Dudes, this is so not REST
jens.mooseyard.com
jens.mooseyard.com
But then I read the article and I have to admit this is a pretty egregious break with how REST services are supposed to work.
There's a case to be made for relaxing the definition of REST to accept "just a bunch of URIs" APIs that don't follow the discoverability/transparency conventions of REST. But you can't call "here's a list of verbs" a REST interface.
What you'd want to see in a Rdio REST interface would be:
* /rdio/api/1/playlists * /rdio/api/1/playlist/foo * /rdio/api/1/artist/1 * /rdio/api/1/artist/1/tracks/2
&c &c &c. But, from what I can tell, none of the state inside Rdio's database is mapped to a URI.
You should disclaim that these URLs could just as well be
* /rdio?q=1&s=63
* /rdio?q=1&s=63-8
* /rdio/4E7?q=1
* /rdio/4E63?q=1
and still be part of a RESTful API.
Nice urls with http verbs but no hateos is what most "REST" APIs are though.
/rdio/api/1/artist/1/tracks/2
/rdio?api=1&artist=1&tracks=2
Does it change how caching might work?
the idea is that you can build a generic 'REST' client and point it to almost anything that is a 'REST API' and it will work, in the same way we can now with web browsers
a REST client will never figure out /api/1/ - which is why resources go in the path and query parameters go in as, well, query parameters :)
from the server side you are right that modern front controllers and routing can make it all the same (heck, I can route resources to different subdomains, if I want), but a proxy does not know what the front controller logic is (ie. working backwards to what part is what).
That's a stupid idea. REST is built on two things: content types and resource location. The content types indicate the meaning and location of the resources, if the client does not understand the content-type, it can not do anything.
You can not "build a generic 'REST' client and point it to almost anything that is a 'REST API' and it will work", that makes no sense. Indeed, that's not what web browsers do: the driver of the API on a website is the user, not the browser.
> a REST client will never figure out /api/1/ - which is why resources go in the path and query parameters go in as, well, query parameters :)
That's completely irrelevant.
here is what my generic REST client looks like:
* HTTP[s] * JSON * Atom and RSS * Caching layer
some controller login in between all of these. I inherit this to implement simple models and controllers in the client and just tell it which resources I want (like fields in a database)
how is that not a generic client?
for twitter I have classes called users and statuses, with each method defined. for google I have similar classes. this all took minutes for me to get working - I thought that was the entire point of REST?
read the original desertion, sec. 5.2.2 connectors:
http://www.ics.uci.edu/~fielding/pubs/dissertation/rest_arch...
> if the client does not understand the content-type, it can not do anything.
generic doesn't mean understanding the payload, it means understanding the protocol
Also, I believe safari will ignore your caching/expires headers if you use GET params.
/rdio/api/1/artist/1/tracks/2/artist/1
I just did a quick check and Django, for example, will just take the second of any duplicated GET params.
http://www.w3.org/TR/REC-html40/interact/forms.html#form-content-type
Sorry sir, I'm not wrong, those would be perfectly valid as part of a REST service, including as alternatives to tptacek's description. Readability of the URLs has nothing at all to do with RESTfullness.
The only thing REST says about the data transmitted (the URL in this case) is that it's the container for the application's state, and most interpretations of this is that the URL is entirely opaque. And indeed, Fielding has suggested/explained this several times.
Those are not REST because they are not resource-based.
Correct REST:
http://blah.com/widget/:id - GET, PUT, POST, DELETE
How does that fit with your /rdio/4E7?q=1 ? Where's the resource? What is the id? What is q anyway?
You can pass parameters with a ? in REST... for example, you might do
GET http://blah.com/resource/123?format=xml
But there's still a clearly identified resource which the GET is performed on.
Sure, technically, by stretching things a bit, you could claim that /rdio?q=1&s=63-8 is a REST GET request for resource id "63-8".
And, technically, you'd be right, much like technically, this is valid C: http://stackoverflow.com/questions/18642/what-is-the-most-un...
