Joshua Gatcomb <[EMAIL PROTECTED]> wrote: > All: > oo1-prop.pasm and oo2-prop.pasm are both broken.
Yep. First WRT calling conventions and - that's a bigger problem - again WRT DOD/GC. The latter is permanently reocurring and needs fixing. 1) The main source of troubles in hash code are the moving chunk buffers. 2) As every collection of buffer memory can move these buffers, buckets can only be addressed by index, which causes one more lookup per hash operation 3) This make hash code slower then it should be I'd suggest to rewrite hash.c to use system memory. Takers wanted. Back to oo*-prop.pasm. I think they are useless as benchmarks, but the loop of setting properties repeatedly would make a good gc.t test. leo
