Would using a WSDL to generate REST clients be the wrong direction?_问答_开发者

I'm out to create a fairly simple web service for intranet use. The service will ultimately be the interface to a DB that will allow me to keep track of what the various internal tools within the company are doing. I'm thinking I want a web service so that various tools (and thus different l开发者_开发技巧anguages) within the organization can easily update the DB without needing to know the schema.

I've read many of the related REST vs SOAP questions already that come up from searching https://stackoverflow.com/search?q=soap+rest but I'm not sure I've found my answer.

My dilemma seems to be that I want the simplicity of REST while also having the code generation abilities of a WSDL, which seems to imply SOAP. It is of utmost importance to me that various internal tools (JAVA, Perl, Python, PHP, C++) be able to talk to this service and it would seem silly to have to rewrite/maintain the interface layer for each of these languages manually when the WSDL route would do that for me. From what I can tell, if the WS needs to change, you would update the WSDL, regenerate the client stubs, and make any changes necessary to the code that uses the stubs (which would need to be done anyway).

For example - say I have a tool written in JAVA that utilizes a RESTful web service. I imagine that the tool will have specific code for dealing with certain URLs, launching the request, doing something with the response, translating that into some data structure if I'd like, etc. Now lets say I also have a Perl tool doing the same thing. Now I'll need Perl code to do the same, make requests on specific URLs get the responses, do something with them, etc. In each case, and thus in C++ and Python and C#, where code cannot be shared, eventually I'll end up with a wrapper classes/methods that hide a lot of that ugliness from me. I'd much rather call a function on a class that returns my data encapsulated in an object instead of having to worry about the URL, the arguments, the response, etc. Sure, maybe its not a lot of code in any particular place but it starts to add up over time. Multiply that out over each of the tools and now when I make a change to the service I must go update the URLs in each CRUD operation and all that goes along with that. I guess I imagine that with a WSDL that is the aspect that is done for you. Your code simply interacts with the stubs. What the stubs do, who cares? Urls, arguments, responses - if anything changes just regenerate the stubs from the WSDL. If that process causes your code to break, so be it, but at least I didn't also have to update all the code that deals with the specifics of making the request and dealing with the response. Is this really not a problem? Perhaps what I need to do is just create a service and a few clients and see what I'm really up against.

While I'm a fairly seasoned programmer with experience in JAVA, Perl, Python, C++, etc, this is the first time I've considered authoring a WS and don't have prior experience with other WSs, so I'm looking for some guidance. Do I just go with WSDL/SOAP and forget about what everybody is saying about how popular, simple, and useful REST is?

I don't get the code generation issue here.

REST rarely requires any code generation of any kind. It's just HTTP requests with simple JSON (or XML) payloads.

You use the existing HTTP libraries (e.g. Apache Commons, or Python urrlib2). You use existing JSON (or XML) libraries. (the Jersey project has a nice JAXB-JSON conversion, for example).

What's "generated"? Our RESTful library in Java and Python are nearly identical and simply make REST requests through the HTTP library.

class OurService {
    def getAResource( String argValue ) {
        path = { "fixed", argValue };
        URI uri= build_path( path );
        return connection.get( uri )

[I've left out the exception handling.]

Are you trying to layer in the complex SOAP interface/implementation separation?

A client "written in JAVA that utilizes a RESTful web service"... A "Perl tool doing the same thing" ... "in C++ and Python and C#".

Correct.

"where code cannot be shared"

Correct. The code cannot be shared. You have to write each client in the appropriate language. Writing some kind of "generator" to create this code from WSDL is (1) a huge amount of work and (2) needless. Each language has unique syntax and unique libraries for making REST requests. But it's so simple and generic that there's hardly anything there.

The canonical example in Python is

class Some_REST_Stub( object ):
    def get_some_resource( self, arg ): # This name is from the WSDL
        uri = "http://host:port/path/to/resource/%s/" % arg # This path is from the WSDL
        data= urllib2.open( uri )
        return json.load( data )

This block of code is shorter than the WSDL required to describe it. And it seems easier to change because the name is the method name and the URI is simply a string.

In most languages the client is approximately that simple. I just wrote a bunch of REST client code in Java. It's wordier, but it's generic stuff. Build a request, parse the JSON that comes back. That's it.

A RESTful WSDL declaration buries two pieces of trivial information in a lot of XML.

It provides an "interface name" for the URI.
It documents the meaning of GET, PUT, POST and DELETE by providing Stub class method names.

It doesn't help you write much code, since there isn't much code. Note that all REST requests have the same generic HttpRequest and HttpResponse structure. The request contains generic an Entities. The response, also, contains a generic Entity that must be parsed. There's very little to REST -- the point is to be as simple as possible.

It largely eliminates the need for WSDL since everything is a generic JSONObject or XML URLEncoded and sent as a string.

"I'd much rather call a function on a class that returns my data encapsulated in an object instead of having to worry about the URL, the arguments, the response, etc."

Correct, you'll have to write a "stub" class. It has almost no code, and it will be shorter than the WSDL required to describe it.

"Multiply that out over each of the tools and now when I make a change to the service I must go update the URLs in each CRUD operation and all that goes along with that."

You can either update the stub class in each language for each client. Or you can update the WSDL and then regenerate each client. It seems like the same amount of work, since the URI is trivially encapsulated in the client code.

"I guess I imagine that with a WSDL that is the aspect that is done for you."

I'm unclear on what's "done". Translating the wordy and complex WSDL into a simple stub class? I suppose this could be helpful. Except the WSDL is bigger than the stub class. I'm suggesting that you'll probably find it easier to simply write the stub class. It's shorter than the WSDL.

"Your code simply interacts with the stubs."

Correct.

"What the stubs do, who cares? Urls, arguments, responses - if anything changes just regenerate the stubs from the WSDL."

Sadly, almost none of this requires any real programming. Generating it from WSDL is more work than simply writing it. URI's a strings. Arguments are generic JSONObjects. Responses are generic HttpResponses including a JSONArray. That's it.

"I didn't also have to update all the code that deals with the specifics of making the request and dealing with the response."

Except, there aren't any interesting specifics of making the request. HTTP is simple, generic stuff. GET, POST, PUT and DELETE are all nearly identical.

Fossill,

I recommend that you do not bother to learn SOAP for this. Ws-* has a very high learning curve and the (unnecessary) complexity will likely eat you alive.

Looking at your skill set (Java, Perl, Python, C++) you should be very satisfied with a REST (or at least HTTP-based) approach. And: you'll get results very fast.

As S.Lott said, do not worry about the code generation. You'll not need it.

For questions, I suggest you join rest-discuss on Yahoo groups: http://tech.groups.yahoo.com/group/rest-discuss/ You usually get immediate help with all things REST there.

Personally, I have yet to see any use case that could benefit from using WS-*.

Jan

The code-generation aspect that you value is a maintenance item, but I question its real worth.

Be it thru a WSDL document or your own grammar documentation for a REST-style implementations, clients have to comply to the published interface. The WS/SOAP (code-generation) development model might have more tools, but I think that's because it's clunky and needs them.

There's no impact in the 'integrateability' of your web service. That's an issue of publishing the formal interface (in whatever form that takes), in either case. And the speed with which you move from an interface change to an implementation change is arguably faster with REST-style services. Firing up (and fighting with) WS-* code generation tools takes time.

FYI - REST does have a WSDL-like auto-generation schema definition called WADL. But almost no one uses it.

Apache CXF has Java support for REST clients that gives you the same sorts of 'code generation' advantages, in many cases, as full SOAP.