But just because it's valid doesn't mean it's in any way good. Those are horrible examples. So no, those URLs could not "just as well" be those nasty examples you give. They could, much more poorly, be those examples, and still be technically valid. By the same reasoning, "Buffalo buffalo Buffalo buffalo buffalo buffalo Buffalo buffalo" is just as good english as Shakespeare.
There is absolutely no difference between path segments and query parameters from the perspective of resource identification.
Your arguments about nice looking URLs in fact have absolutely nothing to do with REST, except that REST says you must use your transport protocol according to its standards, where appropriate. It says nothing about how URLs should look. I you're talking about the HTTP spec, call it by its proper name please. We don't gain much from conflating different terms like this, it's just a source of confusion.
It implies that a resource is different just because you are using a newer version of the API.
My own most recent implementations use Accept: and Content-Type to define version compatibility, which is what content negotiation is supposed to be about.
Versioning in the old web API sense meant changing the way the protocol works, and REST will always be REST and the resources will always be resources.
All that is likely to change between 'API versions' is formats, which you negotiate with headers, and adding new fields, which old clients can either ask not to get through content negotiation or can safely ignore.
APIv2 is not the same resource as APIv1, so why send it to the same location? Similarly, APIv1/user/1 may very well not contain the same information as APIv2/user/1, so are they really the same resource? They likely refer to the same semantic resource, but if user/1 != user/1, they're not the same from a data standpoint, which is what APIs are generally used for.
what 'API' is signaling is what the client is capable of understanding, which is what content negotiation headers are for
I think of it no different to the same resources for web browser consumption (for humans) - I don't place new designs of a website into /design1/ /design2/ etc.
That doesn't make REST "bad", but I think it does start to make the quasi-religious aspect of it less obviously correct.
It's a correctness vs usability debate. I'd love to go for correct, but it's extremely painful to enforce in your users in practice.
Anything else requires the client knowing how to mangle URIs (e.g. adding /avatar.jpg, /contact.vcf).
It has a dubious track record in X as well. X has been able to do content negotiation for decades for the clipboard but even now it's still spotty and buggy, and that's with everything actually running on the same system. It looks easy. Heck, it looks very easy. It isn't. I can't even tell you exactly why it isn't as easy as it looks (beyond the semantic thing I mentioned above), but the state of the world is pretty clear.
eg.?v=1&format=xml
but interesting to note is that why is encoding easy to do as part of content negotiation, yet everything else is hard? you never see ?e=UTF-8 or &c=gzip
The challenge is some HTTP library APIs commonly used for developing clients offer the following interface: - GET $URL
If you are lucky, then it might also support POST and offer a second parameter for the request body. If are really lucky, there is a way to specify arbitrary methods, but this is fairly rare.
The correct method for selecting a specific representation is to use the Accept header on the request. For example, if I were requesting somedomain.com/Groxx I could specify Accept: image/png if I wanted a picture of you, or Accept: text/x-vcard if I wanted your contact information.
"The Accept request-header can be used to specify certain media types which are acceptable for the response."
Similar to how when you access some linked data URLs you can specify rdf/xml or rdf/n3 in the Accept header to receive the same data in a specific format.
It's not used to specify whether you want a picture of someone or their contact information all from the same URL.
Clearly it is a pragmatic trade-off and is popular (Google and Twitter do it). The correct answer will be arrived at by considering the total architecture of your application.
You should also question whether you really need multiple representations for each resource -- in many cases, it's more trouble than it's worth to automatically generate, say, both JSON and XML responses, since it just divides your attention to creating sensible serializations for your API.
If you expect to have clients which cannot modify the headers of their requests, you can support fallbacks like specifying the Accept header in a query parameter. Putting it in the actual URL though is doing a disservice to fully compatible HTTP clients, though.
But if JSON is so inherent to the resource that content negotiation doesn't even matter, it can be fine to include that in the URL, as you might include descriptive filename extensions like ".png" or ".wav" in a URL.
