[tromp]

With my 1988 IOCCC entry

    char*M,A,Z,E=40,J[40],T[40];main(C){for(*J=A=scanf(M="%d",&C);
    --            E;             J[              E]             =T
    [E   ]=  E)   printf("._");  for(;(A-=Z=!Z)  ||  (printf("\n|"
    )    ,   A    =              39              ,C             --
    )    ;   Z    ||    printf   (M   ))M[Z]=Z[A-(E   =A[J-Z])&&!C
    &    A   ==             T[                                  A]
    |6<<27<rand()||!C&!Z?J[T[E]=T[A]]=E,J[T[A]=A-Z]=A,"_.":" |"];}

discussed in more detail in Don Libes' "Obfuscated C and Other Mysteries" [1], I turned out to have rediscovered Eller's algorithm, which only keeps one line of the maze in memory.

[1] https://tromp.github.io/maze.html

[remram]

This segfaults for me somehow.

[tromp]

See https://tromp.github.io/pearls.html for suggested changes to make thus run with modern compilers.