flag truncate_timestamps added to fetcharrow functions

Arrow only allows timestamps with years between 1400 to 9999.
This new flag will allow to truncate the dates preventing exceptions.

Author: JorgeGarciaIrazabal

cpp/turbodbc/Library/src/time_helpers.cpp

@@ -1,5 +1,4 @@
 #include <turbodbc/time_helpers.h>
-
 #ifdef _WIN32
 #include <windows.h>
 #endif
@@ -26,6 +25,21 @@ int64_t timestamp_to_microseconds(char const * data_pointer)
     return (ts - timestamp_epoch).total_microseconds();
 }
 
+int64_t timestamp_to_microseconds_truncated(char const * data_pointer)
+{
+    auto & sql_ts = *reinterpret_cast<SQL_TIMESTAMP_STRUCT const *>(data_pointer);
+    if (sql_ts.year > 9999) {
+        SQL_TIMESTAMP_STRUCT new_sql_ts = {9999, 12, 31, 23, 59, 59, 999999000};
+        return timestamp_to_microseconds(reinterpret_cast<char const *>(&new_sql_ts));
+    }
+    if(sql_ts.year < 1400) {
+        SQL_TIMESTAMP_STRUCT new_sql_ts = {1400, 1, 1, 0, 0, 0, 0};
+        return timestamp_to_microseconds(reinterpret_cast<char const *>(&new_sql_ts));
+    }
+
+    return timestamp_to_microseconds(data_pointer);
+}
+
 
 void microseconds_to_timestamp(int64_t microseconds, char * data_pointer)
 {

cpp/turbodbc/Library/turbodbc/time_helpers.h

@@ -10,6 +10,15 @@ namespace turbodbc {
  */
 int64_t timestamp_to_microseconds(char const * data_pointer);
 
+
+/**
+ * @brief Convert an SQL_TIMESTAMP_STRUCT stored at data_pointer to an
+ *        integer describing the elapsed microseconds since the POSIX epoch
+ *        and truncated to arrow valid range (year 1400 to 9999)
+ */
+int64_t timestamp_to_microseconds_truncated(char const * data_pointer);
+
+
 /**
  * @brief Convert the number of microseconds since the POSIX epoch
  *        to a timestamp and store it in an SQL_TIMESTAMP_STRUCT located

cpp/turbodbc/Test/tests/time_helpers_test.cpp

@@ -8,6 +8,7 @@
 #include <sql.h>
 
 using turbodbc::timestamp_to_microseconds;
+using turbodbc::timestamp_to_microseconds_truncated;
 using turbodbc::microseconds_to_timestamp;
 using turbodbc::nanoseconds_to_timestamp;
 using turbodbc::date_to_days;
@@ -59,6 +60,24 @@ TEST(TimeHelpersTest, MicrosecondsToTimestampForYear4000)
 }
 
 
+TEST(TimeHelpersTest, TimestampToMicrosecondsTruncatedForYear10000)
+{
+    SQL_TIMESTAMP_STRUCT data = {10000, 01, 02, 3, 4, 5, 123456000};
+    // expectation one microsecond before year 10000
+    std::int64_t expected = 253402300799999999;
+    EXPECT_EQ(expected, timestamp_to_microseconds_truncated(reinterpret_cast<char const *>(&data)));
+}
+
+
+TEST(TimeHelpersTest, TimestampToMicrosecondsTruncatedForYear100)
+{
+    SQL_TIMESTAMP_STRUCT data = {100, 01, 02, 3, 4, 5, 0};
+    // expectation january first 1400
+    std::int64_t expected = -17987443200000000;
+    EXPECT_EQ(expected, timestamp_to_microseconds_truncated(reinterpret_cast<char const *>(&data)));
+}
+
+
 TEST(TimeHelpersTest, NanosecondsToTimestampForEpoch)
 {
     std::int64_t const nanoseconds = 0;

cpp/turbodbc_arrow/Library/src/arrow_result_set.cpp

@@ -159,8 +159,8 @@ Status AppendIntsToBuilder(size_t rows_in_batch, std::unique_ptr<ArrayBuilder> c
 
 }
 
-arrow_result_set::arrow_result_set(turbodbc::result_sets::result_set & base, bool strings_as_dictionary, bool adaptive_integers) :
-    base_result_(base), strings_as_dictionary_(strings_as_dictionary), adaptive_integers_(adaptive_integers)
+arrow_result_set::arrow_result_set(turbodbc::result_sets::result_set & base, bool strings_as_dictionary, bool adaptive_integers, bool truncate_timestamps) :
+    base_result_(base), strings_as_dictionary_(strings_as_dictionary), adaptive_integers_(adaptive_integers),  truncate_timestamps_(truncate_timestamps)
 {
 }
 
