|Subject:||Number of items in @INC scales with -j and process-number|
The attached script generates a directory structure that replicates this problem, and then the output of concatting the various per-pid output files shows how some pid's get much larger @INC's, despite the code itself not changing.
But the general problem is essentially
prove -j4 a b c d
Results in a getting 1x lib, b getting 2 x lib, c getting 3x lib, d getting 4xlib, etc ( but not necessarily in that order or distribution, its kinda random )
The only good part is it seems to cap out at -jN, and for some reason it gives the impression once N is reached it goes back to alternating between one and 2x lib.
But its not great.