Hacker News Reader

Comment by derefr | Hacker News Reader

I'm surprised this hasn't been trivialized/automated away yet (or has it?) Some sort of language-neutral REST-protocol description language, that could be used to code-generate client libraries and also generate a test spec and documentation. Or maybe just a standard for media-types in HATEOS response types in specifying REST API "roots", to make them machine-discoverable, in a way where the result could be "cached into" a client-library.

View on HN

This is a shameless plug, so I'll take my beating if it's appropriate. For the last few months, I've been working on a Python library, beekeeper, that's all about that. It essentially consumes a JSON description file that gives all the specs for an API, as well as information about its object structure, and then provides an intuitive, native-Python interface, typically with little to no additional code required.

You can check it out on GitHub at https://github.com/haikuginger/beekeeper, read more at http://beekeeper.rtfd.org, or just install it with pip.

I'm biased, but I think it's pretty cool.

I believe the current most popular flavor is Swagger ( http://swagger.io ) at the moment.

Despite having such quirks as: no Null support =|

IME, especially building APIs for mobile apps: nulls suck anyway!

Yeah, but nonetheless, are sometimes necessary. The suggested way is to just remove the field entirely, which is really a different sort of null.

Also, generic mapping types are second class citizens, id est just wanting to specify string -> string

I appreciate your point of view, but mine is different. For example, at Trulia, with a dozen different apps with millions of users and rich data in the app, there is a "no nulls" policy. It's just a convention: build sparse data structures instead of passing nulls.

So messing around with Python and rest APIs... Often a wrapper library for an API exists on github.. But why use it? If I use 6 APIs in my project I now just added 6 libraries that all do things like auth, connection pooling, error handling a little bit different. Meh. The actual work of code to json to API to json to code is simple (1 line usually), I would rather just learn the API and do it myself. You still need to learn the API to use the wrapper library...

Sounds like the soap protocol. I've had to deal with it recently at work and paired with the suds python library auto generates the api for you

Sounds suspiciously like SOAP to me.

I could maybe see something like http://unirest.io or, to a lesser extent, http://swagger.io making sense.

Have you tried using Swagger-Codegen (https://github.com/swagger-api/swagger-codegen) to generate API clients, documentation (markdown) and test spec?

Disclosure: I'm a top contributor to Swagger-Codegen

I believe this is what AWS does to create all their libraries in the bazillion languages they support.

Honestly, unless the API is complex, why bother?

I feel like abstracting APIs into objects (essentially RPC calls?) creates too much of a dependency on the network. The friction of writing the calls and their handlers helps minimize their use :)

Mainly because I code microservices in at least five languages, and want to be able to touch all the same third-parties from any given one of them.

Reply on news.ycombinator.com