(#178) Fix structured output JSON descriptions #191
Conversation
Properly handle description annotation during schema generation
EugeneTheDev
added
bug
There was a problem hiding this comment.
@EugeneTheDev please also add a test with a lot of different descriptors:
- class descriptor via annotation
- property descriptor via annotation
- property of Serializable type, also with field, each of them annotated
- one field is annotated and also overriden by the map
- one field is NOT annotated but provided through the map
- what else can be an edge case?
| * @param schemaFormat Format of the generated schema, can be simple or detailed. | ||
| * @param maxDepth Maximum recursion depth when generating schema to prevent infinite recursion for circular references. | ||
| * @param descriptions Optional map of serial class names and property names to descriptions. | ||
| * If a property/type is already described with [LLMDescription] annotation, value from the map will override this description. |
There was a problem hiding this comment.
Let's also please add good example in KDoc about how to form the descriptions map.
I mean, "className.propertyName" to "property description"
There was a problem hiding this comment.
Also, maybe let's name it descriptionOverrides if it's an overriding map?
There was a problem hiding this comment.
Regarding testing edge cases, that's what I already modified in tests, to cover these cases you described. I didn't mark each field with annotation though because it looks redundant, and it's also nice to test that description is added only for these fields where it is specified, and is skipped for other fields.
Regarding examples and parameter renaming: sounds good, will do it
Co-authored-by: Damjan Polugic <[email protected]>
Based on #188 and fixes #178 by properly handling
@LLMDescriptionduring JSON schema generation. Also updated the tests to include the annotation too and simplified structured output example to use existing node