fix(json_parse): Clear state after encountering an error (facebookinc…

…ubator#12150)

Summary:

`json_parse` if it encounters an error and throws in a sensitive part of the code , doesnt clear out its previous state when processing the next row. This can cause it to throw when processing the next valid row.

Reviewed By: xiaoxmeng, amitkdutta

Differential Revision: D68534578

velox/functions/prestosql/JsonFunctions.cpp

@@ -238,6 +238,7 @@ class JsonParseFunction : public exec::VectorFunction {
         auto size = prepareInput(value, needNormalizes[row]);
         if (auto error = parse(size, needNormalizes[row])) {
           context.setVeloxExceptionError(row, errors_[error]);
+          clearState();
           return;
         }
         auto outputSize = concatViews(views_, output);
@@ -457,6 +458,12 @@ class JsonParseFunction : public exec::VectorFunction {
     }
   }
 
+  void clearState() const {
+    fields_.clear();
+    sortIndices_.clear();
+    fastSortKeys_.clear();
+  }
+
   mutable folly::once_flag initializeErrors_;
   mutable std::exception_ptr errors_[simdjson::NUM_ERROR_CODES];
   // Padding is needed in case string view is inlined.

velox/functions/prestosql/tests/JsonFunctionsTest.cpp

@@ -338,6 +338,21 @@ TEST_F(JsonFunctionsTest, jsonParse) {
       jsonParse("{\"k1\\"), "Invalid escape sequence at the end of string");
   VELOX_ASSERT_USER_THROW(
       jsonParse("{\"k1\\u"), "Invalid escape sequence at the end of string");
+
+  // Ensure state is cleared after invalid json
+  {
+    data = makeRowVector({makeFlatVector<StringView>({
+        R"({"key":1578377,"name":"Alto Ma\\u00e9 \\"A\\"","type":"cities"})", // invalid json
+        R"([{"k1": "v1" }, {"k2": "v2" }])" // valid json
+    })});
+
+    result = evaluate("try(json_parse(c0))", data);
+
+    expected = makeNullableFlatVector<StringView>(
+        {std::nullopt, R"([{"k1":"v1"},{"k2":"v2"}])"}, JSON());
+
+    velox::test::assertEqualVectors(expected, result);
+  }
 }
 
 TEST_F(JsonFunctionsTest, canonicalization) {