Github user stevedlawrence commented on the issue:
https://github.com/apache/nifi/pull/3130
After some research and reading the [Avro
specification](https://avro.apache.org/docs/1.8.1/spec.html) , I'd agree that
the DFDL infoset does seem somewhat similar to a Record.
DFDL does support all the primitive types (null, boolean, int, long, float,
double, bytes, string) and logical types (date, time, decimal), plus a few
others (integer, byte, short, signed/unsigned) . But as far as complex types,
it only really supports "records" and arrays. Below is the list of things in
Avro that the DFDL infoset does not support:
* It sort of supports enums, but only in the sense that it can validate
that a primitive type is one of the valid enum values via the xsd:restriction.
* Maps. In DFDL, a map would be implemented as a sequence of key/value
pairs, so there wouldn't be any enforcement of unique keys.
* Unions. Each element in the infoset must have an explicit primitive type.
Each element can be optional or nulled, but cannot be a union of multiple
primitive types. In DFDL, that is handled by an xs:choice of two different
elements with different types and some method (often a discriminator) to
determine which branch of the choice.
* Namespaces are slightly different, but probably similar enough. DFDL uses
XML namespacing.
* Aliases are not supported.
* Sort order. DFDL outputs infoset elements in the order in which they
appear in schema.
* The DFDL infoset does not contain the schema. One must keep track of the
associated schema outside of the data.
* The isn't really a concept of different serializations like Avro looks to
have. Instead, the DFDL schema defines the physical data format via DFDL
annotations, which are used to determine how to serialize/deserialize data.
Theoretically, one could have different schemas with the same logical format
but with with different DFDL annotations to describe physical formats, but that
isn't a comment use case we've come across.
The Daffodil devs would be happen to discuss the possibility of integrating
Daffodil/DFDL as an alternative to Avro. Not sure if any of the above
limitations are blockers. The core of Avro and DFDL definitely do seem to have
some overlap.
---
