Sudoku-solving based language benchmark

This is a benchmark of a Sudoku solver written in various languages. The goal is two-fold:

• to compare the performance of different languages
• it's my Hello, world! when I learn a new language

Results

The results are for a 9x9 Sudoku, with 3x3 blocks. The benchmark is run on a 2022 MacBook Pro, with a M1 processor and 16GB of RAM.

For an easy Sudoku

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
make bench-go	7.3 ± 0.6	6.2	9.5	1.18 ± 0.17
make bench-rust	6.2 ± 0.7	5.2	9.2	1.00
make bench-python	421.1 ± 15.4	406.0	453.1	67.95 ± 8.21

For a hard Sudoku

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
make bench-go	32.8 ± 0.7	31.5	34.5	1.69 ± 0.08
make bench-rust	19.4 ± 0.9	18.1	22.7	1.00
make bench-python	7127.5 ± 51.9	7039.1	7212.9	367.95 ± 17.07

For a very hard Sudoku, made especially against backtracking

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
make bench-go	507.8 ± 3.3	503.7	513.2	1.91 ± 0.02
make bench-rust	266.4 ± 2.1	263.8	270.3	1.00
make bench-python	439142.3 ± 4526.4	435062.2	448740.0	1648.47 ± 21.52

Yes, the Python version takes 7 minutes to solve this Sudoku, while the other languages answer in a second.

Reproduce

Every benchmark must have a reproduce section, along with the usual License and Contributing sections. Not doing so is wrong. Hi, TurboPack

Prerequisites

• Hyperfine
• Rust
• Go
• Python
  • pip install numpy rich matplotlib argcomplete

Run

make bench-all

FAQ

As these are my Hello Worlds, the code is probably not idiomatic. I'm open to suggestions and PRs.

"The xxx implementation is not optimized for performance!"

This is your language's fault, not mine: I would be happy to code a faster algorithm, but if the language doesn't make it easy, I'm not going to do it.