Improve speed when listing columns in BigQuery

[ebyhr]

Description

Using mini parser because BigQuery Java SDK doesn't support translating string to BigQuery type as far as I asked Google engineers.

This PR improves listing columns. Test with 169 tables is improved from 22s to 1.5s.

SELECT * FROM bigquery.information_schema.columns WHERE table_schema = 'test';

• Add tests for special column names
• Add tests for unsupported column types and fields

Release notes

(x) Release notes are required, with the following suggested text:

# BigQuery
* TBD. ({issue}`issuenumber`)

[wendigo]

this is testing the same thing twice:

getBigQueryTypeInfo(typeString) is TypeInfoUtils.getTypeInfoFromTypeString(typeString)

[wendigo]

Why do we need that? It's only used for test purpose.

[wendigo]

We don't need a category - this information is already encoded in the sealed class hierarchy. Instead of a switch (typeInfo.getCategory()) we can justswitch (typeInfo)

[wendigo]

Suggested change

	private TypeSignature toTypeSignature(TypeInfo typeInfo)
	{
	switch (typeInfo.getCategory()) {
	case PRIMITIVE:
	Type primitiveType = fromPrimitiveType((PrimitiveTypeInfo) typeInfo);
	if (primitiveType == null) {
	break;
	}
	return primitiveType.getTypeSignature();
	case ARRAY:
	ArrayTypeInfo arrayTypeInfo = (ArrayTypeInfo) typeInfo;
	TypeSignature elementType = toTypeSignature(arrayTypeInfo.getListElementTypeInfo());
	return arrayType(typeParameter(elementType));
	case STRUCT:
	StructTypeInfo structTypeInfo = (StructTypeInfo) typeInfo;
	List<TypeInfo> fieldTypes = structTypeInfo.getAllStructFieldTypeInfos();
	List<String> fieldNames = structTypeInfo.getAllStructFieldNames();
	verify(fieldTypes.size() == fieldNames.size(), "Invalid BigQuery struct type: %s", typeInfo);
	return rowType(Streams.zip(
	fieldNames.stream(),
	fieldTypes.stream().map(this::toTypeSignature),
	TypeSignatureParameter::namedField)
	.collect(Collectors.toList()));
	}
	throw new TrinoException(NOT_SUPPORTED, format("Unsupported BigQuery type: %s", typeInfo));
	}
	private TypeSignature toTypeSignature(TypeInfo typeInfo)
	{
	return switch (typeInfo) {
	case PrimitiveTypeInfo primitiveTypeInfo:
	Type primitiveType = fromPrimitiveType(primitiveTypeInfo);
	if (primitiveType == null) {
	throw new TrinoException(NOT_SUPPORTED, "Unsupported primitive type: " + primitiveTypeInfo);
	}
	yield primitiveType.getTypeSignature();
	case ArrayTypeInfo arrayTypeInfo:
	TypeSignature elementType = toTypeSignature(arrayTypeInfo.getListElementTypeInfo());
	yield arrayType(typeParameter(elementType));
	case StructTypeInfo structTypeInfo:
	List<TypeInfo> fieldTypes = structTypeInfo.getAllStructFieldTypeInfos();
	List<String> fieldNames = structTypeInfo.getAllStructFieldNames();
	verify(fieldTypes.size() == fieldNames.size(), "Invalid BigQuery struct type: %s", typeInfo);
	yield rowType(Streams.zip(
	fieldNames.stream(),
	fieldTypes.stream().map(this::toTypeSignature),
	TypeSignatureParameter::namedField)
	.collect(Collectors.toList()));
	};
	}

[wendigo]

throw new TrinoException("Unsupported type"); then return null handling at the call-site

[wendigo]

Not needed - we already encode the category information in the sealed classes hierarchy

[wendigo]

Do we need hashCode and equals? Where is it used?

[ebyhr]

Closing as @nineinchnick is going to take over this PR.