@Nick Smith Looking forward to it. I had no idea you were one of us with experience teaching kids. Here’s what I did for two years, and it worked great in a 1-on-1 setting: http://akkartik.name/post/mu (Spoiler: it’s absolutely text-based). I saw no problems with (a small number of) keywords. I did see value in minimizing punctuation. One implication: don’t start kids on one of the Lisps. (Even though I love Lisp in general.) I don’t understand your “It’s 90% boxes-and-arrows, 8% Scratch clones, and 2% of unique stuff” comment. Why chase novelty for its own sake? Similarly, I don’t think boxes and arrows or blocks have been given nearly enough rope in the last 5 decades. You’ll need to elaborate on why you think they’re “objectively bad”. It’s not at all obvious that we need to try something different. One could equally make the case that we need to persist with and incrementally improve already-attempted ideas. (I say this as someone pro-text for the most part.)

Nobody here ever said novelty was worth chasing for its own sake. However, I disagree that boxes and arrows haven't been given enough rope. They've been given an excessive amount of rope: there are hundreds of boxes-and-arrows editors out there. They seem to have proven useful in niche scenarios, typically where only small structures are required. As discussed though, when using arrows for dataflow you inevitably end up with an arrow soup as you get larger; the visualisation just doesn't scale. I worked briefly with the world leaders in graph layout research and even they are still struggling with ways to make large graphs readable. But either way, it's not up to a disbeliever to prove the absence of something; it's up to a believer to prove its existence. If you'd like to argue for the merits of boxes-and-arrows, I'm open to discussing them with you.

I agree: block programming à la Scratch, show the same program structure as text. Take a Scratch script, erase all colors to keep only text, you end in something very close to an imperative language. In fact there is one tool that is showing that. I thought it was GP (https://gpblocks.org/) but I can't find pointer ti this feature. Anyway, this (unknown) tool has a slider that progressivly erases blocks to keep only text. Then one can edit text as usual text language.

I agree that Scratch is great for good old fashioned structured imperative programming – with most of of the disadvantages though its broadcasting coroutine/concurrency bits are pretty inspired.

What is abstraction story of box & arrow world? It comes to my mind that you will get huge graph it's equal to huge function.

@Niko Autio Many node-based (aka box-and-arrow) VPLs use some sort of grouping / nesting feature to turn chunks of an existing graph into a new kind of node. Some of them only let you collapse things in place, but the nicer ones let you create a reusable node that you can treat like a function.

Hmm, not sure why I got no notifications about this thread.. @Nick Smith you’re right that there’s a scaling problem with graph visualization. Traditional approaches like Labview certainly feel like a dead end. But there’s newer ideas that are worth exploring, like Pane by @Joshua Horowitz, which has been discussed here before: http://joshuahhh.com/projects/pane. I also brought up Boxer.