public interface for regression tests?

[clarkevans]

Currently, Mux doesn't export http_handler in its public interface, yet, something like it is required to build regression tests that don't open a socket. Indeed, opening a socket isn't really required to test Mux... you just send a Request and get a Response object. Here is a rough stab at a trivial wrapper to make get and post requests where one could type get("/") to print the given response to stdout.

@app APP = ...
loopback(req::Request) = 
  read(IOBuffer(HTTP.handle(Mux.http_handler(APP), req).body), String)
wget(url) = println(loopback(HTTP.Request("GET", url)))
post(url) = println(loopback(HTTP.Request("POST", url)))

Something similar might be nice to include so that application developers could test their interfaces w/o having a webserver active. Generally, we might wish to also consider updating regression tests in this project to use a similar method. There's no real need to test most of the functionality via sockets. At this level in the stack what's matter is functionality via Request/Response.

[cmcaine]

Yes, I think we should at least document how to test without opening sockets.

I don't think this is only useful for regression testing.

All you need to do is response = yourappname.warez(HTTP.Request(blah)) but the warez field is private at the mo. (The http_handler stuff doesn't do all that much, really).

I don't know what the public interface should be, opinions welcome or I'll just experiment :)

The normal way to get a string from a vector of bytes would be String(bytes) or String(take!(bytes)), btw. No need to wrap in an IOBuffer. Or you can copy(bytes) if you don't want it mutated.

[clarkevans]

Perhaps make Mux.App callable?

 (app::App)(req::Request)::Response = app.warez(req)
 (app::App)(method, location) = app(Request(method, location))

Oh, I dislike http_handler magic with regard to converting a Dict to a Response object anyway. That might be a good candidate for removal.

[cmcaine]

I talked about removing mk_response in another PR and might do that.

Making App callable is what I was thinking of doing, too, but I haven't thought it all through yet :)

I'd probably just define this if I did:

(app::App)(x) = app.warex(x)

I don't think it's our responsibility to provide easier ways of creating HTTP request objects, or if we did do that, it can be in a Mux.Request function or something?

[clarkevans]

I think mk_response is rather specialized and could be implemented in user code if someone wanted, so yea, I support removing it as #129 mentions.

Making the definition, (app::App)(x) = app.warex(x) seems obvious/useful.

It's not Mux responsibility to create HTTP request objects. There is one bit of code that checks if an object is showable via "text/html, setting "Content-Type" appropriately -- that might be useful to keep.