[tagrun]

Can't say it's surprising. The von-Neumann architecture computers and their assembly languages are imperative. Functional programming language structs are just unnatural. Being computationally universal, one can simulate one using the other. But to me, it feels like an unwanted abstraction layer.

Analogy #1: if I were going to write a game for PC, I would directly write a game for PC, not a GBA ROM + GBA emulator for PC and make-believe that it's a PC game.

Analogy #2: if I want to write a novel in Spanish, it will not be possible to achieve the quality of a text written in Spanish from the beginning, by say, writing it in Japanese and using a translator (no matter how much you may like Japanese). Some idioms and culture-dependent things will be lost in translation. (Italian or Portuguese might be better, however)

[taeric]

The von-Neumann level nature of the machine really doesn't enter into it, though. I mean, ultimately, yes, somehow your solution has to be run in an imperative way. At the high level "looking at the declarative solution," though, it is irrelevant.

So, to your analogy #1. If such an emulator already existed and was in wide use, you would not be doing any harm to your system to target it.

To extend your analogy to the absurd, it doesn't make sense to write your program out symbolically in a programming language, because at the end of the day it is electrical values in a processor.

[tagrun]

> The von-Neumann level nature of the machine really doesn't enter into it, though.

It does enter when you try to execute the code; which is a translation from declarative to imperative (assembly).

> To extend your analogy to the absurd, it doesn't make sense to write your program out symbolically in a programming language, because at the end of the day it is electrical values in a processor.

Care to tell me one single example of a declarative computer?

I'm more of a hardware engineer than a programmer, and I can't think of any universal logical circuit with memory that is not imperative. Even at the transistor layer, everything is imperative. This includes gate-model quantum computers and biological computers. There are some analog models that work for specific problems, but there certainly are nowhere near being declarative.

And the point was, at the high level, declarative -> imperative translation (which is done by a compiler) will typically be worse that an imperative -> imperative translation.

This is why a very written compute kernels in C outperforms its, say, Haskell equivalent, typically by orders. It's also the reason why scientific computation, signal processing, transcoding software etc. (when raw CPU performance outweighs everything) is typically written in Fortran/C/C++ with a mixture of assembly.

The reason is simple; they map to the underlying hardware well. It is also the reason why assembly languages are imperative.

[taeric]

Apologies, I'm not sure what it is you are trying to say. :(

The fact that the underlying system is von-Neumann seems just completely non sequitor. That is my only point. Every point you make afterwards is just as true of the "Harvard architecture" as it is of the von-Neumann one.

In the rest of what you are saying, it honestly sounds like we agree.