT PyUnicodeObject
99
#define STRINGLIB_SIZEOF_CHAR 1
10+
#define STRINGLIB_MAX_CHAR 0x7Fu
1011
#define STRINGLIB_CHAR Py_UCS1
1112
#define STRINGLIB_TYPE_NAME "unicode"
1213
#define STRINGLIB_PARSE_CODE "U"

‎Objects/stringlib/codecs.h‎

Lines changed: 143 additions & 78 deletions
Original file line numberDiff line numberDiff line change
@@ -15,19 +15,18 @@
1515
# error C 'long' size should be either 4 or 8!
1616
#endif
1717

18-
Py_LOCAL_INLINE(int)
19-
STRINGLIB(utf8_try_decode)(const char *start, const char *end,
20-
STRINGLIB_CHAR *dest,
21-
const char **src_pos, Py_ssize_t *dest_index)
18+
Py_LOCAL_INLINE(Py_UCS4)
19+
STRINGLIB(utf8_decode)(const char **inptr, const char *end,
20+
STRINGLIB_CHAR *dest,
21+
Py_ssize_t *outpos)
2222
{
23-
int ret;
24-
Py_ssize_t n;
25-
const char *s = start;
23+
Py_UCS4 ch;
24+
const char *s = *inptr;
2625
const char *aligned_end = (const char *) ((size_t) end & ~LONG_PTR_MASK);
27-
STRINGLIB_CHAR *p = dest;
26+
STRINGLIB_CHAR *p = dest + *outpos;
2827

2928
while (s < end) {
30-
Py_UCS4 ch = (unsigned char)*s;
29+
ch = (unsigned char)*s;
3130

3231
if (ch < 0x80) {
3332
/* Fast path for runs of ASCII characters. Given that common UTF-8
@@ -48,15 +47,33 @@ STRINGLIB(utf8_try_decode)(const char *start, const char *end,
4847
unsigned long value = *(unsigned long *) _s;
4948
if (value & ASCII_CHAR_MASK)
5049
break;
51-
_p[0] = _s[0];
52-
_p[1] = _s[1];
53-
_p[2] = _s[2];
54-
_p[3] = _s[3];
55-
#if (SIZEOF_LONG == 8)
56-
_p[4] = _s[4];
57-
_p[5] = _s[5];
58-
_p[6] = _s[6];
59-
_p[7] = _s[7];
50+
#ifdef BYTEORDER_IS_LITTLE_ENDIAN
51+
_p[0] = (STRINGLIB_CHAR)(value & 0xFFu);
52+
_p[1] = (STRINGLIB_CHAR)((value >> 8) & 0xFFu);
53+
_p[2] = (STRINGLIB_CHAR)((value >> 16) & 0xFFu);
54+
_p[3] = (STRINGLIB_CHAR)((value >> 24) & 0xFFu);
55+
# if SIZEOF_LONG == 8
56+
_p[4] = (STRINGLIB_CHAR)((value >> 32) & 0xFFu);
57+
_p[5] = (STRINGLIB_CHAR)((value >> 40) & 0xFFu);
58+
_p[6] = (STRINGLIB_CHAR)((value >> 48) & 0xFFu);
59+
_p[7] = (STRINGLIB_CHAR)((value >> 56) & 0xFFu);
60+
# endif
61+
#else
62+
# if SIZEOF_LONG == 8
63+
_p[0] = (STRINGLIB_CHAR)((value >> 56) & 0xFFu);
64+
_p[1] = (STRINGLIB_CHAR)((value >> 48) & 0xFFu);
65+
_p[2] = (STRINGLIB_CHAR)((value >> 40) & 0xFFu);
66+
_p[3] = (STRINGLIB_CHAR)((value >> 32) & 0xFFu);
67+
_p[4] = (STRINGLIB_CHAR)((value >> 24) & 0xFFu);
68+
_p[5] = (STRINGLIB_CHAR)((value >> 16) & 0xFFu);
69+
_p[6] = (STRINGLIB_CHAR)((value >> 8) & 0xFFu);
70+
_p[7] = (STRINGLIB_CHAR)(value & 0xFFu);
71+
# else
72+
_p[0] = (STRINGLIB_CHAR)((value >> 24) & 0xFFu);
73+
_p[1] = (STRINGLIB_CHAR)((value >> 16) & 0xFFu);
74+
_p[2] = (STRINGLIB_CHAR)((value >> 8) & 0xFFu);
75+
_p[3] = (STRINGLIB_CHAR)(value & 0xFFu);
76+
# endif
6077
#endif
6178
_s += SIZEOF_LONG;
6279
_p += SIZEOF_LONG;
@@ -67,87 +84,135 @@ STRINGLIB(utf8_try_decode)(const char *start, const char *end,
6784
break;
6885
ch = (unsigned char)*s;
6986
}
87+
if (ch < 0x80) {
88+
s++;
89+
*p++ = ch;
90+
continue;
91+
}
7092
}
7193

