American Fuzzy Lop VS fishnet

There's some efforts to guide test generation for property based testing to make the instruction pointer explore as large a space as possible.

This effort is more mature in the fuzzing community. See eg American Fuzzy Lop https://github.com/google/AFL

What you're describing, just generating random input to test a program, is sometimes called "blind fuzzing" but the state-of-the-art is far beyond that. Maybe try reading through the documentation of e.g. https://github.com/google/AFL to see what a fuzzer does and why just producing random input isn't even scratching the surface.

for general riscv I used to use this https://github.com/google/AFL I dont know if it supports x64 tho.

Ex ( AFL, WinAFL, HonggFuzz, LibFuzzer, Jazzer )

I wrote this little fuzz test target in order to fuzz it with afl (under ASan and UBSan):

afl, which is trivial to apply to this program:

I made my own version of a TCL interpreter (well, a very TCL like langauge) derived from "picol" available at https://github.com/howerj/pickle. There are many different re-implementations and derivatives of this interpreter but they all seem very "crashy", this one has been significantly hardened by using a fuzzer on it which ran for months called American Fuzzy Lop https://lcamtuf.coredump.cx/afl/ . It is also more suitable for embedded use whilst still not having arbitrary restrictions like many other implementations.

On Linux afl is a very powerful bug-finding tool, and it's a great companion when doing code review. Composes well with ASan and UBSan.

b-, https://lcamtuf.coredump.cx/afl/

> The problem is that a stockfish based bot knows some very strong moves, but deliberately plays bad moves so it’s about the right skill level.

What are you basing this on? To me it seems like difficulty is set by limiting search depth/time: https://github.com/lichess-org/fishnet/blob/master/src/api.r...

The LiChess documentation indicates how many nodes are searched: https://github.com/lichess-org/fishnet/blob/master/doc/protocol.md

The eval graph comes from an analysis done by fishnet. The analysis that you see changing is done locally in the browser on your device. That is why there is a difference.

YOU can help by running this: https://github.com/lichess-org/fishnet

Game analyses are made in a distributed way, but only accepted from users they trust. You can check the tool they use for that here and they instructions to use it.

It's not on the user's device and not on their servers. The game analysis is done by fishnet using donated CPU time.

You can run a program on your computer, so that Lichess can run Stockfish analysis using your CPU https://github.com/lichess-org/fishnet