[vintagedave]

This point was where this changed from crazy/fun to absolutely extraordinary, where calculations of multiple possible positions all occurred in parallel, running a regex over an increasing series of state & variable sets, aka threads:

> And now for my absolute favorite part of the language we've developed. By the magic of regular expressions (and the fact that they perform substitution globally over the entire string), we can run multiple threads simultaneously!

Also:

> What do you want out of a conclusion to a blog post like this? I don't really have much to conclude. I guess I'll just say that I think more people should do entirely pointless things like this. It's a lot of fun, no one cares how long it takes you to finish, no one cares if it works or not, and incidentally, it teaches you more than you wanted to know about dozens of areas of computer science outside your field.

What a wonderful ethos.

[ricardo81]

This makes me wonder whether I could achieve such a thing if I removed all my preoccupations of other stuff.

For me what I take out of it is the power to sit down, focus your mind on something then who knows the lengths of what is possible. That, and the author is clearly very talented/skilled and creative.

[larodi]

Apparently it takes more than skill but also persistence and concentration. Not many schools of thought explain this well .

[PaulHoule]

No, it takes (1) knowing what you learn in a compilers class (or upper level math classes) and (2) putting it to work. He didn't write 80,000 regular expressions, he wrote a compiler that wrote those expressions. Commercial-quality compilers are hard to write but simple compilers are straightforward if you know the fundamentals.

It's like the problem of solving the Rubik's cube. If you look at it in terms of geometry and spatial intuition it's intractable. If you treat it as an abstract algebra problem and break it down into "solve the top of the cube, solve the middle row of the cube, solve the bottom of the cube" and then develop a set of operators that will let you permute certain parts of the cube it takes a moderate amount of skill, persistence and concentration.

CS students do exercises such as writing moderately complex programs for a Turing machine and it's an exercise like what he did.

====

Funny my project last month was a chess engine. For a long time I'd wondered if I could make a decent MCTS chess engine but someone close to me has been getting serious about chess (usually wins against the random person, usually loses at the chess club, is fighting hard to get up the bracket) so writing a program that was a match for him seemed like a fun and meaningful project and I decided to try the conventional route of alpha-beta search.

If you tried to write a chess engine from zero you'd probably struggle, but the lifetime value you get out of education (CS or otherwise) is looking things up the literature, learning from other people's experiences, and putting it to work. So this has been my guide

https://www.chessprogramming.org/Main_Page

I started out with Python with the goal of making something tiny and stylish and started out with a move generator from

https://python-chess.readthedocs.io/en/latest/

and once I got the signs right in alpha-beta negamax (when I had them wrong it discovered https://en.wikipedia.org/wiki/Fool%27s_mate) it beat my tester several times before he regrouped, traded a knight for a good pawn structure, and got his first win.

The next goal is to take it to the chess club which means it has to respect time control. I switched to Java because its faster and because it is easy to have a comms thread interrupt a think thread. The first try wasn't faster because my move ordering was terrible. I was much faster implementing it though because I could cut and paste the test cases I had in Python and if I knew anything at this point it was the signs in negamax.

https://en.wikipedia.org/wiki/Iterative_deepening_depth-firs... is not only good for time control but gets better move ordering and I'm in the middle of implementing that. It would take me years to find out that Iterative Deepening works a lot better than you might expect, the Killer Heuristic, etc. Reading is my superpower.

[larodi]

Okay, as you wish.

Though I've never seen people accomplish stuff like a) understand regex; and b) write a compiler for it without "persistence and concentration.". Perhaps some get it delivered through divine apparition of some kind.

What is it that really triggered your lengthy response and the narrative..., I really fail to understand, sorry.

[3willows]

Agree. The whole point of the article is to generate the long list of regex programmatically.

A quick way to verify this is to download the repo, remove everything other than main.py and regex-chess.json, and the programme will still work.

All the other python files are building up to regex-chess.json, see e.g. the imports and output to write_regex_json.py.