Support reading request body in custom middlewares about conjure-go-runtime

[whickman]

yep that is a mis-characterization on my part, sorry. Request is certainly fully in memory before being written to buffer and we would be creating a second copy.

That's a fair point. Given that and need for signed requests etc, supporting reading the body is reasonable. I'll follow up off the issue on prioritization / implementation

Not a huge fan of the amount of optionality we are providing people in middleware given how much skew that is causing across consumers, but that is a conversation for a different time.

(CGR already supports reading the response body in the error middleware, which seems potentially dangerous to me anyway...)

[whickman]

reading the request body means you'll need to load the entire request into memory, rather than allowing it to be read incrementally, as is done now. For that reason, we want to discourage middleware working with the request (and response) body

[avisbal]

cc: @najork

Is the request body actually read incrementally? It looks like it's encoded to a buffer in bodyMiddleware, which I thought would write the entire encoded request body to the buffer in memory.

bodyMiddleware sets request.GetBody, whose intended purpose is to allow reading the request body more than once client-side, right? I think it's appropriate to note that using that method will increase memory utilization, i.e. because it creates a copy of the request body, but that seems like tradeoff that can be made when authoring or using middleware as opposed to restricting it from being used.