#cweiskevoxpelli, I've had enough API situations in which validating the returned content type quickly told me that either I made an error, or the API sent something strange to my request. reduced debugging time
#voxpellicweiske: my experience instead says that trying to return the same format no matter which situation and then inspecting the data itself and/or response code is preferable
#voxpelliEspecially considering the crazily complicated specs of accept headers etc
#cweiskeif everyone thought that way, then the content-type header would not exist
#cweiskesince you can always inspect the content. hah.
#voxpelliContent-types to me are to control whether response should be eg. HTML, xml or json – not any more than that
#voxpelliYou can only inspect the content if you get an instruction on how you can expect to be able to inspect it
#voxpelliBut true, many clients ignore content types and simply parse the response in different ways until a parser returns a non-error result of the parsing
#cweiskemagic bytes and heuristics always allow you to inspect anything
#voxpelliCan you show some pre-existing work and some arguments on why error-specific content types would add anything and make things better in any way?
#cweiske"reduces debugging time" is my argument, and I only used it in commercial closed-source projects.
#voxpelliThe downside of them are that any clients that expect a json response to have the content type "application/json" will fail to parse the response
#voxpellicweiske: so far I only see an argument that it reduces _your_ debugging time to be fair. It would probably increase mine because I would totally not expect it
#voxpelli(Also, your suggestion is by the way wrong as I can see, the proper mime-type would be application/json+modifier)
#voxpelliRight, um, well – that in itself is a pretty good argument why mime-types are crazily confusing and should just be left as simple as possible ;)
#voxpelliIf one just can deliver such a message then all suck kinds of cases becomes a non-issue from a spec point of view
#voxpelliI guess I could show such a message if I fail to communicate to GitHub or fail to translate it to a Jekyll document
#aaronpktypically the error messages that APIs return are not user-centric, and more often than not just confuse the user rather than help them. it's an interesting idea to have a special property to explicitly return a user-level error message, especially since with Micropub, we have lots of different server implementations that have various restrictions outside of what the spec requires
#beari've always been a fan of { "result": { "code": 5XX, "message": "you did something wrong", "text": "here is a text version of why I decided to not let you do this..." } }
#aaronpki've been following the oauth2 error response format of {"error":"something_wrong", "error_description":"Message to the developer"}
#bear*nods* that works also if the transport is HTTP
#voxpelliaaronpk: I'm just about to head to sleep, but general thought is that looks good, only thing maybe lacking is how to add a non-experimental property – if eg another spec wants to extend Micropub