gistfile1.txt

8:24:20 AM Aaron Gotwalt: we're having the collection-json vs other api forms argument tomorrow
8:24:36 AM Aaron Gotwalt: or hypermedia types, i suppose
8:25:00 AM Aaron Gotwalt: he's concerned about verbosity and that it presents the data in a format that is not ideal for either the client or the server
8:25:33 AM Larry Marburger: that's a feature
8:26:09 AM Larry Marburger: it's designed to be future proof which hand-rolled json won't be
8:26:19 AM Larry Marburger: it's an interchange format
8:26:31 AM Larry Marburger: i may have just made that up
8:26:34 AM Aaron Gotwalt: right, but it has processing cost on either side
8:26:39 AM Aaron Gotwalt: which is less of a concern in ruby-land
8:26:43 AM Aaron Gotwalt: but on ios
8:26:46 AM Aaron Gotwalt: with old devices
8:26:56 AM Larry Marburger: you're not really telling me it's expensive to parse cj, are you?
8:27:03 AM Larry Marburger: there's no way i'd buy that without benchmarks
8:27:06 AM Aaron Gotwalt: not a single data node
8:27:14 AM Aaron Gotwalt: yeah, fair
8:27:29 AM Aaron Gotwalt: i think my biggest cj love actually comes from inline collections
8:27:34 AM Aaron Gotwalt: i implemented it last week
8:27:41 AM Aaron Gotwalt: makes so much sense
8:27:45 AM Larry Marburger: sweet
8:27:57 AM Aaron Gotwalt: allows you to preserve normalization across your endpoints
8:28:14 AM Aaron Gotwalt: but without an insane http request cost
8:31:07 AM Larry Marburger: that
8:31:28 AM Aaron Gotwalt: really, my gripes just come down to the the data representation
8:31:46 AM Aaron Gotwalt: the first thing i do on either side is inject to get back to a hash
8:31:46 AM Larry Marburger: that's why i like it because it's easier that everything uses the same message format
8:32:01 AM Larry Marburger: you have that with anything, though
8:32:08 AM Larry Marburger: it's just abstracted
8:32:12 AM Aaron Gotwalt: well, why not just ship a hash across the wire then?
8:32:22 AM Aaron Gotwalt: metadata on elements?
8:32:35 AM Larry Marburger: then you're talking about remote objects which is completely different 
8:32:59 AM Aaron Gotwalt: you just jumped across a philosophical line that isn't actually real
8:33:00 AM Larry Marburger: or more like rpc
8:33:21 AM Aaron Gotwalt: assume that the data coming across the wire is not a 1:1 representation of whatever's on the other side
8:33:25 AM Aaron Gotwalt: it's not a remote object
8:33:49 AM Aaron Gotwalt: it's serialized in the form of a hash, because it's a friendly representation for both client and server
8:33:57 AM Larry Marburger: oh i thought you were saying why not just serialize objects
8:34:11 AM Aaron Gotwalt: as well as reducing verbosity across the wire, which for 3g networks matters
8:34:51 AM Larry Marburger: but once it's gzipped, i don't think verbosity matters
8:35:12 AM Larry Marburger: or do you mean verbosity in the number of requests needed?
8:35:28 AM Aaron Gotwalt: inline helps solve the request problem
8:35:40 AM Aaron Gotwalt: even if you assume the verbosity argument is bunk due to gzip
8:35:46 AM Aaron Gotwalt: you still have this non-ideal middle format
8:35:52 AM Aaron Gotwalt: that is a middle format of a middle format
8:36:00 AM Larry Marburger: so anything you send is converted to bits
8:36:02 AM Aaron Gotwalt: why go so abstract
8:36:11 AM Larry Marburger: there's always a lower abstraction
8:36:22 AM Aaron Gotwalt: i can't come up with a compelling argument to annoy every developer who has to consume or produce this data
8:36:45 AM Larry Marburger: well you shouldn't use anything you don't think is delivering value
8:37:01 AM Aaron Gotwalt: well, everything until the data node delivers value, from my perspective
8:37:18 AM Larry Marburger: what about that?
