patch 9.2.0145: UTF-8 decoding and length calculation can be improved
Commit:
https://github.com/vim/vim/commit/55464b5d18b5ab8a8f9821949315e2167a9df148
Author: Yasuhiro Matsumoto <[email protected]>
Date: Thu Mar 12 19:15:28 2026 +0000
patch 9.2.0145: UTF-8 decoding and length calculation can be improved
Problem: Vim often calls utf_ptr2char() and utf_ptr2len() separately.
Solution: Refactor UTF-8 hot paths into utf_ptr2char_and_len() to
decode the codepoint and byte length in a single pass.
Fold combining character logic into the same optimized flow.
Improves redraw performance by ~8-10% in UTF-8 heavy
scenarios (Yasuhiro Matsumoto).
closes: #19649
Signed-off-by: Yasuhiro Matsumoto <[email protected]>
Signed-off-by: Christian Brabandt <[email protected]>
diff --git a/src/mbyte.c b/src/mbyte.c
index 41571f476..8370c3fa2 100644
--- a/src/mbyte.c
+++ b/src/mbyte.c
@@ -129,6 +129,9 @@ static int dbcs_char2cells(int c);
static int dbcs_ptr2cells_len(char_u *p, int size);
static int dbcs_ptr2char(char_u *p);
static int dbcs_head_off(char_u *base, char_u *p);
+static inline int utf_ptr2char_and_len(char_u *p, int *lenp);
+static inline int utf_ptr2char_and_len_len(char_u *p, int size, int *lenp);
+static inline int utf_iscomposinglike_char(int c1, int c2);
#ifdef FEAT_EVAL
static int cw_value(int c);
#endif
@@ -1629,13 +1632,14 @@ utf_ptr2cells(
char_u *p)
{
int c;
+ int len;
// Need to convert to a character number.
if (*p >= 0x80)
{
- c = utf_ptr2char(p);
+ c = utf_ptr2char_and_len(p, &len);
// An illegal byte is displayed as <xx>.
- if (utf_ptr2len(p) == 1 || c == NUL)
+ if (len == 1 || c == NUL)
return 4;
// If the char is ASCII it must be an overlong sequence.
if (c < 0x80)
@@ -1670,15 +1674,16 @@ latin_ptr2cells_len(char_u *p UNUSED, int size UNUSED)
utf_ptr2cells_len(char_u *p, int size)
{
int c;
+ int len;
// Need to convert to a wide character.
if (size > 0 && *p >= 0x80)
{
- if (utf_ptr2len_len(p, size) < utf8len_tab[*p])
+ c = utf_ptr2char_and_len_len(p, size, &len);
+ if (len > size)
return 1; // truncated
- c = utf_ptr2char(p);
// An illegal byte is displayed as <xx>.
- if (utf_ptr2len(p) == 1 || c == NUL)
+ if (len == 1 || c == NUL)
return 4;
// If the char is ASCII it must be an overlong sequence.
if (c < 0x80)
@@ -1786,50 +1791,219 @@ dbcs_ptr2char(char_u *p)
/*
* Convert a UTF-8 byte sequence to a character number.
- * If the sequence is illegal or truncated by a NUL the first byte is
- * returned.
- * For an overlong sequence this may return zero.
- * Does not include composing characters, of course.
+ * Returns the character and sets "*lenp" to its byte length.
+ * Illegal bytes are returned as-is with a length of one.
