diff --git a/docs/en/operations/settings/settings-formats.md b/docs/en/operations/settings/settings-formats.md index f8b40cd81ac..fcec0afb8d2 100644 --- a/docs/en/operations/settings/settings-formats.md +++ b/docs/en/operations/settings/settings-formats.md @@ -194,6 +194,17 @@ If enabled, ClickHouse will try to infer type `DateTime64` from string fields in Enabled by default. +## input_format_try_infer_variants {#input_format_try_infer_variants} + +If enabled, ClickHouse will try to infer type [`Variant`](../../sql-reference/data-types/variant.md) in schema inference for text formats when there is more than one possible type for column/array elements. + +Possible values: + +- 0 — Disabled. +- 1 — Enabled. + +Default value: `0`. + ## date_time_input_format {#date_time_input_format} Allows choosing a parser of the text representation of date and time. diff --git a/src/Core/Settings.h b/src/Core/Settings.h index d8837d26e54..479d5939b57 100644 --- a/src/Core/Settings.h +++ b/src/Core/Settings.h @@ -1136,6 +1136,7 @@ class IColumn; M(Bool, input_format_json_defaults_for_missing_elements_in_named_tuple, true, "Insert default value in named tuple element if it's missing in json object", 0) \ M(Bool, input_format_json_throw_on_bad_escape_sequence, true, "Throw an exception if JSON string contains bad escape sequence in JSON input formats. If disabled, bad escape sequences will remain as is in the data", 0) \ M(Bool, input_format_json_ignore_unnecessary_fields, true, "Ignore unnecessary fields and not parse them. Enabling this may not throw exceptions on json strings of invalid format or with duplicated fields", 0) \ + M(Bool, input_format_try_infer_variants, false, "Try to infer the Variant type in text formats when there is more than one possible type for column/array elements", 0) \ M(Bool, type_json_skip_duplicated_paths, false, "When enabled, during parsing JSON object into JSON type duplicated paths will be ignored and only the first one will be inserted instead of an exception", 0) \ M(UInt64, input_format_json_max_depth, 1000, "Maximum depth of a field in JSON. This is not a strict limit, it does not have to be applied precisely.", 0) \ M(Bool, input_format_try_infer_integers, true, "Try to infer integers instead of floats while schema inference in text formats", 0) \ diff --git a/src/Core/SettingsChangesHistory.cpp b/src/Core/SettingsChangesHistory.cpp index 2415323b4a0..fb59577b0f0 100644 --- a/src/Core/SettingsChangesHistory.cpp +++ b/src/Core/SettingsChangesHistory.cpp @@ -71,6 +71,7 @@ static std::initializer_listgetSettingsRef().max_parser_depth; format_settings.client_protocol_version = context->getClientProtocolVersion(); format_settings.date_time_overflow_behavior = settings.date_time_overflow_behavior; + format_settings.try_infer_variant = settings.input_format_try_infer_variants; /// Validate avro_schema_registry_url with RemoteHostFilter when non-empty and in Server context if (format_settings.schema.is_server) diff --git a/src/Formats/FormatSettings.h b/src/Formats/FormatSettings.h index da25da74134..ed178a68b9d 100644 --- a/src/Formats/FormatSettings.h +++ b/src/Formats/FormatSettings.h @@ -35,6 +35,7 @@ struct FormatSettings bool decimal_trailing_zeros = false; bool defaults_for_omitted_fields = true; bool is_writing_to_terminal = false; + bool try_infer_variant = false; bool seekable_read = true; UInt64 max_rows_to_read_for_schema_inference = 25000; diff --git a/src/Formats/SchemaInferenceUtils.cpp b/src/Formats/SchemaInferenceUtils.cpp index ed251fa5d55..e8eab3b4453 100644 --- a/src/Formats/SchemaInferenceUtils.cpp +++ b/src/Formats/SchemaInferenceUtils.cpp @@ -7,6 +7,7 @@ #include #include #include +#include #include #include #include @@ -306,6 +307,33 @@ namespace type_indexes.erase(TypeIndex::UInt64); } + /// if setting 'try_infer_variant' is true then we convert to type variant. + void transformVariant(DataTypes & data_types, TypeIndexesSet & type_indexes) + { + if (checkIfTypesAreEqual(data_types)) + return; + + DataTypes variant_types; + for (const auto & type : data_types) + { + if (const auto * variant_type = typeid_cast(type.get())) + { + const auto & current_variants = variant_type->getVariants(); + variant_types.insert(variant_types.end(), current_variants.begin(), current_variants.end()); + } + else + { + variant_types.push_back(type); + } + } + + auto variant_type = std::make_shared(variant_types); + + for (auto & type : data_types) + type = variant_type; + type_indexes = {TypeIndex::Variant}; + } + /// If we have only date/datetimes types (Date/DateTime/DateTime64), convert all of them to the common type, /// otherwise, convert all Date, DateTime and DateTime64 to String. void transformDatesAndDateTimes(DataTypes & data_types, TypeIndexesSet & type_indexes) @@ -652,7 +680,11 @@ namespace transformDatesAndDateTimes(data_types, type_indexes); if constexpr (!is_json) + { + if (settings.try_infer_variant) + transformVariant(data_types, type_indexes); return; + } /// Check settings specific for JSON formats. @@ -670,6 +702,10 @@ namespace if (settings.json.try_infer_objects_as_tuples) mergeJSONPaths(data_types, type_indexes, settings, json_info); + + if (settings.try_infer_variant) + transformVariant(data_types, type_indexes); + }; auto transform_complex_types = [&](DataTypes & data_types, TypeIndexesSet & type_indexes) @@ -682,7 +718,11 @@ namespace transformNothingComplexTypes(data_types, type_indexes); if constexpr (!is_json) + { + if (settings.try_infer_variant) + transformVariant(data_types, type_indexes); return; + } /// Convert JSON tuples with same nested types to arrays. transformTuplesWithEqualNestedTypesToArrays(data_types, type_indexes); @@ -695,6 +735,9 @@ namespace if (json_info && json_info->allow_merging_named_tuples) mergeNamedTuples(data_types, type_indexes, settings, json_info); + + if (settings.try_infer_variant) + transformVariant(data_types, type_indexes); }; transformTypesRecursively(types, transform_simple_types, transform_complex_types); @@ -861,7 +904,6 @@ namespace if (checkIfTypesAreEqual(nested_types_copy)) return std::make_shared(nested_types_copy.back()); - return std::make_shared(nested_types); } else @@ -1456,6 +1498,15 @@ void transformFinalInferredJSONTypeIfNeededImpl(DataTypePtr & data_type, const F return; } + + if (const auto * variant_type = typeid_cast(data_type.get())) + { + auto nested_types = variant_type->getVariants(); + for (auto & nested_type : nested_types) + transformFinalInferredJSONTypeIfNeededImpl(nested_type, settings, json_info, remain_nothing_types); + data_type = std::make_shared(nested_types); + return; + } } void transformFinalInferredJSONTypeIfNeeded(DataTypePtr & data_type, const FormatSettings & settings, JSONInferenceInfo * json_info) @@ -1535,6 +1586,20 @@ DataTypePtr makeNullableRecursively(DataTypePtr type) return nested_type ? std::make_shared(nested_type) : nullptr; } + if (which.isVariant()) + { + const auto * variant_type = assert_cast(type.get()); + DataTypes nested_types; + for (const auto & nested_type: variant_type->getVariants()) + { + if (!nested_type->lowCardinality() && nested_type->haveSubtypes()) + nested_types.push_back(makeNullableRecursively(nested_type)); + else + nested_types.push_back(nested_type); + } + return std::make_shared(nested_types); + } + if (which.isTuple()) { const auto * tuple_type = assert_cast(type.get()); diff --git a/tests/queries/0_stateless/03150_infer_type_variant.reference b/tests/queries/0_stateless/03150_infer_type_variant.reference new file mode 100644 index 00000000000..a43fa1e1227 --- /dev/null +++ b/tests/queries/0_stateless/03150_infer_type_variant.reference @@ -0,0 +1,31 @@ + ┏━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓ + ┃ arr ┃ toTypeName(arr) ┃ + ┡━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┩ +1. │ ['1','Hello',(32)] │ Array(Variant(String, Tuple( + a Nullable(Int64)))) │ + └────────────────────┴──────────────────────────────────────────────────────┘ + ┏━━━━━━━┳━━━━━━━━━━━━━━━━━━┓ + ┃ x ┃ toTypeName(x) ┃ + ┡━━━━━━━╇━━━━━━━━━━━━━━━━━━┩ +1. │ 42 │ Nullable(String) │ + ├───────┼──────────────────┤ +2. │ Hello │ Nullable(String) │ + └───────┴──────────────────┘ + ┏━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓ + ┃ x ┃ toTypeName(x) ┃ + ┡━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┩ +1. │ [1,2,3] │ Variant(Array(Nullable(Int64)), Tuple( + a Nullable(Int64))) │ + ├─────────┼───────────────────────────────────────────────────────────────┤ +2. │ (42) │ Variant(Array(Nullable(Int64)), Tuple( + a Nullable(Int64))) │ + └─────────┴───────────────────────────────────────────────────────────────┘ + ┏━━━━┳━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓ + ┃ c1 ┃ toTypeName(c1) ┃ c2 ┃ toTypeName(c2) ┃ + ┡━━━━╇━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┩ +1. │ 1 │ Nullable(Int64) │ Hello World! │ Variant(Array(Nullable(Int64)), String) │ + ├────┼─────────────────┼──────────────┼─────────────────────────────────────────┤ +2. │ 2 │ Nullable(Int64) │ [1,2,3] │ Variant(Array(Nullable(Int64)), String) │ + ├────┼─────────────────┼──────────────┼─────────────────────────────────────────┤ +3. │ 3 │ Nullable(Int64) │ 2020-01-01 │ Variant(Array(Nullable(Int64)), String) │ + └────┴─────────────────┴──────────────┴─────────────────────────────────────────┘ diff --git a/tests/queries/0_stateless/03150_infer_type_variant.sql b/tests/queries/0_stateless/03150_infer_type_variant.sql new file mode 100644 index 00000000000..45126ccd471 --- /dev/null +++ b/tests/queries/0_stateless/03150_infer_type_variant.sql @@ -0,0 +1,5 @@ +SET input_format_try_infer_variants=1; +SELECT arr, toTypeName(arr) FROM format('JSONEachRow', '{"arr" : [1, "Hello", {"a" : 32}]}') FORMAT Pretty; +SELECT x, toTypeName(x) FROM format('JSONEachRow', '{"x" : 42}, {"x" : "Hello"}') FORMAT Pretty; +SELECT x, toTypeName(x) FROM format('JSONEachRow', '{"x" : [1, 2, 3]}, {"x" : {"a" : 42}}') FORMAT Pretty; +SELECT c1, toTypeName(c1), c2, toTypeName(c2) FROM format('CSV', '1,Hello World!\n2,"[1,2,3]"\n3,"2020-01-01"\n') FORMAT Pretty; \ No newline at end of file