diff --git a/source/common/utext.cpp b/source/common/utext.cpp
index f16591f..1430c74 100644
--- a/source/common/utext.cpp
+++ b/source/common/utext.cpp
@@ -836,9 +836,15 @@ U_CDECL_END
//------------------------------------------------------------------------------
// 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 };
@@ -878,7 +884,7 @@ struct UTF8Buf {
// 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;
};
@@ -1021,7 +1027,8 @@ utf8TextAccess(UText *ut, int64_t index, UBool forward) {
// Requested index is in this buffer.
u8b = (UTF8Buf *)ut->p; // the current buffer
mapIndex = ix - u8b->toUCharsMapStart;
- ut->chunkOffset = u8b->mapToUChars[mapIndex] - u8b->bufStartIdx;
+ U_ASSERT(mapIndex < (int32_t)sizeof(UTF8Buf::mapToUChars));
+ ut->chunkOffset = u8b->mapToUChars[mapIndex] - u8b->bufStartIdx;
return TRUE;
}
@@ -1287,6 +1294,10 @@ fillReverse:
// 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);
}
@@ -1300,7 +1311,10 @@ fillReverse:
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
@@ -1327,6 +1341,7 @@ fillReverse:
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 {
@@ -1356,6 +1371,7 @@ fillReverse:
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;
@@ -1530,6 +1546,7 @@ utf8TextMapIndexToUTF16(const UText *ut, int64_t index64) {
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;
diff --git a/source/test/intltest/utxttest.cpp b/source/test/intltest/utxttest.cpp
index 17180cf..41ec461 100644
--- a/source/test/intltest/utxttest.cpp
+++ b/source/test/intltest/utxttest.cpp
@@ -59,6 +59,8 @@ UTextTest::runIndexedTest(int32_t index, UBool exec,
if (exec) Ticket6847(); break;
case 5: name = "Ticket10562";
if (exec) Ticket10562(); break;
+ case 6: name = "Ticket12888";
+ if (exec) Ticket12888(); break;
default: name = ""; break;
}
}
@@ -1483,3 +1485,62 @@ void UTextTest::Ticket10562() {
utext_close(usText);
}
+// Ticket 12888: bad handling of illegal utf-8 containing many instances of the archaic, now illegal,
+// six byte utf-8 forms. Original implementation had an assumption that
+// there would be at most three utf-8 bytes per UTF-16 code unit.
+// The five and six byte sequences map to a single replacement character.
+
+void UTextTest::Ticket12888() {
+ const char *badString =
+ "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
+ "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
+ "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
+ "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
+ "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
+ "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
+ "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
+ "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
+ "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
+ "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
+ "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
+ "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
+ "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
+ "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
+ "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
+ "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
+ "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
+ "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
+ "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80"
+ "\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80\xfd\x80\x80\x80\x80\x80";
+
+ UErrorCode status = U_ZERO_ERROR;
+ LocalUTextPointer ut(utext_openUTF8(NULL, badString, -1, &status));
+ TEST_SUCCESS(status);
+ for (;;) {
+ UChar32 c = utext_next32(ut.getAlias());
+ if (c == U_SENTINEL) {
+ break;
+ }
+ }
+ int32_t endIdx = utext_getNativeIndex(ut.getAlias());
+ if (endIdx != (int32_t)strlen(badString)) {
+ errln("%s:%d expected=%d, actual=%d", __FILE__, __LINE__, strlen(badString), endIdx);
+ return;
+ }
+
+ for (int32_t prevIndex = endIdx; prevIndex>0;) {
+ UChar32 c = utext_previous32(ut.getAlias());
+ int32_t currentIndex = utext_getNativeIndex(ut.getAlias());
+ if (c != 0xfffd) {
+ errln("%s:%d (expected, actual, index) = (%d, %d, %d)\n",
+ __FILE__, __LINE__, 0xfffd, c, currentIndex);
+ break;
+ }
+ if (currentIndex != prevIndex - 6) {
+ errln("%s:%d: wrong index. Expected, actual = %d, %d",
+ __FILE__, __LINE__, prevIndex - 6, currentIndex);
+ break;
+ }
+ prevIndex = currentIndex;
+ }
+}
diff --git a/source/test/intltest/utxttest.h b/source/test/intltest/utxttest.h
index db32ae0..5c29434 100644
--- a/source/test/intltest/utxttest.h
+++ b/source/test/intltest/utxttest.h
@@ -34,6 +34,7 @@ public:
void Ticket5560();
void Ticket6847();
void Ticket10562();
+ void Ticket12888();
private:
struct m { // Map between native indices & code points.