*/
- int
-utf_ptr2char(char_u *p)
+ static inline int
+utf_ptr2char_and_len(char_u *p, int *lenp)
{
int len;
+ int c;
- if (p[0] < 0x80) // be quick for ASCII
+ if (p[0] < 0x80)
+ {
+ *lenp = p[0] == NUL ? 0 : 1;
return p[0];
+ }
len = utf8len_tab_zero[p[0]];
- if (len > 1 && (p[1] & 0xc0) == 0x80)
+ if (len <= 1 || (p[1] & 0xc0) != 0x80)
{
- if (len == 2)
- return ((p[0] & 0x1f) << 6) + (p[1] & 0x3f);
- if ((p[2] & 0xc0) == 0x80)
- {
- if (len == 3)
- return ((p[0] & 0x0f) << 12) + ((p[1] & 0x3f) << 6)
- + (p[2] & 0x3f);
- if ((p[3] & 0xc0) == 0x80)
+ *lenp = 1;
+ return p[0];
+ }
+
+ c = ((p[0] & 0x1f) << 6) + (p[1] & 0x3f);
+ if (len == 2)
+ {
+ *lenp = 2;
+ return c;
+ }
+ if ((p[2] & 0xc0) != 0x80)
+ {
+ *lenp = 1;
+ return p[0];
+ }
+
+ c = ((p[0] & 0x0f) << 12) + ((p[1] & 0x3f) << 6) + (p[2] & 0x3f);
+ if (len == 3)
+ {
+ *lenp = 3;
+ return c;
+ }
+ if ((p[3] & 0xc0) != 0x80)
+ {
+ *lenp = 1;
+ return p[0];
+ }
+
+ c = ((p[0] & 0x07) << 18) + ((p[1] & 0x3f) << 12)
+ + ((p[2] & 0x3f) << 6) + (p[3] & 0x3f);
+ if (len == 4)
+ {
+ *lenp = 4;
+ return c;
+ }
+ if ((p[4] & 0xc0) != 0x80)
+ {
+ *lenp = 1;
+ return p[0];
+ }
+
+ c = ((p[0] & 0x03) << 24) + ((p[1] & 0x3f) << 18)
+ + ((p[2] & 0x3f) << 12)
+ + ((p[3] & 0x3f) << 6) + (p[4] & 0x3f);
+ if (len == 5)
+ {
+ *lenp = 5;
+ return c;
+ }
+ if ((p[5] & 0xc0) != 0x80)
+ {
+ *lenp = 1;
+ return p[0];
+ }
+
+ *lenp = 6;
+ return ((p[0] & 0x01) << 30) + ((p[1] & 0x3f) << 24)
+ + ((p[2] & 0x3f) << 18)
+ + ((p[3] & 0x3f) << 12)
+ + ((p[4] & 0x3f) << 6) + (p[5] & 0x3f);
+}
+
+/*
+ * Like utf_ptr2char_and_len(), but never reads beyond "size" bytes.
+ * For an incomplete sequence "*lenp" is set to the expected length.
+ */
+ static inline int
+utf_ptr2char_and_len_len(char_u *p, int size, int *lenp)
+{
+ int len;
+ int c;
+
+ if (size < 1)
+ {
+ *lenp = 1;
+ return NUL;
+ }
+ if (p[0] < 0x80)
+ {
+ *lenp = 1;
+ return p[0];
+ }
+
+ len = utf8len_tab_zero[p[0]];
+ if (len <= 1)
+ {
+ *lenp = 1;
+ return p[0];
+ }
+ if (len > size)
+ {
+ int i;
+
+ // Incomplete sequence: validate continuation bytes within range.