I backed away from using content-negotiation for an API, part of which would have served data and part content.
e.g. to access file metadata, get /path/to/file with Accept: api/v1 and do a 'normal' GET to get the file contents.
I came across some suggestions that some browsers (recent IE I think, unsure) which suggested that cacheing was broken wrt Accept header - which made me worry that people hitting accessing the resource might get the wrong version.
I don't know if your api is vulnerable to this, if you've come across the problem or have any deployment wisdom to share.
What most people talk about when they say REST are good HTTP practices. Meaningful URLs, proper methods, etc... are all about using HTTP properly. We can call some of these views other acronyms like ROA but it's really just appreciation for proper and modern HTTP stacks.
REST was extracted from HTTP. It's not that HTTP was influenced by REST, it's the exact opposite, REST is a formalization of HTTP.
If it were an extraction of anything, it wouldn't have been HTTP but existing ideas in scaling web architectures in general at the time.
The crowd that mistakes ROA for REST (I think) sees the stuff in REST that relates directly to development and gets excited - but the stuff related to making their app scale, they ignore. If you don't want to learn a little bit about everything, at least follow the rules that relate to the things you don't know about.
I personally find that sometimes I have trouble spotting the "oh! I forgot to make it DELETE instead of GET" bugs. When the parameters of the API are all in the url and/or query, it feels better encapsulated.
(promise I'm not flaming, and I fully expect to get told, yes it matters, and here's why! ;)
I often think that, via REST APIs and content negotiations, API can become browseable and self-descriptive but I've yet to see any example of such a thing. Furthermore, just because an API is restful doesn't mean it's well-designed, and doesn't mean it has descriptive APIs (the artist's track could just as well be in rdio/api/64D89CAC?q=42, that wouldn't inherently make the API non-restful)
It should be trivial with AtomPub/GData/OData APIs. REST itself is not specific enough for that.
Note: Requiring that the call are OAuth'ed, doesn't affect its RESTness.
If you're expecting a REST API yes. It also significantly impairs the discovery and the cacheability of the API calls.
What REST brings to table is exchanging representations. For one thing, it makes your API actually simpler, since you don't even need a list of 'methods' like rdio has, all you need to know is the representations and methods stay the same—which should come natural if you know HTTP.
The other nice thing that you get for free is that it accommodates for unreliable connectivity. In case of failure you can just retry. OTOH RPC abstracts such things away. For example if you have a call like "addSomethingToList" and you retry that method, you risk having duplicates.
Except those are of course guidelines, the API implementor is free to entirely disregard them (much to the danger of everybody)
> For one thing, it makes your API actually simpler, since you don't actually need a list of 'methods' like rdio has, all you need to know is the representations and methods stay the same—which should be natural if you know HTTP.
Instead of a list of methods, you get a list of methods (which do not have to stay the same, by the way) and a tree of resources. Theoretically you should be able to crawl a REST API using only the document types and the root URL, I don't think I've ever seen a "public" API letting me that.
> OTOH RPC abstract such things away. For example if you have call like "addSomethingToList" and you retry that method you risk having duplicates.
Yes, because if you POST the same thing twice you don't risk fucking things up...
Oh you should be using PUT, not POST, you say? Good luck getting that one in browsers.
Witness the Google Web Accelerator debacle.
> Except those are of course guidelines
Well, obviously.
> Instead of a list of methods, you get a list of methods (which do not have to stay the same, by the way) and a tree of resources.
Well, yes. But those are already established and if you're exchanging representations as REST demands, you shouldn't need anything more than GET, POST, PUT, DELETE and PATCH to manipulate data.
> Theoretically you should be able to crawl a REST API using only the document types and the root URL, I don't think I've ever seen a "public" API letting me that.
Sun tried to do that for their cloud computing API, I think they're the ones that created the term RESTful where you'd use elegant concepts like <link> in your representations to advertise available actions.
> Yes, because if you POST the same thing twice you don't risk fucking things up...
If you do it properly with versioning/unique ids where appropriate, then all your calls should be re-triable or at least fail.
> Oh you should be using PUT, not POST, you say? Good luck getting that one in browsers.
