Read body for all requests with a Content-Length header

[jgraham]

The framing of a HTTP request doesn't depend on the method [1], so any request with a non-zero Content-Length has a corresponding body. This does not change in the case of idempotent methods like GET; what is different is that for such methods it is incorrect for the server's response to depend on the content of this body.

At the moment GET and HEAD are incorrectly special cased in hyper [2] leading to problems if you send a GET request with a body (the unread body bleeds into the following request unless the connection is closed). In these cases the body data, if any, must be read from the network but could be discarded and not made avaliable through the request API.

[1] https://tools.ietf.org/html/rfc7230#section-3.3
[2] https://github.com/hyperium/hyper/blob/master/src/server/request.rs#L42

[seanmonstar]

Good catch. I've also noticed that for requests that should have a body, such as POST, and the user doesn't read it out, it bleeds into the next request also.

[reem]

We could write a Drop impl for Request that tries to read out the remainder of the body.

[seanmonstar]

cc #197

[seanmonstar]

The patch was reverted, so not fixed.

[reem]

If we are going to do this, it has to be configurable (the max amount we read before shutting down the connection).

[durka]

We should fix this, or advertise the behavior in blinking red marquee in the documentation. I had no idea where to start when I ended up with order-dependent 404 errors in my application (because one POST form submission had content, and the next one had an invalid HTTP method due to this bug).

What scenarios motivate not wanting to read the entire request out of the network buffer?