The current implementation of array.pmc is rather inefficient and limited - and really slow.
So I rewrote the base routines almost from scratch and have currently a file named list.c, which I will commit (yep cvs ci, thanks to Dan, Steve and others, who made this possible) in the next future - if people agree. Some features: - list.c is based roughly on principles of intlist.c (chunk based, fast index set/get, push/pop/shift/unshift) - can handle <null> arrays, no chunk allocation first - can store packed char, short, int, num, PMC* (void*), ... - different growing policies: - growing (min alloc 4 items/chunk -> max 1024 items/chunk (both are defines but above are good limits in tests) - or fixed 1024 items/chunk for arrays first accessed like "set P0[500], I0" - same speed as intlist, ~double speed for big # (10^6) items - can handle sparse arrays, saving many MBs for very sparse arrays Currently I'm testing it as an intlist substitute (via wrapper functions built in), it runs all intlist tests + 2 more WRT sparse access. So, if people are ok with this, i'll commit it as list.c in parrot's root (+ some more t/pmc/intlist.t). Transition of array.pmc may then follow later. It's currently not intended to replace intlist.c, I don't know the planned usage patterns for intlist (rx stack?). Some optimizations in list.c might also improve intlist.c so that this would be a special and therefore faster array with limited capability. If conserving memory is an issue, list.c is much more efficient for very small and very big arrays. We will see. I know, that Peter is working on a similar array substitute, so I announce the status of my list.c - and when he has a better array, we can always change to it. Or we take the best things out of the two and improve it even more. Comments welcome, leo
