Wrong class name when using $ref in json schema

[razem-io]

We have the following issue:

alliance.json

{
  "type": "object",
  "properties": {
    "tag": {
      "type": "string"
    },
    "name": {
      "type": "string"
    },
    "points": {
      "type": "integer"
    },
    "diplomacy": {
      "type": "object",
      "properties": {
        "treaties": {
          "type": "array",
          "items": {
               "$ref": "alliance_treaty.json"
          }
        },
        "ownOffers": {
          "type": "array",
          "items": {
               "$ref": "alliance_treaty.json"
          }
        },
        "foreignOffers": {
          "type": "array",
          "items": {
               "$ref": "alliance_treaty.json"
          }
        }
      }
    }
  }
}

alliance_treaty.json

{
  "type": "object",
  "properties": {
    "allianceId": {
      "type": "integer"
    },
    "otherAllianceId": {
      "type": "integer"
    },
    "type": {
      "type": "integer"
    },
    "status": {
      "type": "integer"
    } 
  }
}

We expect the output to be like this:

• Alliance.java
• AllianceTreaty.java

Actual result is:

• Alliance.java
• Treaty.java

This only happens when using "$ref" otherwise the class gets generated correctly. Is this a bug or are we missing something?

Thanks for your time!
Julian

[joelittlejohn]

When $ref is used the name of the type is driven but the property name (in this case treaties becomes treaty). This is discussed further under #22.

If you want to work around this you could use the javaType extension property to force Treaty to become AllianceTreaty.

[razem-io]

javaType works as a workaround. Thank you for the fast answer!

However, in our case the property does not seem to be the problem. The other workaround we figured out is renaming the file alliance_treaty.json to _alliance_treaty.json. This also results in the correct output. But only if we also generate the _alliance_treaty.json with the SchemaMapper. At first we excluded the $ref json files which results in the wrong output. Perhaps it is a mix of both. Property name and file name?

[joelittlejohn]

When using $ref the class names can appear non-deterministic and this is because the output is affected by the order in which in the input schemas are read.

Imagine you point jsonschema2pojo at a directory of schemas. If _alliance_treaty.json is read first then of course this input file causes AllianceTreaty.java to be created. When the plugin then reads another schema alliance.json it creates Alliance.java. When the $ref value is encountered, the plugin can determine that this is a reference to a type that already exists (a schema that has been encountered before), so it reuses the type. Files are read in alphabetical order so _alliance_treaty.json and alliance.json produce the output you want.

Now if you have alliance_treaty.json and alliance.json then alliance.json is read first. It creates Alliance.java and the property name dictates that treaties should produce a java class name of Treaty. Now when we have finished generation for alliance.json we move on to alliance_treaty.json. Again, the plugin detects that this schema has been encountered before and it doesn't generate any new type. So now we have Alliance.java and Treaty.java.

Ideally, I'd like to move to a pre-processor model where we build the entire tree of known schemas and only begin generation after this tree is understood. For now though, you can see that without javaType you can get some unexpected results.

As to why the property name is used when generating a new type, in many cases this is preferable, e.g. if the schema is not a $ref, or if the $ref is a fragment that points to another node in the same file (or a fragment in another file). We probably need to try to differentiate between these cases to understand whether property name should be used or not - unfortunately this may also appear a bit magical at times.

[razem-io]

Thank you for the detailed answer and your time. We will stick with javaType then.
Should have thought about the alphabetical ordering before writing a comment.

[dadrus]

This is related to #481. Please check the commit linked to it.