Fixed error in lowerUTF8 and upperUTF8 functions [#CLICKHOUSE-2]

dbms/src/Functions/LowerUpperUTF8Impl.h

@@ -1,5 +1,6 @@
 #include <Columns/ColumnString.h>
 #include <Poco/UTF8Encoding.h>
+#include <Common/UTF8Helpers.h>
 
 #if __SSE2__
 #include <emmintrin.h>
@@ -9,6 +10,11 @@
 namespace DB
 {
 
+namespace ErrorCodes
+{
+    extern const int BAD_ARGUMENTS;
+}
+
 namespace
 {
     /// xor or do nothing
@@ -90,10 +96,9 @@ struct LowerUpperUTF8Impl
         array(data.data(), data.data() + data.size(), res_data.data());
     }
 
-    static void vector_fixed(const ColumnString::Chars & data, size_t /*n*/, ColumnString::Chars & res_data)
+    static void vector_fixed(const ColumnString::Chars &, size_t, ColumnString::Chars &)
     {
-        res_data.resize(data.size());
-        array(data.data(), data.data() + data.size(), res_data.data());
+        throw Exception("Functions lowerUTF8 and upperUTF8 cannot work with FixedString argument", ErrorCodes::BAD_ARGUMENTS);
     }
 
     /** Converts a single code point starting at `src` to desired case, storing result starting at `dst`.
@@ -129,16 +134,28 @@ struct LowerUpperUTF8Impl
         {
             static const Poco::UTF8Encoding utf8;
 
-            if (const auto chars = utf8.convert(to_case(utf8.convert(src)), dst, src_end - src))
+            int src_sequence_length = UTF8::seqLength(*src);
+
+            int src_code_point = utf8.queryConvert(src, src_end - src);
+            if (src_code_point > 0)
             {
-                src += chars;
-                dst += chars;
-            }
-            else
-            {
-                ++src;
-                ++dst;
+                int dst_code_point = to_case(src_code_point);
+                if (dst_code_point > 0)
+                {
+                    int dst_sequence_length = utf8.convert(dst_code_point, dst, src_end - src);
+
+                    /// We don't support cases when lowercase and uppercase characters occupy different number of bytes in UTF-8.
+                    /// As an example, this happens for ß and ẞ.
+                    if (dst_sequence_length == src_sequence_length)
+                    {
+                        src += dst_sequence_length;
+                        dst += dst_sequence_length;
+                        return;
+                    }
+                }
             }
+
+            *dst++ = *src++;
         }
     }
 
@@ -149,7 +166,7 @@ private:
     static void array(const UInt8 * src, const UInt8 * src_end, UInt8 * dst)
     {
 #if __SSE2__
-        const auto bytes_sse = sizeof(__m128i);
+        static constexpr auto bytes_sse = sizeof(__m128i);
         auto src_end_sse = src + (src_end - src) / bytes_sse * bytes_sse;
 
         /// SSE2 packed comparison operate on signed types, hence compare (c < 0) instead of (c > 0x7f)

dbms/tests/queries/0_stateless/00761_lower_utf8_bug.reference

@@ -0,0 +1 @@
+1

dbms/tests/queries/0_stateless/00761_lower_utf8_bug.sql

@@ -0,0 +1 @@
+SELECT lowerUTF8('\xF0') = lowerUTF8('\xF0');

dbms/tests/queries/bugs/fuzzy.sql

@@ -6,9 +6,6 @@ SELECT globalNotIn(['"wh'], [NULL]);
 SELECT globalIn([''], [NULL])
 SELECT ( SELECT toDecimal128([], rowNumberInBlock()) ) , lcm('', [[(CAST(('>A') AS String))]]);
 SELECT truncate(895, -16);
-SELECT (CAST((lowerUTF8('a7\xwK>-')) AS String)), [6935];
-SELECT upperUTF8(sipHash128('\0')), [], ['xD2jG'];
-SELECT upperUTF8(SHA256(''));
 SELECT arrayEnumerateUniq(anyHeavy([]), []);
 SELECT notIn([['']], [[NULL]]);
 SELECT subtractDays((CAST((-5263074.47) AS DateTime)), -737895);