You can create browser API (with proper CSRF protection etc. btw) that acts as a [server-side] client to your REST API. Also I see most usage of any REST APIs from browsers coming from Javascript based clients, which usually don't impose such limitations.
REST and pretty URLs are very much orthogonal indeed.
Oh yes you have. Ever browsed a ftp server with your browser? That is a REST API, using HTTP and FTP for transportation and HTML and WHATEVER for the actual data. You can start from the root of the server (or more commonly, from /pub/) and crawl through the file hierarchy as much as you like.
But your point is valid. Not many APIs allow to "crawl" the data, even though one of the four guiding principles for interface is: Hypermedia as the engine of application state "If it is likely that the client will want to access related resources, these should be identified in the representation returned, for example by providing their URIs in sufficient context, such as hypertext links."
REST is an architecture style, not a protocol. HTTP is protocol well suited for use in a RESTful system. The API in question violates REST (atleast) by not using the methods provided by HTTP.
Pedantry: because they're supposed to be idempotent, POST, PUT, and DELETE should only have side-effects the first time they're issued against a given resource.
Also, RFC 2616 uses "SHOULD NOT", as opposed to "MUST NOT" to describe the behavior of the "safe methods". To wit, "in particular, the convention has been established that the GET and HEAD methods SHOULD NOT have the significance of taking an action other than retrieval."
This is why your web browser issues "need to resend a form" warnings when you return to earlier pages.
Thanks for the correction.
Yes. You can use libraries when the service you're trying to interact with works according to some standard.
An example: I have an app I wrote at work that is a JavaScript frontend powered by a REST API backend. The REST API tries to be a good REST citizen: proper HTTP implementation (201 + Location when something is created, 202 for "I will get to that eventually", etc.), consistent URL conventions, and a consistent data format.
The end result is an API that is easy for humans to understand, but also easy for computers to understand: the simple CRUD code was about 10 lines of backbone.js, because my API doesn't break any conventions. This means that the code to interact with it already exists, and all I had to do was make the web page look pretty.
This is a good use of programmer resources, now and in the future. In the present, I didn't have to do any work to make a pretty GUI. In the future, someone that wants to pull data from my API will not have to swear at me -- everything you can do is easily discoverable by starting from the root resource.
I died a little inside every time I wrote a 'Restful' servlet there.
Google created GData2 as an extension to AtomPub format, which only specifies a common envelope for resources (Atom Syndication Format) and how to talk about collections of resources (AtomPub itself).
There is another notable extension to AtomPub format from Microsoft called OData.
1. custom methods let you get/put specific fields, without having to transfer the whole thing or split resources up as you add new methods.
2. custom methods avoid needing to implement complex or heavy state mutations on the client. the example was rotating an image 90 degrees; you wouldn't want to download an entire full-res image to your wimpy battery-powered phone, rotate it, and re-upload it when you can just do that on the server.
(... of course this should just update some orientation metadata instead of resampling the image -- it's just an example)
GET /resources/picture/orientation => 0
PUT /resources/picture/orientation?value=90
To be fair, I thought we were at this point already. It seems safer to assume that when one advertises a "RESTful API" they are really referring to a "REST-like" API which basically means "Hey, we don't use sessions!"
While I believe that the term is already diluted to that point, I still think it's valid to point out when people are doing it wrong.
My main problem with its dilution is that it's usually totally meaningless -- usually other terms like RPC or just plain old "proper HTTP" accurately describe what someone is calling REST, to there point where its supposed RESTfulness can be nothing more than for the sake of buzzword compliance.
Most REST API consumers today are more like CURL than a web browser. In the CURL-like case, the RPC pattern works quite well. HATEOS and other browser-centric REST ideas often just don't fit naturally into a simple API consuming app. People need to move on.
