Fix schema inference in case of empty messages in Protobuf/CapnProto formats

[Avogar]

Changelog category (leave one):

• Bug Fix (user-visible misbehavior in official stable or prestable release)

Changelog entry (a user-readable short description of the changes that goes to CHANGELOG.md):

Fix bug in schema inference in case of empty messages in Protobuf/CapnProto formats that allowed to create column with empty Tuple type. Closes #39051
Add 2 new settings input_format_{protobuf/capnproto}_skip_fields_with_unsupported_types_in_schema_inference that allow to skip fields with unsupported types while schema inference for Protobuf and CapnProto formats

Information about CI checks: https://clickhouse.com/docs/en/development/continuous-integration/

[ekpdt]

Our current workaround for this bug is to use schema inference to get a preliminary schema on our (extremely) complex Protobuf messages and then edit the inferred schema to remove illegal empty tuples.

This patch would make it impossible for us to use schema inference at all, which is unfortunate.

Would you consider allowing some way to DESCRIBE a protobuf-based schema via the File engine but prohibit creating any actual tables? I admit this seems like a weird solution because CH will still create a schema that is illegal.

If you integrate this patch, we will probably pre-process our .proto files to remove empty messages.

[Avogar]

Our current workaround for this bug is to use schema inference to get a preliminary schema on our (extremely) complex Protobuf messages and then edit the inferred schema to remove illegal empty tuples.

This patch would make it impossible for us to use schema inference at all, which is unfortunate.

Would you consider allowing some way to DESCRIBE a protobuf-based schema via the File engine but prohibit creating any actual tables? I admit this seems like a weird solution because CH will still create a schema that is illegal.

If you integrate this patch, we will probably pre-process our .proto files to remove empty messages.

What if I add special setting like input_format_protobuf_skip_columns_with_unsupported_types_in_schema_inference to silently skip bad messages (like empty tuples) instead of throwing an exception? Will it help?

Like, if we see a field that is an empty message and this setting is enabled, we will just skip this field while schema inference

[ekpdt]

What if I add special setting like input_format_protobuf_skip_columns_with_unsupported_types_in_schema_inference to silently skip bad messages (like empty tuples) instead of throwing an exception? Will it help?

That would be perfect, thanks.

[tavplubix]

New test is flaky with Ordinary database: https://s3.amazonaws.com/clickhouse-test-reports/0/8e98b7d47dbbcfe88c8d420b35794854e93a6361/stateless_tests__release__databaseordinary_.html

#39623