72-
if (ch < 0x80) {
73-
s++;
74-
*p++ = ch;
75-
continue;
76-
}
77-
78-
n = utf8_code_length[ch];
79-
80-
if (s + n > end) {
81-
/* unexpected end of data: the caller will decide whether
82-
it's an error or not */
83-
goto _error;
94+
if (ch < 0xC2) {
95+
/* invalid sequence
96+
\x80-\xBF -- continuation byte
97+
\xC0-\xC1 -- fake 0000-007F */
98+
goto InvalidStart;
8499
}
85100

86-
switch (n) {
87-
case 0:
88-
/* invalid start byte */
89-
goto _error;
90-
case 1:
91-
/* internal error */
92-
goto _error;
93-
case 2:
94-
if ((s[1] & 0xc0) != 0x80)
101+
if (ch < 0xE0) {
102+
/* \xC2\x80-\xDF\xBF -- 0080-07FF */
103+
Py_UCS4 ch2;
104+
if (end - s < 2) {
105+
/* unexpected end of data: the caller will decide whether
106+
it's an error or not */
107+
break;
108+
}
109+
ch2 = (unsigned char)s[1];
110+
if ((ch2 & 0xC0) != 0x80)
95111
/* invalid continuation byte */
96-
goto _error;
97-
ch = ((s[0] & 0x1f) << 6) + (s[1] & 0x3f);
112+
goto InvalidContinuation;
113+
ch = (ch << 6) + ch2 -
114+
((0xC0 << 6) + 0x80);
98115
assert ((ch > 0x007F) && (ch <= 0x07FF));
99116
s += 2;
117+
if (STRINGLIB_MAX_CHAR <= 0x007F ||
118+
(STRINGLIB_MAX_CHAR < 0x07FF && ch > STRINGLIB_MAX_CHAR))
119+
goto Overflow;
100120
*p++ = ch;
101-
break;
121+
continue;
122+
}
102123

103-
case 3:
104-
/* Decoding UTF-8 sequences in range \xed\xa0\x80-\xed\xbf\xbf
105-
will result in surrogates in range d800-dfff. Surrogates are
106-
not valid UTF-8 so they are rejected.
107-
See http://www.unicode.org/versions/Unicode5.2.0/ch03.pdf
108-
(table 3-7) and http://www.rfc-editor.org/rfc/rfc3629.txt */
109-
if ((s[1] & 0xc0) != 0x80 ||
110-
(s[2] & 0xc0) != 0x80 ||
111-
((unsigned char)s[0] == 0xE0 &&
112-
(unsigned char)s[1] < 0xA0) ||
113-
((unsigned char)s[0] == 0xED &&
114-
(unsigned char)s[1] > 0x9F)) {
124+
if (ch < 0xF0) {
125+
/* \xE0\xA0\x80-\xEF\xBF\xBF -- 0800-FFFF */
126+
Py_UCS4 ch2, ch3;
127+
if (end - s < 3) {
128+
/* unexpected end of data: the caller will decide whether
129+
it's an error or not */
130+
break;
131+
}
132+
ch2 = (unsigned char)s[1];
133+
ch3 = (unsigned char)s[2];
134+
if ((ch2 & 0xC0) != 0x80 ||
135+
(ch3 & 0xC0) != 0x80) {
115136
/* invalid continuation byte */
116-
goto _error;
137+
goto InvalidContinuation;
138+
}
139+
if (ch == 0xE0) {
140+
if (ch2 < 0xA0)
141+
/* invalid sequence
142+
\xE0\x80\x80-\xE0\x9F\xBF -- fake 0000-0800 */
143+
goto InvalidContinuation;
117144
}
118-
ch = ((s[0] & 0x0f) << 12) + ((s[1] & 0x3f) << 6) + (s[2] & 0x3f);
145+
else if (ch == 0xED && ch2 > 0x9F) {
146+
/* Decoding UTF-8 sequences in range \xED\xA0\x80-\xED\xBF\xBF
147+
will result in surrogates in range D800-DFFF. Surrogates are
148+
not valid UTF-8 so they are rejected.
149+
See http://www.unicode.org/versions/Unicode5.2.0/ch03.pdf
150+
(table 3-7) and http://www.rfc-editor.org/rfc/rfc3629.txt */
151+
goto InvalidContinuation;
152+
}
153+
ch = (ch << 12) + (ch2 << 6) + ch3 -
154+
((0xE0 << 12) + (0x80 << 6) + 0x80);
119155
assert ((ch > 0x07FF) && (ch <= 0xFFFF));
120156
s += 3;
157+
if (STRINGLIB_MAX_CHAR <= 0x07FF ||
158+
(STRINGLIB_MAX_CHAR < 0xFFFF && ch > STRINGLIB_MAX_CHAR))
159+
goto Overflow;
121160
*p++ = ch;
122-
break;
161+
continue;
162+
}
123163

