Structured CloudEvent fails to convert to `io.cloudevents.CloudEvent` in the functional model (data lost / "Unknown encoding")

[radhakrishnan10192]

Describe the bug

When a structured-mode CloudEvent (Content-Type: application/cloudevents+json, whole envelope in the body) is consumed by a function whose target type is the CNCF io.cloudevents.CloudEvent, message conversion fails.

CloudEventMessageUtils.toCanonical(...) transforms the structured message into a binary-shaped message (data extracted to the payload, attributes moved to ce-* headers), but:

1. it does not reset the contentType header — the outgoing message keeps application/cloudevents+json even though the payload is no longer the serialized envelope; and
2. it leaves the extracted data as a deserialized Map, not serialized bytes.

Downstream, the CNCF io.cloudevents.spring.messaging.CloudEventMessageConverter re-decides structured-vs-binary purely from contentType. Because the stale application/cloudevents+json remains, it takes the structured reader and tries to re-parse the (already extracted) payload as the envelope:

io.cloudevents.rw.CloudEventRWException: Could not parse. Unknown encoding. Invalid content type or spec version
	at io.cloudevents.core.message.impl.MessageUtils.parseStructuredOrBinaryMessage(MessageUtils.java:69)
	at io.cloudevents.spring.messaging.CloudEventMessageConverter.createMessageReader(CloudEventMessageConverter.java:60)
	at io.cloudevents.spring.messaging.CloudEventMessageConverter.fromMessage(CloudEventMessageConverter.java:45)
	at org.springframework.cloud.function.context.config.SmartCompositeMessageConverter.fromMessage(SmartCompositeMessageConverter.java:137)
	at org.springframework.cloud.function.context.catalog.SimpleFunctionRegistry$FunctionInvocationWrapper.convertInputMessageIfNecessary(SimpleFunctionRegistry.java:1488)

Even after the contentType is corrected so the binary reader is selected, a second symptom remains: the CNCF binary reader only reads byte[]/String payloads (CloudEventMessageConverter#getBinaryData returns null for anything else), so the Map payload is dropped and the delivered cloudEvent.getData() is null while the attributes are present.

This only affects the io.cloudevents.CloudEvent target type. A plain POJO target works, because Spring's own message converters can bind a Map payload to a POJO — which is why the existing CloudEventFunctionTests#testStructuredPojoToPojoDefaultOutputAttributeProvider passes and this case is not covered.

Affected versions

• spring-cloud-function-context:4.3.x. CloudEventMessageUtils.toCanonical / buildBinaryMessageFromStructuredMap are byte-identical across these.
• io.cloudevents:cloudevents-spring: 2.2.1 (the optional dependency declared by spring-cloud-function-context);
• Java: 21.

Sample CloudEvent (structured mode)

Content-Type: application/cloudevents+json, body:

{
  "specversion" : "1.0",
  "type" : "org.springframework",
  "source" : "https://spring.io/",
  "id" : "A234-1234-1234",
  "datacontenttype" : "application/json",
  "data" : {
    "version" : "1.0",
    "releaseName" : "Spring Framework",
    "releaseDate" : "24-03-2004"
  }
}

Steps to reproduce

Minimal function with an io.cloudevents.CloudEvent input type (the cloudevents-spring converter is auto-registered by ContextFunctionCatalogAutoConfiguration$CloudEventsMessageConverterConfiguration):

@EnableAutoConfiguration
@Configuration
static class TestConfiguration {
    @Bean
    Function<Message<CloudEvent>, Message<CloudEvent>> echoCloudEvent() {
        return Function.identity();
    }
}

@Test
void structuredToCloudEventTypedPayload() {
    String payload = /* the structured JSON envelope above */;

    Function<Object, Object> function = context.getBean(FunctionCatalog.class).lookup("echoCloudEvent");

    Message<String> inputMessage = MessageBuilder
            .withPayload(payload)
            .setHeader(MessageHeaders.CONTENT_TYPE, "application/cloudevents+json")
            .build();

    // structured mode: attributes are in the body, not the headers
    assertThat(CloudEventMessageUtils.isCloudEvent(inputMessage)).isFalse();

    Message<CloudEvent> result = (Message<CloudEvent>) function.apply(inputMessage); // throws
}

Actual behavior

function.apply(...) throws CloudEventRWException: Could not parse. Unknown encoding during input conversion. (If contentType is corrected in isolation, the call succeeds but the delivered CloudEvent has getData() == null.)

Expected behavior

The function receives a fully-populated io.cloudevents.CloudEvent — correct specversion, type, source, id, datacontenttype, and non-null data — consistent with how the same structured event is handled when the target type is a POJO.

Root cause

In CloudEventMessageUtils.buildBinaryMessageFromStructuredMap(...), after extracting data and moving attributes to ce-* headers, the original headers are copied back onto the outgoing message — re-applying contentType=application/cloudevents+json — and the data remains a Map. The binary-mode branch of toCanonical already resets contentType to the data content-type; the structured branch does not, and it does not serialize the data.

Proposed fix

In buildBinaryMessageFromStructuredMap, after building the binary-mode message:

1. reset MessageHeaders.CONTENT_TYPE to the data content-type (mirroring the binary-mode branch), and
2. re-serialize the extracted data to bytes using the data content-type when it is not already byte[]/String.

With both applied, the structured event converts correctly to io.cloudevents.CloudEvent (attributes + data), and all existing CloudEvent tests still pass. Happy to open a PR with a CloudEventFunctionTests case covering the io.cloudevents.CloudEvent target type.

[radhakrishnan10192]

@olegz Hi Oleg, i have encountered this issue and i think its a potential issue but would like your review. Thanks

[radhakrishnan10192]

For better reproducer we can use this sample project

[radhakrishnan10192]

this could be a potential fix for this.