Merge pull request #1 from pagespeed/oschaaf-patch
Apply upstream change
diff --git a/source/common/utext.cpp b/source/common/utext.cpp
index e0cc5b7..70acd42 100644
--- a/source/common/utext.cpp
+++ b/source/common/utext.cpp
@@ -1177,9 +1177,15 @@
//------------------------------------------------------------------------------
// Chunk size.
-// Must be less than 85, because of byte mapping from UChar indexes to native indexes.
-// Worst case is three native bytes to one UChar. (Supplemenaries are 4 native bytes
-// to two UChars.)
+// Must be less than 42 (256/6), because of byte mapping from UChar indexes to native indexes.
+// Worst case there are six UTF-8 bytes per UChar.
+// obsolete 6 byte form fd + 5 trails maps to fffd
+// obsolete 5 byte form fc + 4 trails maps to fffd
+// non-shortest 4 byte forms maps to fffd
+// normal supplementaries map to a pair of utf-16, two utf8 bytes per utf-16 unit
+// mapToUChars array size must allow for the worst case, 6.
+// This could be brought down to 4, by treating fd and fc as pure illegal,
+// rather than obsolete lead bytes. But that is not compatible with the utf-8 access macros.
//
enum { UTF8_TEXT_CHUNK_SIZE=32 };
@@ -1219,7 +1225,7 @@
// Requires two extra slots,
// one for a supplementary starting in the last normal position,
// and one for an entry for the buffer limit position.
- uint8_t mapToUChars[UTF8_TEXT_CHUNK_SIZE*3+6]; // Map native offset from bufNativeStart to
+ uint8_t mapToUChars[UTF8_TEXT_CHUNK_SIZE*6+6]; // Map native offset from bufNativeStart to
// correspoding offset in filled part of buf.
int32_t align;
};
@@ -1362,6 +1368,7 @@
// Requested index is in this buffer.
u8b = (UTF8Buf *)ut->p; // the current buffer
mapIndex = ix - u8b->toUCharsMapStart;
+ U_ASSERT(mapIndex < (int32_t)sizeof(UTF8Buf::mapToUChars));
ut->chunkOffset = u8b->mapToUChars[mapIndex] - u8b->bufStartIdx;
return TRUE;
@@ -1632,6 +1639,10 @@
// Can only do this if the incoming index is somewhere in the interior of the string.
// If index is at the end, there is no character there to look at.
if (ix != ut->b) {
+ // Note: this function will only move the index back if it is on a trail byte
+ // and there is a preceding lead byte and the sequence from the lead
+ // through this trail could be part of a valid UTF-8 sequence
+ // Otherwise the index remains unchanged.
U8_SET_CP_START(s8, 0, ix);
}
@@ -1645,7 +1656,10 @@
UChar *buf = u8b->buf;
uint8_t *mapToNative = u8b->mapToNative;
uint8_t *mapToUChars = u8b->mapToUChars;
- int32_t toUCharsMapStart = ix - (UTF8_TEXT_CHUNK_SIZE*3 + 1);
+ int32_t toUCharsMapStart = ix - sizeof(UTF8Buf::mapToUChars) + 1;
+ // Note that toUCharsMapStart can be negative. Happens when the remaining
+ // text from current position to the beginning is less than the buffer size.
+ // + 1 because mapToUChars must have a slot at the end for the bufNativeLimit entry.
int32_t destIx = UTF8_TEXT_CHUNK_SIZE+2; // Start in the overflow region
// at end of buffer to leave room
// for a surrogate pair at the
@@ -1672,6 +1686,7 @@
if (c<0x80) {
// Special case ASCII range for speed.
buf[destIx] = (UChar)c;
+ U_ASSERT(toUCharsMapStart <= srcIx);
mapToUChars[srcIx - toUCharsMapStart] = (uint8_t)destIx;
mapToNative[destIx] = (uint8_t)(srcIx - toUCharsMapStart);
} else {
@@ -1705,6 +1720,7 @@
do {
mapToUChars[sIx-- - toUCharsMapStart] = (uint8_t)destIx;
} while (sIx >= srcIx);
+ U_ASSERT(toUCharsMapStart <= (srcIx+1));
// Set native indexing limit to be the current position.
// We are processing a non-ascii, non-native-indexing char now;
@@ -1885,6 +1901,7 @@
U_ASSERT(index>=ut->chunkNativeStart+ut->nativeIndexingLimit);
U_ASSERT(index<=ut->chunkNativeLimit);
int32_t mapIndex = index - u8b->toUCharsMapStart;
+ U_ASSERT(mapIndex < (int32_t)sizeof(UTF8Buf::mapToUChars));
int32_t offset = u8b->mapToUChars[mapIndex] - u8b->bufStartIdx;
U_ASSERT(offset>=0 && offset<=ut->chunkLength);
return offset;
