[
https://issues.apache.org/jira/browse/AVRO-4093?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17901195#comment-17901195
]
Oscar Westra van Holthe - Kind commented on AVRO-4093:
------------------------------------------------------
You've just demonstrated the need for conflict resolution in the new union
specification. The reason: values in different types can have the same (plain)
JSON representation. Somehow, one must be able to choose the right union type.
Now, choosing one at random or by smallest hash value is out of the question,
as the chosen type must be predictable. So we've chosen the simplest possible
one: treat the union as an ordered list and choose the first matching type.
Note that the "union is a list of types" is not new: we previously demanded
that the default value matches the first union type.
This is (IMHO) the best solution for the fields {{enum}}, {{string_enum}}, and
{{int_long}}.
> New union defaulting rules is dangerous.
> ----------------------------------------
>
> Key: AVRO-4093
> URL: https://issues.apache.org/jira/browse/AVRO-4093
> Project: Apache Avro
> Issue Type: Bug
> Components: spec
> Affects Versions: 1.12.0
> Reporter: Huw Campbell
> Priority: Major
>
> {{In version 1.12, the specification for unions changed, so that the default
> value must be one of the values instead of the first one.}}
> {{While this sounds ok at first blush, it opens a whole kettle of worms with
> regards to schemas which appear correct but contain bugs.}}
> {{For example, this schema has four examples which I want defaults for, in
> all cases, I want the second one, but the first will be picked.}}
>
> {color:#a31515}{ "type": "record"{color}
> {color:#a31515}, "name": "demo1"{color}
> {color:#a31515}, "fields" :{color}
> {color:#a31515} [ { "name": "enum"{color}
> {color:#a31515} , "type":{color}
> {color:#a31515} [ \{ "type": "enum", "name": "suit", "symbols": [ "diamonds",
> "hearts", "clubs", "spades" ] }{color}
> {color:#a31515} , \{ "type": "enum", "name": "tools", "symbols": [ "shovels",
> "spades", "hammers", "drills" ] }{color}
> {color:#a31515} ]{color}
> {color:#a31515} , "doc": "I want tools to be the default, but if I select
> spades we get a suit!"{color}
> {color:#a31515} , "default": "spades"{color}
> {color:#a31515} }{color}
> {color:#a31515} , { "name": "string_enum"{color}
> {color:#a31515} , "type":{color}
> {color:#a31515} [ \{ "type": "string" }{color}
> {color:#a31515} , \{ "type": "enum", "name": "tools", "symbols": [ "shovels",
> "spades", "hammers", "drills" ] }{color}
> {color:#a31515} ]{color}
> {color:#a31515} , "doc": "I want tools to be the default, but this will be a
> string"{color}
> {color:#a31515} , "default": "drill"{color}
> {color:#a31515} }{color}
> {color:#a31515} , { "name": "int_long"{color}
> {color:#a31515} , "type": [ "int", "long"]{color}
> {color:#a31515} , "doc": "I can't make this a long"{color}
> {color:#a31515} , "default": 400000000{color}
> {color:#a31515} }{color}
> {color:#a31515} , { "name": "records_ambiguity"{color}
> {color:#a31515} , "type":{color}
> {color:#a31515} [ { "name": "xs"{color}
> {color:#a31515} , "type":{color}
> {color:#a31515} { "type": "record"{color}
> {color:#a31515} , "name": "xs"{color}
> {color:#a31515} , "fields" : [ \{ "name": "z", "type": [ "null", "int" ],
> "default": null }, \{ "name": "w", "type": [ "null", "int" ], "default": null
> } ]{color}
> {color:#a31515} }{color}
> {color:#a31515} }{color}
> {color:#a31515} , { "name": "ys"{color}
> {color:#a31515} , "type":{color}
> {color:#a31515} { "type": "record"{color}
> {color:#a31515} , "name": "ys"{color}
> {color:#a31515} , "fields" : [ \{ "name": "a", "type": "int" } , \{ "name":
> "c", "type": "string" } ]{color}
> {color:#a31515} }{color}
> {color:#a31515} }{color}
> {color:#a31515} ]{color}
> {color:#a31515} , "doc": "I want ys, hence the fields a and c, xs has
> everything with defaults, so it matches and the items are discarded."{color}
> {color:#a31515} , "default": \{ "a": 2, "c": "yes" }{color}
> {color:#a31515} }{color}
> {color:#a31515} ]{color}
> {color:#a31515}}
>
> {{Other things which it can get wrong are "fixed", fields which have names,
> so two are fine, but they could be parsed as each other, "bytes", and so on.}}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
