Show HN: A virtual machine made with Google Sheets formulas (no script)

The basic algorithm is "parentheses counting": walk the string starting from an opening bracket, counting +1 per `[` and -1 per `]`; when you reach 0 you're at the matching closing bracket.

First, lets split the code to one char per row:

    =MID($A$1,D3,1)

(D3,D4,... are simply 1,2,3,...)

Now, lets parse `[` as 1, `]` as -1, anything else as 0:

    =IF(E3="[",1,if(E3="]",-1,0))

Now, we will need one column per character in the code, where in the i'th column the first row is the i'th character's value and each row we go down accumulates the value of the next character.

    =G2+OFFSET($F2,G$1,0)
    =G3+OFFSET($F3,G$1,0)
    ...

(Column F is the 1,0,-1s. Row 1 is 0,1,2,3..., this time horizontally, which we use to translate horizontal motion into vertical motion)

Now for the i'th row of the "] match" column, lets find the index of the first 0 in the i'th column, and add `i-1` to get an index from the beginning of the code instead of from the opening bracket:

    =MATCH(0,offset($G$3,0,D3-1,18,1),0)+D3-1
    =MATCH(0,offset($G$3,0,D4-1,18,1),0)+D4-1
    ...

And all that's left is reverse-search this table to find the matching opening brackets to every closing bracket, and while we're at it lets clean up all irrelevant rows to the empty string:

    =IF(F3=1,C3,IF(F3=-1,MATCH(D3,$B$3:$B$20,0),""))

(D3 is the index of the closing bracket, $B$3:$B$20 is interestingly the column we're currently in - Sheets does not complain as long as there is no actual circular referencing going on)

Voila! Column B now contains for every bracket its matching bracket's index in the code, and for anything else, "".

Ah. I thought "." must be printing (as it's something similar to that in Forth [guess it's ".s" though not "."]), but I was confused by the comma because I didn't understand how you'd be reading from stdin here. Actually I'm still confused. What are you reading in?

We were bootstrapped and only sold in Israel, so there is nothing about us in English.

It started with a programming course I founded for data analysts in my army unit that taught them enough Python and sql and requests and win32com to be able to automate significant parts of their own jobs, creating a lot of value, and I wanted to figure out how to teach the same course outside.

The answer was "nobody outside the army knows how to deal with losing an employee for 3 weeks", so we cut the material again and again and even at the minimum we could reach - 80 hours - we couldn't make a single sell.

So my cofounder quit and I started playing with different models of teaching, and found out that there are huge benefits to teaching with a computer game (more fun, more engaging, more self-driven, but mostly - you get so much data and can optimize your lessons so fast... I really don't understand why e.g. Codecademy don't do a lot more data-based micro-optimization).

I partnered with some people that sold after-class activities to middle schools and we sold a lot more courses than we planned, so I had a frantic year of supervising the teaching of 14 courses while trying to amend my weekend prototype to teach the whole material. On the way I developed a bunch of other nice teaching activities.

But working alone sucked, so I put it on hold and went looking for the best team I could find for my "real" startup (this one was started with the explicit intention of failing but learning as much as I can, as I was pretty clueless in business at the time).

I strongly believe in educational technology as an efficient way to improve the world (though a bit less than I use to as a way to get rich). I fully intend to return to working in the field when I'm no longer needed at FeezBack (sorry, again we're Israel-only at the moment, so no links).

Sure! I emailed you. Or we can discuss it publicly here :-)

First of all, I must agree about the way I presented it, and admit that when j45 posted that first comment I checked if I could still edit my post and it turned out I couldn't.

I've used Brainfuck in some educational contexts (like the talk on virtual machines that I gave in PyCon IL a few weeks ago and in a local python meetup before that), although not yet with complete beginners.

It's useful in some contexts and not others. It's especially useful when some of your audience has heard of it before.

It's especially useful because it lets me mention other "war stories", like this one (students love war stories): https://news.ycombinator.com/item?id=7943514

Before I teach asm, I always take 1.5 hours to teach an exercise I once developed called the "shirts exercise", which is about optimizing code for a simple and very intuitive vm that looks a lot less like real asm than Brainfuck does (for example, it has conservation of material). Real asm has so much incidental complexity and I prefer to show the beauty and motivation for learning it before we start learning about number encoding etc'.

The reason to use this spreadsheet is that it's written in a format that my students would be able to follow and understand. So if I tell them "this is exactly how a computer is built", I hope to demistify a bit the idea of a computer. A good programmer is a programmer that is not afraid to dig into abstractions.

The creator of Brainfuck, Urban Müller, made a once in a lifetime presentation about it recently. It was at a company event where we had the pleasure of watching him talk about how he developed it. If you are interested you can check it out at the livestream recording that was posted in youtube (position 2:52) https://youtu.be/gjm9irBs96U?t=8722

Lol, truestory