Here is the simplest|minimal|readable python3 resolver (naive backtracing, recursive):
sqr = lambda g,x,y: g[y*9+x:y*9+x+3] + g[y*9+x+9:y*9+x+12] + g[y*9+x+18:y*9+x+21]
col = lambda g,x: g[x::9]
row = lambda g,y: g[y*9:y*9+9]
free = lambda g,x,y: set("123456789") - set(col(g,x) + row(g,y) + sqr(g,(x//3)*3,(y//3)*3))
def resolv(g):
i=g.find(".")
if i>=0:
for elem in free(g,i%9,i//9):
ng=resolv( g[:i] + elem + g[i+1:] )
if ng: return ng
else:
return gSome grids are available in g_simples.txt (a grid by line of 81 chars, empty cases are .)
The idea of the repo, is to compare differents languages at "run times". Currently, there a c/mojo/nim/java/js/rust versions. So every version implements the same algorithm, without using specials optimisations provided by the language itself ... and try to resolve the first 100 grids !!!
On my computer (Intel® N100 × 4 / ubuntu 23.10), I got :
node18.13: 0m 44s
py3.7: 0m 40s
py3.11: 0m 34s
pypy3.10: 0m 15s
codon0.16: 0m 23s
java22: 0m 21s (openjdk)
Nim1.6.14: 0m 11s (release mode)
Nim1.6.14: 0m 8s (danger mode)
mojo0.4.0: 0m 7s
c/gcc13.2: 0m 2.7s
rust1.71: 0m 32s (not specialized version)
SPECIALIZED versions (with specialized types/structures by languages)
=====================
mojo0.4.0: 0m 3.1s (sudoku_specialized.mojo : with specialized types)
rust1.71: 0m 1.2s (sudoku_specialized.rs : with ultra-specialized rust types/apis)
BTW, Other experiments results :
mojo0.4.0: 0m 2s (experiments/mojodojodev.mojo) (different algo)
BTW2, tests with an optimized algo are availables too.