Overallocate target buffer for normalization more early. Fixes #834676.
Backported to 2.3.
diff --git a/Lib/test/test_normalization.py b/Lib/test/test_normalization.py
index 046dca6..0cbc2b4 100644
--- a/Lib/test/test_normalization.py
+++ b/Lib/test/test_normalization.py
@@ -84,5 +84,8 @@
continue
assert X == NFC(X) == NFD(X) == NFKC(X) == NFKD(X), c
+ # Check for bug 834676
+ normalize('NFC',u'\ud55c\uae00')
+
if __name__ == "__main__":
test_main()
diff --git a/Modules/unicodedata.c b/Modules/unicodedata.c
index d266ad7..311db29 100644
--- a/Modules/unicodedata.c
+++ b/Modules/unicodedata.c
@@ -311,12 +311,14 @@
stack[stackptr++] = *i++;
while(stackptr) {
Py_UNICODE code = stack[--stackptr];
- if (!space) {
- space = PyString_GET_SIZE(result) + 10;
- if (PyUnicode_Resize(&result, space) == -1)
+ /* Hangul Decomposition adds three characters in
+ a single step, so we need atleast that much room. */
+ if (space < 3) {
+ int newsize = PyString_GET_SIZE(result) + 10;
+ space += 10;
+ if (PyUnicode_Resize(&result, newsize) == -1)
return NULL;
- o = PyUnicode_AS_UNICODE(result) + space - 10;
- space = 10;
+ o = PyUnicode_AS_UNICODE(result) + newsize - space;
}
/* Hangul Decomposition. */
if (SBase <= code && code < (SBase+SCount)) {