Plus you can even write your own reader literals to enhance your data format. Aero does this well/horribly depending on your taste but it's a cool mechanism for declarative formats.

Yep, two types in fact, line comments with ; and a discard sequence - you can tag code that is read (so it must be correct), but then discarded. We have edn files that are well documented with comments.

https://github.com/edn-format/edn#comments

Yeah but still you have to admit, there's quite a lot more friction and ambiguity in that approach. Do I represent a map as

    [
      ["key", "value"],
      ["otherKey", "otherValue"]
    ]  
  

or

    [
      {"key": "key", "value": "value"},
      {"key": "otherKey", "value":"otherValue"}
    ]  

or something else? And how can I distinguish between values which just happen to look like maps but are actually not maps? And what if the service I'm talking to does it differently? What if I'm comparing or sorting two JSON values, don't I have to modify the equality/hashcode/ordering logic now to interpret

    [
      ["key", "value"],
      ["otherKey", "otherValue"]
    ] 

and

    [
      ["otherKey", "otherValue"],
      ["key", "value"]
    ] 

as being equal and implement a new hashcode/ordering which treats them as equal?

Quite a lot easier when this is just native to the format.

JS Map/Set use referential equality, which is pretty worthless if you aren't dealing with primitives.

You may have missed the other 10 reasons I listed besides performance.

Also, no JS code will be faster than Java. While both have insanely good JIT compilers that can output some truly, often surprisingly great machine code, java is all-around faster, if for no other reason, due to its killer GCs.

>That's like saying that C++ is a better backend platform

C++ isn't even a platform, period. It provides no runtime or anything else other than the language. If you build on top of the JVM ecosystem, similar to NET you get a cross platform, extremely performant ecosystem with millions if not billions invested into it basically for free.

And of course you can build a billion dollar company on a crappy tech stack, but if you had built it on a good one you'd still be a billion dollar company and be even better off. What a strange argument, you ought to make the best technical choice you can, it's a core part of anything you built. And getting a 10x performance on the JVM vs say Python, or smooth, non error prone concurrency in Clojure is a significant benefit.

In particular in business applications were you usually deal with data transformations, the single-threaded, mutable state type of design of some languages is awful. Clojure in particular was exactly made for this practical use case.