+ for (i = 1; i < size; ++i)
+ if ((p[i] & 0xc0) != 0x80)
{
- if (len == 4)
- return ((p[0] & 0x07) << 18) + ((p[1] & 0x3f) << 12)
- + ((p[2] & 0x3f) << 6) + (p[3] & 0x3f);
- if ((p[4] & 0xc0) == 0x80)
- {
- if (len == 5)
- return ((p[0] & 0x03) << 24) + ((p[1] & 0x3f) << 18)
- + ((p[2] & 0x3f) << 12) + ((p[3] & 0x3f) << 6)
- + (p[4] & 0x3f);
- if ((p[5] & 0xc0) == 0x80 && len == 6)
- return ((p[0] & 0x01) << 30) + ((p[1] & 0x3f) << 24)
- + ((p[2] & 0x3f) << 18) + ((p[3] & 0x3f) << 12)
- + ((p[4] & 0x3f) << 6) + (p[5] & 0x3f);
- }
+ *lenp = 1;
+ return p[0];
}
- }
+ *lenp = len;
+ return p[0];
+ }
+ if ((p[1] & 0xc0) != 0x80)
+ {
+ *lenp = 1;
+ return p[0];
+ }
+
+ c = ((p[0] & 0x1f) << 6) + (p[1] & 0x3f);
+ if (len == 2)
+ {
+ *lenp = 2;
+ return c;
+ }
+ if ((p[2] & 0xc0) != 0x80)
+ {
+ *lenp = 1;
+ return p[0];
}
- // Illegal value, just return the first byte
- return p[0];
+
+ c = ((p[0] & 0x0f) << 12) + ((p[1] & 0x3f) << 6) + (p[2] & 0x3f);
+ if (len == 3)
+ {
+ *lenp = 3;
+ return c;
+ }
+ if ((p[3] & 0xc0) != 0x80)
+ {
+ *lenp = 1;
+ return p[0];
+ }
+
+ c = ((p[0] & 0x07) << 18) + ((p[1] & 0x3f) << 12)
+ + ((p[2] & 0x3f) << 6) + (p[3] & 0x3f);
+ if (len == 4)
+ {
+ *lenp = 4;
+ return c;
+ }
+ if ((p[4] & 0xc0) != 0x80)
+ {
+ *lenp = 1;
+ return p[0];
+ }
+
+ c = ((p[0] & 0x03) << 24) + ((p[1] & 0x3f) << 18)
+ + ((p[2] & 0x3f) << 12)
+ + ((p[3] & 0x3f) << 6) + (p[4] & 0x3f);
+ if (len == 5)
+ {
+ *lenp = 5;
+ return c;
+ }
+ if ((p[5] & 0xc0) != 0x80)
+ {
+ *lenp = 1;
+ return p[0];
+ }
+
+ *lenp = 6;
+ return ((p[0] & 0x01) << 30) + ((p[1] & 0x3f) << 24)
+ + ((p[2] & 0x3f) << 18)
+ + ((p[3] & 0x3f) << 12)
+ + ((p[4] & 0x3f) << 6) + (p[5] & 0x3f);
+}
+
+ static inline int
+utf_iscomposinglike_char(int c1, int c2)
+{
+ if (utf_iscomposing(c2))
+ return TRUE;
+#ifdef FEAT_ARABIC
+ if (!arabic_maycombine(c2))
+ return FALSE;
+ return arabic_combine(c1, c2);
+#else
+ (void)c1;
+ return FALSE;
+#endif
+}
+
+/*
+ * Convert a UTF-8 byte sequence to a character number.
+ * If the sequence is illegal or truncated by a NUL the first byte is
+ * returned.
+ * For an overlong sequence this may return zero.
+ * Does not include composing characters, of course.
+ */
+ int
+utf_ptr2char(char_u *p)
+{
+ int len;
+
+ return utf_ptr2char_and_len(p, &len);
}
/*
@@ -1953,25 +2127,29 @@ utfc_ptr2char(
int len;
int c;
int cc;
+ int cc_len;
int i = 0;
- c = utf_ptr2char(p);
- len = utf_ptr2len(p);
+ c = utf_ptr2char_and_len(p, &len);
// Only accept a composing char when the first char isn't illegal.
