MultipartFormDataOutput as an entity does not populate boundary in Content-Type header

[mqchen-ws]

Version: 6.2.9.Final

Example code

    public static void main(String[] args) {
        final Client client = ResteasyClientBuilder.newClient();
        final var multipart = new MultipartFormDataOutput();
        multipart.addFormData("foo", "foo-value", MediaType.TEXT_PLAIN_TYPE);
        multipart.addFormData("bar", "bar-value", APPLICATION_OCTET_STREAM_TYPE, "bar.pdf");
        final var multipartEntity = Entity.entity(multipart, MULTIPART_FORM_DATA_TYPE);
        final Invocation invocation = client.target("localhost:8080/test")
                .request()
                .buildPost(multipartEntity);
        // breakpoint here + inspect invocation
        System.out.println(invocation);
    }

The entity is of type MultipartFormdataOutput and it has a boundary parameter which looks like a randomly generated UUID. However, in Invocation.headers the Content-Type is set to only multipart/form-data, without the boundary.

Explicitly constructing the media type and setting the boundary as a parameter, instead of using MULTIPART_FORM_DATA_TYPE, works:

final var mediaType = new MediaType("multipart", "form-data", Map.of("boundary", multipart.getBoundary()));

Since the entity already knows the boundary, it seems like the buildPost should be able to set the boundary without explicitly constructing the media type?

[jamezp]

What version of RESTEasy are you using?

[mqchen-ws]

6.2.9.Final

[jamezp]

Sorry, I just noticed that it was the first thing in the question 🤦. I'll have a look at this.

[mqchen-ws]

No, you were right and I didn't have it there originally. I edited it for anyone looking at it in the future 😅

[jamezp]

@mqchen-ws Are you seeing some kind of an error? While the headers on the invocation may not have the boundary set, it's set in the request. If I dump the request headers I see something like:

Connection: "Keep-Alive"
User-Agent: "Apache-HttpClient/4.5.14 (Java/11.0.24)"
Host: "localhost:8081"
Content-Type: "multipart/form-data; boundary=e81e31c2-8239-4287-a17e-92c419a27dce"
Accept-Encoding: "gzip, deflate"
Content-Length: "484"

[mqchen-ws]

Ah! I simplified it down to this in my test:

        final var client = new ResteasyClientBuilderImpl()
//                .httpEngine(new JettyClientEngine(new HttpClient()))
                .build();

If that line is uncommented out and the jetty engine is used, I don't see the boundary on the header. But without that line, the ApacheHttpClient seems to do the right thing.

[jamezp]

Okay good to know. That sounds like a bug in the JettyClientEngine to me. I'll try and create a reproducer and test for it, then file an issue and get it fixed.

For my own curiosity, why use the jetty-client? I'm asking because I really want to change the default client to be something other than Apache HTTP Client 4.x and support http/2.

[mqchen-ws]

For my own curiosity, why use the jetty-client? I'm asking because I really want to change the default client to be something other than Apache HTTP Client 4.x and support http/2.

That I am not entirely sure, it was set up this way before I started working on the code 😅. I can try asking around.

[stevenschlansker]

@jamezp , we initially went with Jetty since we'd already selected it for our server (good experiences at prior projects) and liked the "clean room" design that has good support for async (as well as sync) from the start. Jetty also has good extensions for websockets and http/2, although we haven't in the end used either much yet. We also had positive experiences with the maintainers over time.

I'd had "okay" experiences with apache httpclient - it usually works at the end of the day, but really felt the "legacy" aspects at times.

I'm puzzled that multipart entities in particular don't work - maybe it needs special handling?

[jamezp]

Thank you @stevenschlansker. That is great information :)

I'm not too sure yet why the headers are behaving differently. I don't know if it's a RESTEasy issue or a Jetty Client issue. I'm leaning towards RESTEasy, but I've not yet looked at it in detail.

[jamezp]

Thank you for all the information on this. I've created https://issues.redhat.com/browse/RESTEASY-3532.

[jamezp]

This should fix it #4282. The issue was the headers were set before the content was written, which seems to be where he boundary is added to the header.