@@ -213,12 +213,14 @@ Status append_to_bool_builder(size_t rows_in_batch, std::unique_ptr<ArrayBuilder
     return typed_builder->AppendValues(data_ptr, rows_in_batch, valid_bytes);
 }
 
-Status append_to_timestamp_builder(size_t rows_in_batch, std::unique_ptr<ArrayBuilder> const& builder, cpp_odbc::multi_value_buffer const& input_buffer, uint8_t*) {
+Status append_to_timestamp_builder(size_t rows_in_batch, std::unique_ptr<ArrayBuilder> const& builder, cpp_odbc::multi_value_buffer const& input_buffer, uint8_t*, bool truncate_timestamp) {
     auto typed_builder = static_cast<TimestampBuilder*>(builder.get());
     for (std::size_t j = 0; j < rows_in_batch; ++j) {
         auto element = input_buffer[j];
         if (element.indicator == SQL_NULL_DATA) {
             ARROW_RETURN_NOT_OK(typed_builder->AppendNull());
+        } else if (truncate_timestamp) {
+            ARROW_RETURN_NOT_OK(typed_builder->Append(turbodbc::timestamp_to_microseconds_truncated(element.data_pointer)));
         } else {
             ARROW_RETURN_NOT_OK(typed_builder->Append(turbodbc::timestamp_to_microseconds(element.data_pointer)));
         }
@@ -288,7 +290,7 @@ Status arrow_result_set::process_batch(size_t rows_in_batch, std::vector<std::un
                 ARROW_RETURN_NOT_OK(append_to_bool_builder(rows_in_batch, columns[i], buffers[i].get(), valid_bytes.data()));
                 break;
             case turbodbc::type_code::timestamp:
-                ARROW_RETURN_NOT_OK(append_to_timestamp_builder(rows_in_batch, columns[i], buffers[i].get(), valid_bytes.data()));
+                ARROW_RETURN_NOT_OK(append_to_timestamp_builder(rows_in_batch, columns[i], buffers[i].get(), valid_bytes.data(), truncate_timestamps_));
                 break;
             case turbodbc::type_code::date:
                 ARROW_RETURN_NOT_OK(append_to_date_builder(rows_in_batch, columns[i], buffers[i].get(), valid_bytes.data()));

cpp/turbodbc_arrow/Library/src/python_bindings.cpp

@@ -9,9 +9,9 @@ using turbodbc_arrow::arrow_result_set;
 namespace {
 
 arrow_result_set make_arrow_result_set(std::shared_ptr<turbodbc::result_sets::result_set> result_set_pointer,
-    bool strings_as_dictionary, bool adaptive_integers)
+    bool strings_as_dictionary, bool adaptive_integers, bool truncate_timestamps)
 {
-	return arrow_result_set(*result_set_pointer, strings_as_dictionary, adaptive_integers);
+	return arrow_result_set(*result_set_pointer, strings_as_dictionary, adaptive_integers, truncate_timestamps);
 }
 
 void set_arrow_parameters(turbodbc::cursor & cursor, pybind11::object const & pyarrow_table)

cpp/turbodbc_arrow/Library/turbodbc_arrow/arrow_result_set.h

@@ -27,7 +27,7 @@ class PYBIND11_EXPORT arrow_result_set {
      *        in the base result set in a row-based fashion
      */
     arrow_result_set(turbodbc::result_sets::result_set & base, bool strings_as_dictionary,
-        bool adaptive_integers);
+        bool adaptive_integers, bool truncate_timestamps);
 
     /**
      * @brief Retrieve a native (C++) Arrow Table which contains
@@ -58,6 +58,7 @@ class PYBIND11_EXPORT arrow_result_set {
     turbodbc::result_sets::result_set & base_result_;
     bool strings_as_dictionary_;
     bool adaptive_integers_;
+    bool truncate_timestamps_;
 };
 
 }

cpp/turbodbc_arrow/Test/tests/arrow_result_set_test.cpp

@@ -29,6 +29,7 @@ namespace {
 
     bool const strings_as_strings = false;
     bool const strings_as_dictionaries = true;
+    bool const truncate_timestamps = true;
 
     bool const plain_integers = false;
     bool const compressed_integers = true;
@@ -129,7 +130,7 @@ class ArrowResultSetTest : public ::testing::Test {
             auto schema = std::make_shared<arrow::Schema>(expected_fields);
             std::shared_ptr<arrow::Table> expected_table = arrow::Table::Make(schema, expected_arrays);
 