8:37:39 AM Aaron Gotwalt: it throws out the elegance of native json, replacing it with a soap-like k/v structure
8:37:46 AM Aaron Gotwalt: no hierarchy kills me
8:37:58 AM Aaron Gotwalt: the lack of sub-arrays
8:38:14 AM Larry Marburger: the only thing i can say is it's designed that way for a reason 
8:38:20 AM Larry Marburger: but it's not a one size fits all solution 
8:38:23 AM Aaron Gotwalt: yeah
8:38:40 AM Aaron Gotwalt: that i get. i'm just trying to understand what those macro reasons are
8:38:50 AM Larry Marburger: and if you're really running into annoyances in your business logic because of CJ, that's an architecture problem
8:38:52 AM Aaron Gotwalt: because i'm going to have to defend my perspective on c+j to sam tomorrow
8:39:10 AM Larry Marburger: cj is a wire format. your business logic should not care at all about that
8:39:57 AM Larry Marburger: the fact that a json object can mostly be turned into a ruby hash is kind of an accident
8:40:53 AM Larry Marburger: you should still have something sitting on the boundary of your service that knows how to convert to and from whatever format you use
8:40:54 AM Aaron Gotwalt: well, they implicitly contain similar design objectives
8:41:58 AM Larry Marburger: right but converting json to a hash and passing it around may be easy but that doesn't mean it's a good idea from a design perspective 
8:43:38 AM Larry Marburger: what i'm saying is you want some object that hides the underlying format
8:44:17 AM Aaron Gotwalt: hmm
8:45:05 AM Aaron Gotwalt: and where do you draw the line of over-abstraction?
8:45:10 AM Aaron Gotwalt: i mean
8:45:35 AM Aaron Gotwalt: assume we generate a json hash for data elements, rather than k/v
8:45:49 AM Aaron Gotwalt: assume that it's not a representation of the back-end data structures
8:46:12 AM Aaron Gotwalt: but "an object that hides the underlying format"
8:46:18 AM Aaron Gotwalt: and that we send it down the wire
8:46:53 AM Aaron Gotwalt: is that not a morally equivalent data structure? we still have to agree on how to parse it, but the end result is approximately the same.
8:47:59 AM Larry Marburger: does it really matter to you if the format is sent in js object format or nested arrays?
8:48:16 AM Aaron Gotwalt: practical example
8:48:22 AM Aaron Gotwalt: you have a node that contains an array of strings
8:48:22 AM Larry Marburger: in ruby, you'd do JSON(body) and get a hash
8:48:31 AM Aaron Gotwalt: how do you represent that? it isn't clear
8:48:35 AM Larry Marburger: so if you has CollectionJSON(body) and get a hash, wouldn't it be equal?
8:48:41 AM Aaron Gotwalt: aside from another http request
8:48:53 AM Aaron Gotwalt: to a nested resource that's a collection containing just strings
8:49:02 AM Aaron Gotwalt: which is hella complicated for an array of strings
8:50:27 AM Larry Marburger: does inline help you?
8:50:58 AM Larry Marburger: i guess i'd need to see the exact data
8:53:00 AM Aaron Gotwalt: i think i just arrive at it feeling aggressively prescriptive in its abstraction religion, but without an actual forseeable payout
8:54:24 AM Larry Marburger: it's designed for easy parsing
8:54:37 AM Aaron Gotwalt: so is a blob of json
8:54:38 AM Larry Marburger: that's why everything's an array
8:55:08 AM Larry Marburger: not really. you have to have something that makes sense of the object from the resulting json string.
8:55:49 AM Larry Marburger: for example
8:56:02 AM Aaron Gotwalt: I'm gonna put this up on branch
8:56:05 AM Aaron Gotwalt: this is a really interesting thread
8:56:21 AM Larry Marburger: you have an api that returns stuff. it could either be one or more things. do you return the one thing as a single object or do you return it as an array of one?
8:56:32 AM Aaron Gotwalt: oh, i love the collection bit
8:56:34 AM Aaron Gotwalt: basically
8:56:40 AM Aaron Gotwalt: i'm onboard until the data element
8:56:45 AM Larry Marburger: that's what i mean by ease of parsing. you always know what to expect.
8:56:52 AM Aaron Gotwalt: at which point it goes too far
8:57:12 AM Larry Marburger: whereas if you're hand-rolling, you either have an unspoken agreement between client and server or you just don't think about it (that's how i operated)
8:57:31 AM Larry Marburger: the format of the data element matches the format of everything else
8:58:22 AM Larry Marburger: links, data, items, etc. are all arrays of objects.
8:59:53 AM Larry Marburger: i think one of the main reasons for using arrays of objects and not a single object is because it's easier to convert text to a js string than to a js object key
9:00:40 AM Larry Marburger: i remember having that conversation but i don't exactly remember the reasoning