github-actions[bot] commented on code in PR #63192:
URL: https://github.com/apache/doris/pull/63192#discussion_r3233846088
##########
be/src/format/parquet/vparquet_column_reader.cpp:
##########
@@ -103,6 +119,753 @@ static void fill_array_offset(FieldSchema* field,
ColumnArray::Offsets64& offset
}
}
+static constexpr int64_t UNIX_EPOCH_DAYNR = 719528;
+static constexpr int64_t MICROS_PER_SECOND = 1000000;
+
+static int64_t variant_date_value(const VecDateTimeValue& value) {
+ return value.daynr() - UNIX_EPOCH_DAYNR;
+}
+
+static int64_t variant_date_value(const DateV2Value<DateV2ValueType>& value) {
+ return value.daynr() - UNIX_EPOCH_DAYNR;
+}
+
+static int64_t variant_datetime_value(const VecDateTimeValue& value) {
+ int64_t timestamp = 0;
+ value.unix_timestamp(×tamp, cctz::utc_time_zone());
+ return timestamp * MICROS_PER_SECOND;
+}
+
+static int64_t variant_datetime_value(const DateV2Value<DateTimeV2ValueType>&
value) {
+ int64_t timestamp = 0;
+ value.unix_timestamp(×tamp, cctz::utc_time_zone());
+ return timestamp * MICROS_PER_SECOND + value.microsecond();
+}
+
+static int64_t variant_datetime_value(const TimestampTzValue& value) {
+ int64_t timestamp = 0;
+ value.unix_timestamp(×tamp, cctz::utc_time_zone());
+ return timestamp * MICROS_PER_SECOND + value.microsecond();
+}
+
+static int find_child_idx(const FieldSchema& field, std::string_view name) {
+ for (int i = 0; i < field.children.size(); ++i) {
+ if (field.children[i].lower_case_name == name) {
+ return i;
+ }
+ }
+ return -1;
+}
+
+static bool is_variant_wrapper_field(const FieldSchema& field) {
+ auto type = remove_nullable(field.data_type);
Review Comment:
This wrapper detection is still too broad for nested typed-only objects. A
valid typed-only VARIANT can have a nested object field whose user key is named
`typed_value` (for example `v.typed_value.nested.typed_value` representing
`{"nested":{"typed_value":1}}`). When `can_direct_read_typed_value()` or
`append_direct_typed_column_to_batch()` reaches the `nested` struct with
`allow_variant_wrapper=true`, this predicate treats that ordinary nested object
as a structural wrapper and unwraps the user field, so the value is inserted at
path `nested` instead of `nested.typed_value`. The earlier top-level collision
fix enters the top-level container with `allow_variant_wrapper=false`, but this
nested case still goes through the wrapper path. Please make wrapper detection
context-aware for nested typed-only structs as well, and add coverage for a
nested user key named `typed_value`/`value`.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