124-
case 4:
125-
if ((s[1] & 0xc0) != 0x80 ||
126-
(s[2] & 0xc0) != 0x80 ||
127-
(s[3] & 0xc0) != 0x80 ||
128-
((unsigned char)s[0] == 0xF0 &&
129-
(unsigned char)s[1] < 0x90) ||
130-
((unsigned char)s[0] == 0xF4 &&
131-
(unsigned char)s[1] > 0x8F)) {
164+
if (ch < 0xF5) {
165+
/* \xF0\x90\x80\x80-\xF4\x8F\xBF\xBF -- 10000-10FFFF */
166+
Py_UCS4 ch2, ch3, ch4;
167+
if (end - s < 4) {
168+
/* unexpected end of data: the caller will decide whether
169+
it's an error or not */
170+
break;
171+
}
172+
ch2 = (unsigned char)s[1];
173+
ch3 = (unsigned char)s[2];
174+
ch4 = (unsigned char)s[3];
175+
if ((ch2 & 0xC0) != 0x80 ||
176+
(ch3 & 0xC0) != 0x80 ||
177+
(ch4 & 0xC0) != 0x80) {
132178
/* invalid continuation byte */
133-
goto _error;
179+
goto InvalidContinuation;
180+
}
181+
if (ch == 0xF0) {
182+
if (ch2 < 0x90)
183+
/* invalid sequence
184+
\xF0\x80\x80\x80-\xF0\x80\xBF\xBF -- fake 0000-FFFF */
185+
goto InvalidContinuation;
134186
}
135-
ch = ((s[0] & 0x7) << 18) + ((s[1] & 0x3f) << 12) +
136-
((s[2] & 0x3f) << 6) + (s[3] & 0x3f);
137-
assert ((ch > 0xFFFF) && (ch <= 0x10ffff));
187+
else if (ch == 0xF4 && ch2 > 0x8F) {
188+
/* invalid sequence
189+
\xF4\x90\x80\80- -- 110000- overflow */
190+
goto InvalidContinuation;
191+
}
192+
ch = (ch << 18) + (ch2 << 12) + (ch3 << 6) + ch4 -
193+
((0xF0 << 18) + (0x80 << 12) + (0x80 << 6) + 0x80);
194+
assert ((ch > 0xFFFF) && (ch <= 0x10FFFF));
138195
s += 4;
196+
if (STRINGLIB_MAX_CHAR <= 0xFFFF ||
197+
(STRINGLIB_MAX_CHAR < 0x10FFFF && ch > STRINGLIB_MAX_CHAR))
198+
goto Overflow;
139199
*p++ = ch;
140-
break;
200+
continue;
141201
}
202+
goto InvalidStart;
142203
}
143-
ret = 0;
144-
goto _ok;
145-
_error:
146-
ret = -1;
147-
_ok:
148-
*src_pos = s;
149-
*dest_index = p - dest;
150-
return ret;
204+
ch = 0;
205+
Overflow:
206+
Return:
207+
*inptr = s;
208+
*outpos = p - dest;
209+
return ch;
210+
InvalidStart:
211+
ch = 1;
212+
goto Return;
213+
InvalidContinuation:
214+
ch = 2;
215+
goto Return;
151216
}
152217

153218
#undef LONG_PTR_MASK

‎Objects/stringlib/ucs1lib.h‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -7,6 +7,7 @@
77< D963 code class="diff-text syntax-highlighted-line">
#define STRINGLIB(F) ucs1lib_##F
88
#define STRINGLIB_OBJECT PyUnicodeObject
99
#define STRINGLIB_SIZEOF_CHAR 1
10+
#define STRINGLIB_MAX_CHAR 0xFFu
1011
#define STRINGLIB_CHAR Py_UCS1
1112
#define STRINGLIB_TYPE_NAME "unicode"
1213
#define STRINGLIB_PARSE_CODE "U"

‎Objects/stringlib/ucs2lib.h‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -7,6 +7,7 @@
77
#define STRINGLIB(F) ucs2lib_##F
88
#define STRINGLIB_OBJECT PyUnicodeObject
99
#define STRINGLIB_SIZEOF_CHAR 2
10+
#define STRINGLIB_MAX_CHAR 0xFFFFu
1011
#define STRINGLIB_CHAR Py_UCS2
1112
#define STRINGLIB_TYPE_NAME "unicode"
1213
#define STRINGLIB_PARSE_CODE "U"

‎Objects/stringlib/ucs4lib.h‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -7,6 +7,7 @@
77
#define STRINGLIB(F) ucs4lib_##F
88
#define STRINGLIB_OBJECT PyUnicodeObject
99
#define STRINGLIB_SIZEOF_CHAR 4
10+
#define STRINGLIB_MAX_CHAR 0x10FFFFu
1011
#define STRINGLIB_CHAR Py_UCS4
1112
#define STRINGLIB_TYPE_NAME "unicode"
1213
#define STRINGLIB_PARSE_CODE "U"

‎Objects/stringlib/undef.h‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,6 +1,7 @@
11
#undef FASTSEARCH
22
#undef STRINGLIB
33
#undef STRINGLIB_SIZEOF_CHAR
4+
#undef STRINGLIB_MAX_CHAR
45
#undef STRINGLIB_CHAR
56
#undef STRINGLIB_STR
67
#undef STRINGLIB_LEN

0 commit comments

Comments
 (0)
0