- if ((len > 1 || *p < 0x80)
- && p[len] >= 0x80
- && UTF_COMPOSINGLIKE(p, p + len))
+ if ((len > 1 || *p < 0x80) && p[len] >= 0x80)
{
- cc = utf_ptr2char(p + len);
- for (;;)
+ cc = utf_ptr2char_and_len(p + len, &cc_len);
+ if (utf_iscomposinglike_char(c, cc))
{
- pcc[i++] = cc;
- if (i == MAX_MCO)
- break;
- len += utf_ptr2len(p + len);
- if (p[len] < 0x80 || !utf_iscomposing(cc = utf_ptr2char(p + len)))
- break;
+ for (;;)
+ {
+ pcc[i++] = cc;
+ if (i == MAX_MCO)
+ break;
+ len += cc_len;
+ if (p[len] < 0x80)
+ break;
+ cc = utf_ptr2char_and_len(p + len, &cc_len);
+ if (!utf_iscomposing(cc))
+ break;
+ }
}
}
@@ -1994,27 +2172,28 @@ utfc_ptr2char_len(
int len;
int c;
int cc;
+ int cc_len;
int i = 0;
- c = utf_ptr2char(p);
- len = utf_ptr2len_len(p, maxlen);
+ c = utf_ptr2char_and_len_len(p, maxlen, &len);
// Only accept a composing char when the first char isn't illegal.
- if ((len > 1 || *p < 0x80)
- && len < maxlen
- && p[len] >= 0x80
- && UTF_COMPOSINGLIKE(p, p + len))
+ if ((len > 1 || *p < 0x80) && len < maxlen && p[len] >= 0x80)
{
- cc = utf_ptr2char(p + len);
- for (;;)
+ cc = utf_ptr2char_and_len_len(p + len, maxlen - len, &cc_len);
+ if (cc_len <= maxlen - len && utf_iscomposinglike_char(c, cc))
{
- pcc[i++] = cc;
- if (i == MAX_MCO)
- break;
- len += utf_ptr2len_len(p + len, maxlen - len);
- if (len >= maxlen
- || p[len] < 0x80
- || !utf_iscomposing(cc = utf_ptr2char(p + len)))
- break;
+ for (;;)
+ {
+ pcc[i++] = cc;
+ if (i == MAX_MCO)
+ break;
+ len += cc_len;
+ if (len >= maxlen || p[len] < 0x80)
+ break;
+ cc = utf_ptr2char_and_len_len(p + len, maxlen - len, &cc_len);
+ if (cc_len > maxlen - len || !utf_iscomposing(cc))
+ break;
+ }
}
}
@@ -2057,14 +2236,8 @@ utfc_char2bytes(int off, char_u *buf)
utf_ptr2len(char_u *p)
{
int len;
- int i;
- if (*p == NUL)
- return 0;
- len = utf8len_tab[*p];
- for (i = 1; i < len; ++i)
- if ((p[i] & 0xc0) != 0x80)
- return 1;
+ utf_ptr2char_and_len(p, &len);
return len;
}
@@ -2102,19 +2275,8 @@ utf_byte2len_zero(int b)
utf_ptr2len_len(char_u *p, int size)
{
int len;
- int i;
- int m;
- len = utf8len_tab[*p];
- if (len == 1)
- return 1; // NUL, ascii or illegal lead byte
- if (len > size)
- m = size; // incomplete byte sequence.
- else
- m = len;
- for (i = 1; i < m; ++i)
- if ((p[i] & 0xc0) != 0x80)
- return 1;
+ utf_ptr2char_and_len_len(p, size, &len);
return len;
}
@@ -2127,10 +2289,10 @@ utf_ptr2len_len(char_u *p, int size)
utfc_ptr2len(char_u *p)
{
int len;
+ int c;
+ int cc;
+ int cc_len;
int b0 = *p;
-#ifdef FEAT_ARABIC
- int prevlen;
-#endif
if (b0 == NUL)
return 0;
@@ -2138,7 +2300,7 @@ utfc_ptr2len(char_u *p)
return 1;
// Skip over first UTF-8 char, stopping at a NUL byte.