-            turbodbc_arrow::arrow_result_set ars(rs, strings_as_dictionary, adaptive_integers);
+            turbodbc_arrow::arrow_result_set ars(rs, strings_as_dictionary, adaptive_integers, truncate_timestamps);
             std::shared_ptr<arrow::Table> table;
             ASSERT_OK(ars.fetch_all_native(&table, false));
             ASSERT_TRUE(expected_table->Equals(*table));
@@ -151,7 +152,7 @@ TEST_F(ArrowResultSetTest, SimpleSchemaConversion)
         "int_column", turbodbc::type_code::integer, size_unimportant, true}};
     EXPECT_CALL(rs, do_get_column_info()).WillRepeatedly(testing::Return(expected));
 
-    turbodbc_arrow::arrow_result_set ars(rs, strings_as_strings, plain_integers);
+    turbodbc_arrow::arrow_result_set ars(rs, strings_as_strings, plain_integers, truncate_timestamps);
     auto schema = ars.schema();
     ASSERT_EQ(schema->num_fields(), 1);
     auto field = schema->field(0);
@@ -191,7 +192,7 @@ TEST_F(ArrowResultSetTest, AllTypesSchemaConversion)
         std::make_shared<arrow::Field>("nonnull_int_column", arrow::int64(), false)
     };
 
-    turbodbc_arrow::arrow_result_set ars(rs, strings_as_strings, plain_integers);
+    turbodbc_arrow::arrow_result_set ars(rs, strings_as_strings, plain_integers, truncate_timestamps);
     auto schema = ars.schema();
 
     ASSERT_EQ(schema->num_fields(), 12);
@@ -225,7 +226,7 @@ TEST_F(ArrowResultSetTest, SingleBatchSingleColumnResultSetConversion)
     EXPECT_CALL(rs, do_get_buffers()).WillOnce(testing::Return(expected_buffers));
     EXPECT_CALL(rs, do_fetch_next_batch()).WillOnce(testing::Return(OUTPUT_SIZE)).WillOnce(testing::Return(0));
 
-    turbodbc_arrow::arrow_result_set ars(rs, strings_as_strings, plain_integers);
+    turbodbc_arrow::arrow_result_set ars(rs, strings_as_strings, plain_integers, truncate_timestamps);
     std::shared_ptr<arrow::Table> table;
     ASSERT_OK(ars.fetch_all_native(&table, false));
     ASSERT_TRUE(expected_table->Equals(*table));

python/turbodbc/cursor.py

@@ -307,7 +307,7 @@ def _numpy_batch_generator(self):
             first_run = False
             yield result_batch
 
-    def fetcharrowbatches(self, strings_as_dictionary=False, adaptive_integers=False):
+    def fetcharrowbatches(self, strings_as_dictionary=False, adaptive_integers=False, truncate_timestamps=False):
         """
         Fetches rows in the active result set generated with ``execute()`` or
         ``executemany()`` as an iterable of arrow tables.
@@ -320,6 +320,8 @@ def fetcharrowbatches(self, strings_as_dictionary=False, adaptive_integers=False
                 smallest possible integer type in which all values can be
                 stored. Be aware that here the type depends on the resulting
                 data.
+        :param truncate_timestamps: If true, instead of throwing if a timestamp is not between
+                the valid years 1400 and 9999, a truncated value will be returned.
 
         :return: generator of ``pyarrow.Table``
         """
@@ -329,7 +331,8 @@ def fetcharrowbatches(self, strings_as_dictionary=False, adaptive_integers=False
             rs = make_arrow_result_set(
                 self.impl.get_result_set(),
                 strings_as_dictionary,
-                adaptive_integers)
+                adaptive_integers,
+                truncate_timestamps)
             first_run = True
             while True:
                 table = rs.fetch_next_batch()
@@ -341,7 +344,7 @@ def fetcharrowbatches(self, strings_as_dictionary=False, adaptive_integers=False
         else:
             raise Error(_NO_ARROW_SUPPORT_MSG)
 
-    def fetchallarrow(self, strings_as_dictionary=False, adaptive_integers=False):
+    def fetchallarrow(self, strings_as_dictionary=False, adaptive_integers=False, truncate_timestamps=False):
         """
         Fetches all rows in the active result set generated with ``execute()`` or
         ``executemany()``.
@@ -355,6 +358,9 @@ def fetchallarrow(self, strings_as_dictionary=False, adaptive_integers=False):
                 stored. Be aware that here the type depends on the resulting
                 data.
 
+        :param truncate_timestamps: If true, instead of throwing if a timestamp is not between
+                the valid years 1400 and 9999, a truncated value will be returned.
+
         :return: ``pyarrow.Table``
         """
         self._assert_valid_result_set()
@@ -363,7 +369,8 @@ def fetchallarrow(self, strings_as_dictionary=False, adaptive_integers=False):
             return make_arrow_result_set(
                     self.impl.get_result_set(),
                     strings_as_dictionary,
-                    adaptive_integers).fetch_all()
+                    adaptive_integers,
+                    truncate_timestamps).fetch_all()
         else:
             raise Error(_NO_ARROW_SUPPORT_MSG)