[
https://issues.apache.org/jira/browse/AVRO-3979?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Chanzo Bryan updated AVRO-3979:
-------------------------------
Description:
When a reader schema has a type mismatch with a writer schema where the writer
schema type is a union, the location information for the schema compatibility
is lost in Python. This was fixed in Java as a part of AVRO-3612.
Given a reader record schema with _int_ field {_}a{_}:
{code:java}
{
"type": "record",
"name": "Record1",
"fields": [
{
"name": "a",
"type": "int"
}
]
}{code}
and a writer record schema with an _optional int_ (union of _int_ and
{_}null{_}) field {_}a{_}:
{code:java}
{
"type": "record",
"name": "Record1",
"fields": [
{
"name": "a",
"type": ["null", "int"],
"default": "null"
}
]
} {code}
I would expect to get the follow message and location on invoking
{_}get_compatibility(reader, writer){_}:
{code:java}
message: "reader type: int not compatible with writer type: null"
location: "/fields/0/type/0" {code}
Instead, we currently get:
{code:java}
message: "reader type: int not compatible with writer type: null"
location: "/" {code}
losing all location information.
was:
When a reader schema has a type mismatch with a writer schema where the writer
schema type is a union, the location information for the schema compatibility
is lost in Python. This was fixed in Java as a part of AVRO-3612.
Given a reader record schema with _int_ field {_}a{_}:
{code:java}
{
"type": "record",
"name": "Record1",
"fields": [
{
"name": "a",
"type": "int"
}
]
}{code}
and a writer record schema with an _optional int_ (union of _int_ and
{_}null{_}) field {_}a{_}:
{code:java}
{
"type": "record",
"name": "Record1",
"fields": [
{
"name": "a",
"type": ["null", "int"],
"default": "null"
}
]
} {code}
I would expect to get the follow message and location on invoking
{_}get_compatibility(reader, writer){_}:
{code:java}
message: "reader type: int not compatible with writer type: null"
location: "/fields/0/type/0" {code}
Instead, we currently get:
{code:java}
message: "reader type: int not compatible with writer type: null"
location: "/" {code}
losing all location information.
> Python SchemaCompatibility doesn't set location properly for type mismatch
> with writer unions
> ---------------------------------------------------------------------------------------------
>
> Key: AVRO-3979
> URL: https://issues.apache.org/jira/browse/AVRO-3979
> Project: Apache Avro
> Issue Type: Bug
> Components: python
> Affects Versions: 1.12.0, 1.11.3, 1.11.4
> Reporter: Chanzo Bryan
> Priority: Major
> Original Estimate: 20m
> Remaining Estimate: 20m
>
> When a reader schema has a type mismatch with a writer schema where the
> writer schema type is a union, the location information for the schema
> compatibility is lost in Python. This was fixed in Java as a part of
> AVRO-3612.
> Given a reader record schema with _int_ field {_}a{_}:
> {code:java}
> {
> "type": "record",
> "name": "Record1",
> "fields": [
> {
> "name": "a",
> "type": "int"
> }
> ]
> }{code}
>
> and a writer record schema with an _optional int_ (union of _int_ and
> {_}null{_}) field {_}a{_}:
> {code:java}
> {
> "type": "record",
> "name": "Record1",
> "fields": [
> {
> "name": "a",
> "type": ["null", "int"],
> "default": "null"
> }
> ]
> } {code}
>
> I would expect to get the follow message and location on invoking
> {_}get_compatibility(reader, writer){_}:
> {code:java}
> message: "reader type: int not compatible with writer type: null"
> location: "/fields/0/type/0" {code}
>
> Instead, we currently get:
> {code:java}
> message: "reader type: int not compatible with writer type: null"
> location: "/" {code}
>
> losing all location information.
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