- len = utf_ptr2len(p);
+ c = utf_ptr2char_and_len(p, &len);
// Check for illegal byte.
if (len == 1 && b0 >= 0x80)
@@ -2148,19 +2310,22 @@ utfc_ptr2len(char_u *p)
* Check for composing characters. We can handle only the first six, but
* skip all of them (otherwise the cursor would get stuck).
*/
-#ifdef FEAT_ARABIC
- prevlen = 0;
-#endif
+ if (p[len] < 0x80)
+ return len;
+
+ cc = utf_ptr2char_and_len(p + len, &cc_len);
+ if (!utf_iscomposinglike_char(c, cc))
+ return len;
+
for (;;)
{
- if (p[len] < 0x80 || !UTF_COMPOSINGLIKE(p + prevlen, p + len))
- return len;
-
// Skip over composing char
-#ifdef FEAT_ARABIC
- prevlen = len;
-#endif
- len += utf_ptr2len(p + len);
+ len += cc_len;
+ if (p[len] < 0x80)
+ return len;
+ cc = utf_ptr2char_and_len(p + len, &cc_len);
+ if (!utf_iscomposing(cc))
+ return len;
}
}
@@ -2174,9 +2339,9 @@ utfc_ptr2len(char_u *p)
utfc_ptr2len_len(char_u *p, int size)
{
int len;
-#ifdef FEAT_ARABIC
- int prevlen;
-#endif
+ int c;
+ int cc;
+ int cc_len;
if (size < 1 || *p == NUL)
return 0;
@@ -2184,7 +2349,7 @@ utfc_ptr2len_len(char_u *p, int size)
return 1;
// Skip over first UTF-8 char, stopping at a NUL byte.
- len = utf_ptr2len_len(p, size);
+ c = utf_ptr2char_and_len_len(p, size, &len);
// Check for illegal byte and incomplete byte sequence.
if ((len == 1 && p[0] >= 0x80) || len > size)
@@ -2194,32 +2359,24 @@ utfc_ptr2len_len(char_u *p, int size)
* Check for composing characters. We can handle only the first six, but
* skip all of them (otherwise the cursor would get stuck).
*/
-#ifdef FEAT_ARABIC
- prevlen = 0;
-#endif
+ if (len >= size || p[len] < 0x80)
+ return len;
+
+ cc = utf_ptr2char_and_len_len(p + len, size - len, &cc_len);
+ if (cc_len > size - len || !utf_iscomposinglike_char(c, cc))
+ return len;
+
while (len < size)
{
- int len_next_char;
-
- if (p[len] < 0x80)
+ // Skip over composing char
+ len += cc_len;
+ if (len >= size || p[len] < 0x80)
break;
-
- /*
- * Next character length should not go beyond size to ensure that
- * UTF_COMPOSINGLIKE(...) does not read beyond size.
- */
- len_next_char = utf_ptr2len_len(p + len, size - len);
- if (len_next_char > size - len)
+ cc = utf_ptr2char_and_len_len(p + len, size - len, &cc_len);
+ if (cc_len > size - len)
break;
-
- if (!UTF_COMPOSINGLIKE(p + prevlen, p + len))
+ if (!utf_iscomposing(cc))
break;
-
- // Skip over composing char
-#ifdef FEAT_ARABIC
- prevlen = len;
-#endif
- len += len_next_char;
}
return len;
}
diff --git a/src/version.c b/src/version.c
index d539758f5..da639bb06 100644
--- a/src/version.c
+++ b/src/version.c
@@ -734,6 +734,8 @@ static char *(features[]) =
static int included_patches[] =
{ /* Add new patch number below this line */
+/**/
+ 145,
/**/
144,
/**/
--
--
You received this message from the "vim_dev" maillist.
Do not top-post! Type your reply below the text you are replying to.
For more information, visit http://www.vim.org/maillist.php
---
You received this message because you are subscribed to the Google Groups
"vim_dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To view this discussion visit
https://groups.google.com/d/msgid/vim_dev/E1w0ljO-002C7c-UZ%40256bit.org.
