Inserting to variant containing String will lead to only String being inserted

[SkytAsul]

Describe the bug

If I correctly understood, currently ClickHouse-connect passes Variant data as String so ClickHouse server infers the type by itself. However, if the Variant contains the String type (e.g. Variant(Int64, Float32, String)), all data will be inserted as String.

Steps to reproduce

1. Create a table containing a Variant(Int64, Float32, String) column
2. Insert some int or float data using ClickHouse-connect
3. Run SELECT variantType(yourColumn) and see that all rows have String type

Expected behaviour

The correct type is kept

Configuration

Environment

• clickhouse-connect version: 0.8.18
• Python version: 3.12.3

ClickHouse server

• ClickHouse Server version: 25.7.1
• ClickHouse Server non-default settings, if any:

"profile","max_threads","1"
"profile","log_queries","1"
"profile","log_queries_min_type","QUERY_FINISH"
"profile","log_queries_min_query_duration_ms","0"
"profile","log_queries_cut_to_length","10000"
"profile","http_max_field_value_size","262144"
"profile","allow_introspection_functions","1"
"profile","use_query_cache","0"
"profile","query_cache_nondeterministic_function_handling","ignore"
"profile","query_cache_system_table_handling","ignore"
"profile","query_cache_min_query_runs","2"
"profile","query_cache_min_query_duration","100"
"profile","query_cache_ttl","30"
"profile","use_query_condition_cache","0"
"server","max_thread_pool_size","10000"
"server","default_database","default"
"server","max_server_memory_usage","30281777971"
"server","max_server_memory_usage_to_ram_ratio","0.9"
"server","max_concurrent_queries","1000"
"server","uncompressed_cache_size","8589934592"
"server","query_condition_cache_size","106700800"
"server","concurrent_threads_soft_limit_num","0"
"server","concurrent_threads_soft_limit_ratio_to_cores","2"
"server","concurrent_threads_scheduler","fair_round_robin"
"server","async_load_databases","1"
"server","keep_alive_timeout","10"
"server","total_memory_profiler_step","4194304"
"server","total_memory_tracker_sample_probability","0"
"server","validate_tcp_client_information","0"
"server","global_profiler_real_time_period_ns","1000000000"
"server","global_profiler_cpu_time_period_ns","100000000"
"server","dictionaries_lazy_load","1"
"server","wait_dictionaries_load_at_startup","1"

[joe-clickhouse]

Hi @SkytAsul, thanks for filing this!

Unfortunately it’s a known limitation in the current driver—everything is inserted as a String (see the known limitations section of the CHANGELOG). There aren’t any technical blockers per se; it’s mostly a matter of implementation priority. To support true numeric (or other) variants we’d need to:

• Detect each Python value’s type (int, float, etc.)
• Map to the proper Variant sub-type
• Emit the correct discriminator byte per element
• Serialize the value in the Variant’s native binary format

I’ve added this to our backlog as an enhancement. Please let us know if you have any questions or want to help contribute!