MockWebserver fails for empty body requests with "Transfer-Encoding: chunked" header

[bclozel]

I'm using an HTTP client that sends a "Transfer-Encoding: chunked" HTTP request header for all requests, including with empty bodies. When testing that client with MockWebServer, I'm getting the following:

MockWebServer[54824] connection from /127.0.0.1 crashed
java.lang.IllegalArgumentException: Request must not have a body: GET /pojos HTTP/1.1
    at okhttp3.mockwebserver.MockWebServer.readRequest(MockWebServer.java:631)
    at okhttp3.mockwebserver.MockWebServer.access$1500(MockWebServer.java:99)
    at okhttp3.mockwebserver.MockWebServer$4.processOneRequest(MockWebServer.java:498)
    at okhttp3.mockwebserver.MockWebServer$4.processConnection(MockWebServer.java:460)
    at okhttp3.mockwebserver.MockWebServer$4.execute(MockWebServer.java:401)
    at okhttp3.internal.NamedRunnable.run(NamedRunnable.java:33)
    at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
    at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
    at java.lang.Thread.run(Thread.java:745)

Why is this enforced? Is there a strong reason or something in the HTTP spec that qualifies those requests as invalid? Could we relax that check if not necessary?

[JakeWharton]

The RFC states that some servers might reject a GET with a body but that it otherwise has no defined semantics. Does the body have contents of 0\r\n?

[bclozel]

I guess you're right. The RFC states that:

The presence of a message body in a request is signaled by a Content-Length or Transfer-Encoding header field. Request message framing is independent of method semantics, even if the method does not define any use for a message body.

So in that case, I guess the client should at least send a 0\r\n body to send an empty body if nothing has been sent.

Right now the client sends the following:

GET /greeting HTTP/1.1\r\n
transfer-encoding: chunked\r\n
host: localhost:50419\r\n
\r\n
0\r\n
\r\n

Is there something wrong with this request?

[JakeWharton]

Doesn't look like it, no. We can potentially update MWS to correctly ignore bodies on non-body HTTP methods, but we'll have to look more closely at the RFC to see if there are specific HTTP methods where the body is completely forbidden before doing so.

[bclozel]

The spec is actually quite liberal. GET requests may have bodies but they can be ignored by the server. AFAIK, the only request method that forbids request bodies is HEAD.

[anderius]

I have the same problem. Would it be a solution to make it configurable?

[yschimke]

Won't fix. We reject for clients because we are opionated, doing the same here.