The integrity of some acronyms is worth defending. I wouldn't want 'ACID' diluted to mean anything less than its strict definition, and I think the same holds true for 'REST'. The benefit of these distinctions is twofold: 1) we encourage the advantages of good RESTful practices and 2) we can communicate consistently without spending unnecessary effort qualifying our meaning. If REST means 'Anything but SOAP' then at some point we'll have to come up with another shorthand for 'Real REST' or spend lots of time re-stating the principles of REST every time we need to communicate architectural strategies that use them.
I should be able to say 'Singleton' or 'MVC' or 'ACID' and know that there is enough integrity to those ideas that a competent coder will know what I mean without me having to pseudo-code or UML the concept for them. The same should hold true for REST.
So you're telling me I should call my XML-RPC APIs REST?
-- Roy Fielding @ http://roy.gbiv.com/untangled/2008/rest-apis-must-be-hyperte...
If you want a bucket of cold water over the head on this topic: http://roy.gbiv.com/untangled/2008/rest-apis-must-be-hyperte...
And what's the use case anyway? Are you going to perform those requests from other domains and authenticating via cookies on your own? How do you protect from CSRF doing that? What's wrong with OAuth 2.0 + JS?
It is not. It's quite complex actually. And I'm betting what most people who believe they have a clue think of "REST" is actually "RPC over pretty urls".
Actually some Google engineers do a very good job of explaining how AtomPub and GData APIs work just with short Youtube videos.
http://www.youtube.com/watch?v=Sv5woNs9WRE
Being accurate with a technical term is not a 'fetish.'
This is a shame, because RESTful APIs are (for me, anyway) usually a joy to build and to use.
I mean, come on, how can you say you have REST (with R standing for Representational) if you have section called “Methods” in your API documentation. HTTP API sounds okay, but in fact their API is not even HTTP specific (actually, it abuses HTTP).
Well except RDIO's API is not any simpler to use than SOAP.
It's RPC to and through, let's call a procedure call a procedure call.
All the WSDL and SOAP crap gets auto-generated for you. Thereby eliminating in a single stroke REST's advantage.
Now, it is true that there is some stuff†† you can do with SOAP that you can't do with JAX-WS, but really unless you like making life difficult for yourself then don't.
Especially for a big Java to Java solution there doesn't seem to be any compelling reason to go with REST anymore. Even if everything you do maps exactly to the four SQL ops (select, update, insert, delete) it is still just as easy to use the annotation as it would be to use REST.
†C# has something similar I think.
††Example: JAX-WS is limited to what you can do with a method and parameters in Java. In your SOAP schema you can specify for instance that the array you get passed shall have between 1 and 3 members, two is okay but 4 is right out. Whereas in Java (and hence JAX-WS) if you have an Array parameter, you can't specify that it must have a certain number of elements in it.
Just one line for SOAP from stored procedure or function.
Is it beautiful? I don't think so. SOAP remains SOAP, ugly remains ugly.
Is adding new HTTP methods (like PROPFIND, MOVE, COPY, etc.) considered to be RESTful? I always thought you were supposed to structure all actions around the main 6, and add extra functionality in the data payload or query parameters.
Does he have a source for this? Coincidentally I had just read the HTTP 0.9 the other day and I didn't see "method" being used that way at all. I think method, in HTTP, just means "a means to retrieve / put data", not method in the object oriented sense.
Rdo: "Yeah, we have POST!"
Thought Palace: "What about the rest?"
For APIs that need regular browser support, it's fine to add fallbacks like query params to specify PUT or DELETE. But these are hacks.
"HTTP API that’s not as hard to use as SOAP"
My current employer calls our API ...
... RESTful like .
I'm not so sure that "object-oriented" is the right term here, but totally agree with the premise: this isn't REST. It's more like XML-RPC Jr.
They're particularly bad since a strict interpretation of REST can still be valuable, yet it makes REST itself seem overly nebulous and fluffy.
People advocating terms like "RESTish" or looser definitions of REST seem to think that REST is an ultimate goal of any decent web API. It's not, REST doesn't need to be used everywhere, your API isn't crap if it's not REST. So stop calling it REST like you're embarrassed to admit it's "just" RPC.
(Sorry for ranting in reply to you -- you seem to get it)