To be honest, I don't see a problem. Yet that won't hinder me to dream. There is the NIR survey. I don't understand half the options, but they make me think: If it can offer a REPL and that REPL runs everything written in plain NIM as if it where a scripting language, what could be done with that? Use nim as a scripting language for prototyping. No need to compile (yet).
Then is this IC thing. Could it compile a part of the script(s) written above? Could it compile just the Nim std libs, or any pure Nim lib, I need to run that script? Could, with the help of NIR, auto-binding from Nim script to compiled Nim be done? As it is possible to write DSL's in macro's it would mean one can write a DSL based program that is compiled but understands the Nim script DSL without the need of writing a separate parser. It would also mean that a user of that program does not need to compile it, but could. In a thread on sliced seq's speed ElegantBeef wrote `at zero cost`. Searching the forum for `zero cost` it turns op in various topics. To what extend can `at zero cost` be applied to the sd lib? There are several efforts for multi threaded applications. Are they "simple" enough? Are they usable `at zero cost` for the user? The last meaning, can it be (made) available at the flick of a switch? `for ...` `parallelfor ....` without a thought. In similar vain `simdfor ....`. To what extend can the std lib be "simd-ified" to be extremely usable without cryptic names etc? Dream on old man. It adds quite some obfuscation/sugar/magic to the language. Underlying algorithms etc. should be visible. It's not killer libraries, it's killer features to write rather spiffy libraries easily.
