Copy the UTF-8 valid flag upon interning a string

Author: nielsdos

Zend/zend_string.c

@@ -191,6 +191,19 @@ ZEND_API zend_string* ZEND_FASTCALL zend_interned_string_find_permanent(zend_str
 	return zend_interned_string_ht_lookup(str, &interned_strings_permanent);
 }
 
+static zend_string* ZEND_FASTCALL zend_init_string_for_interning(zend_string *str, bool persistent) {
+	uint32_t flags = 0;
+	if (ZSTR_IS_VALID_UTF8(str)) {
+		flags = IS_STR_VALID_UTF8;
+	}
+	zend_ulong h = ZSTR_H(str);
+	zend_string_delref(str);
+	str = zend_string_init(ZSTR_VAL(str), ZSTR_LEN(str), persistent);
+	GC_ADD_FLAGS(str, flags);
+	ZSTR_H(str) = h;
+	return str;
+}
+
 static zend_string* ZEND_FASTCALL zend_new_interned_string_permanent(zend_string *str)
 {
 	zend_string *ret;
@@ -208,10 +221,7 @@ static zend_string* ZEND_FASTCALL zend_new_interned_string_permanent(zend_string
 
 	ZEND_ASSERT(GC_FLAGS(str) & GC_PERSISTENT);
 	if (GC_REFCOUNT(str) > 1) {
-		zend_ulong h = ZSTR_H(str);
-		zend_string_delref(str);
-		str = zend_string_init(ZSTR_VAL(str), ZSTR_LEN(str), 1);
-		ZSTR_H(str) = h;
+		str = zend_init_string_for_interning(str, true);
 	}
 
 	return zend_add_interned_string(str, &interned_strings_permanent, IS_STR_PERMANENT);
@@ -249,10 +259,7 @@ static zend_string* ZEND_FASTCALL zend_new_interned_string_request(zend_string *
 	}
 #endif
 	if (GC_REFCOUNT(str) > 1) {
-		zend_ulong h = ZSTR_H(str);
-		zend_string_delref(str);
-		str = zend_string_init(ZSTR_VAL(str), ZSTR_LEN(str), 0);
-		ZSTR_H(str) = h;
+		str = zend_init_string_for_interning(str, false);
 	}
 
 	ret = zend_add_interned_string(str, &CG(interned_strings), 0);

ext/zend_test/tests/strings_marked_as_utf8.phpt

@@ -2,8 +2,6 @@
 Check that strings are marked as valid UTF-8
 --EXTENSIONS--
 zend_test
---XFAIL--
-Flag is not set in assignment when concatenating known UTF-8 strings
 --FILE--
 <?php
 echo "Empty strings:\n";
@@ -49,8 +47,14 @@ $s = "f" . "o";
 var_dump($s);
 var_dump(zend_test_is_string_marked_as_valid_utf8($s));
 
+// The "foo" string matches with a "Foo" class which is registered by the zend_test extension.
+// That class name does not have the "valid UTF-8" flag because class names in general
+// don't have to be UTF-8. As the "foo" string here goes through the interning logic,
+// the string gets replaced by the "foo" string from the class, which does
+// not have the "valid UTF-8" flag. We therefore choose a different test case: "fxo".
+// The previous "foo" test case works because it is not interned.
 echo "Multiple concatenation known valid UTF-8 in assignment:\n";
-$s = "f" . "o" . "o";
+$s = "f" . "x" . "o";
 var_dump($s);
 var_dump(zend_test_is_string_marked_as_valid_utf8($s));
 
@@ -124,7 +128,7 @@ Concatenation known valid UTF-8 in assignment:
 string(2) "fo"
 bool(true)
 Multiple concatenation known valid UTF-8 in assignment:
-string(3) "foo"
+string(3) "fxo"
 bool(true)
 Concatenation known valid UTF-8 string with empty string in variables:
 bool(true)