Closed world systems let you control (and prove) a lot, but you can only do the things your own "world" has built an idea up about. Also "easier" to build DB schemas for, etc. But hopeless for extensibility when talking with the rest of the world.
Meanwhile, completely free-for-all-json, coordinating on meaning is nigh impossible.
@gcupc *your* closed world may not have all the things in *my* closed world, and yet there are a lot of things we can communicate about.
Say you are a game developer and I am a vegan chef. You say "My boss was really bugging me that I need to improve my A* pathfinding algorithm." Well I don't know what a pathfinding algorithm is, but I can empathize about your boss situation. If I told you I made a good BBQ tempeh and you don't know what tempeh is, you at least might know BBQ.
I haven't officially changed my name yet, but for the sake of paperwork minimization and local friend confusion maximization, I will be switching my middle name to my spouse's last name rather than hyphenating.
@cwebber Oh, sorry.
Yeah that sounds like a very pleasant bureaucracy trip
@gcupc I can at least operate on the things I know about, in my mind, and you can in yours.
The context bit comes in because our conversations are usually in contexts. Programs need that too, so we can know that "run" a program is not "run" a mile.
Even if the world is wide and open, within our closed minds, we can do a lot to understand and interact with each other. Indeed, we must, or we will only talk to ourselves.
I think that's, at best, a formula for making everyone a little bit unhappy.
And it's also a bad analogy. Humans are very good with ambiguity and working around missing knowledge. Computers are very bad at it.
@gcupc That's why the context maps to exact term URIs (Universal Resource Indicators). You and I at least can try to achieve rough consensus on what those specific terms mean.
"You and I" can, but our computer programs can't. All a JSON-LD context tells my program is that it's definitely entering territory it knows nothing about.
@gcupc And that's fine. Your program doesn't know about it, so now your program can ignore that thing for its own logic, but shove it somewhere so that when sharing it or etc, it can present it as part of the whole artifact in case another entity does.
Maybe you don't know what Tempeh is, but my friend Alice does, and now she can understand more when you relay it to her.
Right, so in practice, what this means is that basing ActivityPub on JSON-LD ensures that no two federated social networks will be fully compatible. And I know you consider this a good idea, but I can't understand why.
(That's leaving aside all of the critical interoperability parts of the spec that are left unspecified.)
@gcupc Do you consider it a good idea that at standards-writing time we assume that we can write all possible behavior for all time? I'm not so smug as to believe I can anticipate all needs of the fediverse.
We didn't define a document type for VirtualRealityRoom, for instance, but maybe in the future that's the most popular object type.
Should we say, "AP/AS2 didn't define it, thus it can't exist"?
@gcupc Closed world networked systems always get expanded into open world systems whether you like it or not. Do you plan for it or don't you?
Real standards get revised from time to time. And an implementation can specify which version of the standard it implements.
@gcupc Sorry but upgrading between standards revisions that don't build in support for extensibility barely happens. At my time in standards the most frequent regret I hear people say is "Well we thought people would be able to upgrade to API X.2, but it never happened... it turns out we were stuck on API X.1 forever."
@gcupc There's a network effect in non-extensible systems where if it's "all or nothing", movement is an insanely hard thing to do.
Look at IPv4 -> IPv6 as *still* an ongoing effort.
I am amazed to hear that no one uses HTTP 1.1 or HTTP/2. Or RFC 5322 for that matter.
@gcupc smtp is a great example of a) a standard that hasn't been updated but it would be nice if it could be and b) people awkwardly hacking in extensibility... x-foo-header
(I'm not totally clear on how inference works here, but in more conventional inference systems you have the concept of inference libraries with publicly exported predicate names whose actual implementation might change.)
Being able to merge inaccessible and contradictory logical rules is straightforward when names are canonical, so long as your operations also assume open world.
(In other words, while we can expand possibilities, we can't operate based on the assumption that we have expanded all possibilities. Prolog's '/+' operator can't be treated as a 'not' operator, in other words, because 'impossible to prove' is concretely different from 'untrue' rather than more abstractly.)