Merge r184965 - Crash under ICU with ASAN during editing/selection/mo…

…ve-by-word-visually-crash-test-5.html

https://bugs.webkit.org/show_bug.cgi?id=145429
<rdar://problem/20992218>

Reviewed by Alexey Proskuryakov.

WebKit uses some strings which contain the lower 8-bits of UTF-16 (thereby saving space). However,
ICU doesn't understand this encoding. When we want to use ICU functions with strings in this encoding,
we create a UTextProvider which converts our encoded strings to UTF-16 for ICU, one chunk at a time.
This object contains a vtable which we populate to perform the conversion.

The WebKit function which actually returns the UTF-16 chunks has two relevant arguments: an index into
the encoded string which ICU is requesting, and a direction from that index which ICU is interested
in. This function populates a "chunk" which is characterized by a pointer to a buffer, the length of
the populated data in the buffer, and an offset into the chunk which represents the index that the
requested character was put into.

When ICU requests data going backward, we fill in the chunk accordingly, with the requested character
all the way at the end. We then set the offset equal to the length of the buffer. However, this length
value is stale from the previous time the function ran. Therefore, ICU was reading the wrong index in
the chunk when expecting the requested character.

Covered by editing/selection/move-by-word-visually-crash-test-5.html.

* platform/text/icu/UTextProviderLatin1.cpp:
(WebCore::uTextLatin1Access):

Source/WebCore/ChangeLog

@@ -1,3 +1,32 @@
+2015-05-28  Myles C. Maxfield  <mmaxfield@apple.com>
+
+        Crash under ICU with ASAN during editing/selection/move-by-word-visually-crash-test-5.html
+        https://bugs.webkit.org/show_bug.cgi?id=145429
+        <rdar://problem/20992218>
+
+        Reviewed by Alexey Proskuryakov.
+
+        WebKit uses some strings which contain the lower 8-bits of UTF-16 (thereby saving space). However,
+        ICU doesn't understand this encoding. When we want to use ICU functions with strings in this encoding,
+        we create a UTextProvider which converts our encoded strings to UTF-16 for ICU, one chunk at a time.
+        This object contains a vtable which we populate to perform the conversion.
+
+        The WebKit function which actually returns the UTF-16 chunks has two relevant arguments: an index into
+        the encoded string which ICU is requesting, and a direction from that index which ICU is interested
+        in. This function populates a "chunk" which is characterized by a pointer to a buffer, the length of
+        the populated data in the buffer, and an offset into the chunk which represents the index that the
+        requested character was put into.
+
+        When ICU requests data going backward, we fill in the chunk accordingly, with the requested character
+        all the way at the end. We then set the offset equal to the length of the buffer. However, this length
+        value is stale from the previous time the function ran. Therefore, ICU was reading the wrong index in
+        the chunk when expecting the requested character.
+
+        Covered by editing/selection/move-by-word-visually-crash-test-5.html.
+
+        * platform/text/icu/UTextProviderLatin1.cpp:
+        (WebCore::uTextLatin1Access):
+
 2015-05-29  Brady Eidson  <beidson@apple.com>
 
         Review feedback followup for r185003.

Source/WebCore/platform/text/icu/UTextProviderLatin1.cpp

@@ -104,7 +104,7 @@ static UBool uTextLatin1Access(UText* uText, int64_t index, UBool forward)
         }
         if (index >= length && uText->chunkNativeLimit == length) {
             // Off the end of the buffer, but we can't get it.
-            uText->chunkOffset = uText->chunkLength;
+            uText->chunkOffset = static_cast<int32_t>(index - uText->chunkNativeStart);
             return FALSE;
         }
     } else {
@@ -136,7 +136,7 @@ static UBool uTextLatin1Access(UText* uText, int64_t index, UBool forward)
         if (uText->chunkNativeStart < 0)
             uText->chunkNativeStart = 0;
 
-        uText->chunkOffset = uText->chunkLength;
+        uText->chunkOffset = static_cast<int32_t>(index - uText->chunkNativeStart);
     }
     uText->chunkLength = static_cast<int32_t>(uText->chunkNativeLimit - uText->chunkNativeStart);