error when using parquet broker load to table with generated column from MAP

[Frefreak]

Steps to reproduce the behavior (Required)

CREATE TABLE `test` (
  `a` varchar(65533),
  `b` map<varchar(65533), varchar(65533)>
) ENGINE = OLAP
DUPLICATE KEY (`a`)
DISTRIBUTED BY RANDOM
ORDER BY(`a`);

CREATE TABLE `test_gen` (
  `a` varchar(65533),
  `b` map<varchar(65533), varchar(65533)>,
  `c` varchar(65533) as b['c']
) ENGINE = OLAP
DUPLICATE KEY (`a`)
DISTRIBUTED BY RANDOM
ORDER BY(`a`);

with an arbitrary csv like this:

aaa,"{'a':'11','b':'22','c':'3\'3'}"
bbb,"{'a':'111','b':'222','c':'33\'3'}"

Convert to parquet, and try to load with something like:

LOAD LABEL logs.poc1 ( DATA INFILE("file:///poc.parquet") INTO TABLE test FORMAT AS "parquet" ( a, b)) WITH BROKER allin1broker;
LOAD LABEL logs.poc2 ( DATA INFILE("file:///poc.parquet") INTO TABLE test_gen FORMAT AS "parquet" ( a, b)) WITH BROKER allin1broker;

The generated parquet when inspected with parquet-tool show this (also attached, need to rename):
poc.parquet.txt

❯ parquet-tools show poc.parquet
+-----+---------------------------------------------+
| a   | b                                           |
|-----+---------------------------------------------|
| aaa | [('a', '11'), ('b', '22'), ('c', "3'3")]    |
| bbb | [('a', '111'), ('b', '222'), ('c', "33'3")] |
+-----+---------------------------------------------+

❯ parquet-tools inspect poc.parquet

############ file meta data ############
created_by: parquet-cpp-arrow version 16.1.0
num_columns: 3
num_rows: 2
num_row_groups: 1
format_version: 2.6
serialized_size: 1954


############ Columns ############
a
key
value

############ Column(a) ############
name: a
path: a
max_definition_level: 1
max_repetition_level: 0
physical_type: BYTE_ARRAY
logical_type: String
converted_type (legacy): UTF8
compression: SNAPPY (space_saved: -6%)

############ Column(key) ############
name: key
path: b.key_value.key
max_definition_level: 2
max_repetition_level: 1
physical_type: BYTE_ARRAY
logical_type: String
converted_type (legacy): UTF8
compression: SNAPPY (space_saved: -6%)

############ Column(value) ############
name: value
path: b.key_value.value
max_definition_level: 3
max_repetition_level: 1
physical_type: BYTE_ARRAY
logical_type: String
converted_type (legacy): UTF8
compression: SNAPPY (space_saved: -2%)

Expected behavior (Required)

both load succeed

Real behavior (Required)

the first load to table test without generated column succeed, but the later with the c generated column field failed with error:

type:ETL_RUN_FAIL; msg:Cannot cast '<slot 4>' from VARCHAR to MAP<VARCHAR(65533),VARCHAR(65533)> 

StarRocks version (Required)

3.3.1-2b87854 (all in one docker)

BTW its seems unclear in the docs on how to streamload a csv to table with MAP field. I tried several format but it either reports errors or the map isn't correctly parsed. One way I find is using str_to_map in the columns header but that format seems to be hard to convert to when my value string contains many symbols like ':', '|' etc, not exactly sure how to escape.