Well, guess what
The community stdlib has one and it's good, so they spared me the effort ​​

I was about to add unit tests for the IRI parser and found a nice JSON file with lots of test cases. I have JSON parsing, so that's good. What I don't have is a nice way to decode the generic json type to a specific type presentation, like with Aeson's FromJSON. I could still get all the test cases from the JSON file without it, but the rest of the application needs to handle JSON as well and then a decoder abstraction would be really good to have. So writing a decoder module is next ​:brdFlat:​

edit: Never mind again. I keep missing stuff because I keep looking for Haskell-like solutions, but Koka is simply different.

The application crashes when the JSON is too large. I bet it's because Koka's parser combinator library makes every single step backtrackable.

It's actually not too difficult to avoid backtracking if you make the subparsers smarter. I just didn't do it because it also makes them less readable, but oh well.
So right now I parse booleans like this, for instance: string("true") || string("false"), which means it tries the first parser, and on failure the second one (and therefore the consumed input must get "un-consumed").
The smarter solution is to pick the right branch directly so there is no "un-consuming" necessary. Like this:

val c = one-of("tf")
if c == 't' then
  string("rue")
else
  string("alse")

I removed all backtracking and it can handle larger JSON files now, but I still need to remove a very large chunk of the tests file to avoid the crash. ​​

Looks like it's a known issue with ctl effects (which parsers use): github.com/koka-lang/koka/issues/279
There's one more idea I want to try, but otherwise I'll just accept it for now

@volpeon@icy.wyvern.rip How large exactly? ​​

IRI parser works ​​
Only the IRI resolution algorithm left, and then I can finally get to JSON-LD

@volpeon@icy.wyvern.rip oh no, gonna loose volpeon to JSON-LD again

will it be like the last time you tried that?(i.e. we don't hear from you at all for a week or so) ​​

I revisited the JSON parser and wrote my own parsing library. It's not too different from the one in the stdlib, but it handles some cases more efficiently and that makes a huge difference.
The old parser caused a stack overflow with a 144kb JSON file. The new parser can handle a 1mb JSON test file, but still causes a stack overflow with a 5mb file.