commonMain.com.xebia.functional.openai.OneOfOps.kt Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of xef-openai-client-jvm Show documentation
Show all versions of xef-openai-client-jvm Show documentation
Building applications with LLMs through composability in Kotlin
package com.xebia.functional.openai
import kotlin.reflect.KClass
import kotlinx.serialization.SerializationException
import kotlinx.serialization.json.JsonElement
class OneOfSerializationException(
val payload: JsonElement,
val errors: Map, SerializationException>,
override val message: String =
"""
Failed to deserialize Json: $payload.
Errors: ${
errors.entries.joinToString(separator = "\n") { (type, error) ->
"$type - failed to deserialize: ${error.stackTraceToString()}"
}
}
"""
.trimIndent()
) : SerializationException(message)
/**
* OpenAI makes a lot of use of oneOf types (sum types, or unions types), but it **never** relies on
* a discriminator field to differentiate between the types.
*
* Typically, what OpenAI does is attach a common field like `type` (a single value enum). I.e.
* `MessageObjectContentInner` has a type field with `image` or `text`. Depending on the `type`
* property, the other properties will be different.
*
* Due to the use of these fields, it **seems** there are no overlapping objects in the schema. So
* to deserialize these types, we can try to deserialize each type and return the first one that
* succeeds. In the case **all** fail, we throw [OneOfSerializationException] which includes all the
* attempted types with their errors.
*
* This method relies on 'peeking', which is not possible in KotlinX Serialization. So to achieve
* peeking, we first deserialize the raw Json to JsonElement, which safely consumes the buffer. And
* then we can attempt to deserialize the JsonElement to the desired type, without breaking the
* internal parser buffer.
*/
internal fun attemptDeserialize(
json: JsonElement,
vararg block: Pair, (json: JsonElement) -> A>
): A {
val errors = linkedMapOf, SerializationException>()
block.forEach { (kclass, f) ->
try {
return f(json)
} catch (e: SerializationException) {
errors[kclass] = e
}
}
throw OneOfSerializationException(json, errors)
}