HTTP binary mode does not percent-encode or decode header values (spec 3.1.3.2)

Start by reading http/basic/src/main/java/io/cloudevents/http/impl/HttpMessageWriter.java and the corresponding HTTP binary-mode reader, then check the linked specification discussion before deciding the applicable behavior. Use the subject and string-extension examples in the issue to define focused interoperability checks; done means both directions handle the specified header values consistently without contradicting the agreed specification.

Summary

The HTTP binary-mode binding does not percent-encode or percent-decode header values, which HTTP Protocol Binding 3.1.3.2 requires. An event sent by a conformant SDK is read back with the escapes intact, and an event sent by this SDK is rejected or mis-decoded by one.

Filed alongside cloudevents/spec#1397, which has the cross-SDK picture, and cloudevents/sdk-go#1331 for the same defect there. If the spec discussion changes the rule, this issue goes away with it, so it may be worth waiting on that before acting.

Spec text

Space (U+0020), Double-quote (U+0022), Percent (U+0025), and any characters outside the printable ASCII range of U+0021-U+007E inclusive [are percent-encoded]

and

When performing percent-decoding (when decoding an HTTP message to a CloudEvent), values that have been unnecessarily percent-encoded MUST be accepted, but encoded byte sequences which are invalid in UTF-8 MUST be rejected.

Where

http/basic/src/main/java/io/cloudevents/http/impl/HttpMessageWriter.java passes the attribute value straight to the header consumer:

putHeader.accept(headerName, value);

and the reader takes it back unchanged. A repository-wide search for URLEncoder or any percent-encoding helper returns nothing, so neither direction is implemented.

I read this from the source rather than running it, unlike the Go case which I executed. Please correct me if a caller-supplied writer is expected to do the encoding.

Expected

For an event with subject of a b and a string extension pct of 100%:

ce-subject: a%20b
ce-pct: 100%25

and on receive, a%20b decodes to a b.

Impact

Only binary mode, and only for values containing space, double quote, percent or non-ASCII, so in practice subject and string extensions. Against sdk-csharp, which does implement the rule:

Compatibility note

Implementing this is a wire-format change for anyone whose attribute values contain those characters, and a Java sender fixed in isolation would then be misread by an unfixed Java receiver. That is a real migration concern and part of why I filed the spec issue first rather than sending a patch here.

Measured 2026-09-21. I found this checking a C++ SDK for cross-SDK interoperability rather than as a Java SDK user.