[
https://issues.apache.org/jira/browse/AVRO-3012?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17556717#comment-17556717
]
Kalle Niemitalo commented on AVRO-3012:
---------------------------------------
If a schema with an unrecognized logical type were parsed as a LogicalSchema
rather than as the base schema, would it be possible to give it a LogicalType
whose ConvertToBaseValue and ConvertToLogicalValue methods just pass the data
through without conversions? I suppose LogicalType.GetCSharpType would then
have to return null or throw an exception, and its only caller CodeGen.getType
would have to special-case this somehow. That doesn't really seem better than
special-casing LogicalSchema.LogicalType == null.
Would it be useful for avrogen to emit the unsupported logicalType into the
_SCHEMA property? Doing so would require parsing it as a LogicalSchema (or
smuggling "logicalType" in Schema.Props and changing
PrimitiveSchema.WriteJson). [Parsing Canonical
Form|https://avro.apache.org/docs/1.11.0/spec.html#Parsing+Canonical+Form+for+Schemas]
apparently does not include logicalType, but I don't know whether
schema-registry software (e.g. by Confluent) cares about it.
> Unknown logical types are not ignored during deserialization but lead to an
> exception.
> --------------------------------------------------------------------------------------
>
> Key: AVRO-3012
> URL: https://issues.apache.org/jira/browse/AVRO-3012
> Project: Apache Avro
> Issue Type: Bug
> Components: csharp
> Affects Versions: 1.9.2, 1.10.1
> Reporter: Lucas Heimberg
> Priority: Major
>
> By specification (see
> [https://avro.apache.org/docs/current/spec.html#Logical+Types] ), unknown
> logical types and logical types with invalid specification should be ignored
> during deserialization and instead, the object of the underlying Avro type
> should be returned.
> The C# implementation of Avro however raises an exception when such a logical
> type is found during parsing of a schema string. Therefore, it is not
> possible to parse a schema with a logical type using a version of the C#
> implementation of Avro that does not supports this logical type yet.
> In particular, it is therefore as an example not possible to consume a binary
> encoded Avro datum that makes use of the logical type UUID (as available in
> Avro 1.10.1) with a deserializer using C# Avro 1.9.2. (by specification, the
> deserializer should fall back to the underlying string representation of the
> UUID).
> This severly limits the downwards compability promised by the Avro
> specification.
>
> During schema parsing, a LogicalSchema instance is created using the
> GetFromLogicalSchema method of the LogicalTypeFactory. This method has an
> optional parameter to ignore invalid logical types, which is not used by the
> caller.
> It is not completely obvious to me, whether an unsupported logical type
> should still be parsed as a LogicalSchema, although without a LogicalType, or
> whether it should be directly parsed as the underlying Avro type (in the
> implementation also called BaseSchema) - i.e., whether the downgrading to the
> underlying Avro type should happen already during parsing or later during
> deserialization using the schema.
> For the second case, a fix to the problem would also require an update to the
> ReadLogical method of the GenericReader class in order to support the case
> that the LogicalType of the writer schema is null.
>
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
