I am zero filling the free heap after garbage collector's heap compaction phase. Start address depends on the size of the referenced data and end address depends on the number of object handles. Typically I am zero filling about 20k. Luckily garbage collection will anyway require interrupts disabled, so that I can mess with sp. With 8 word block size the inner loop now takes 11 cycles per 8 words.
I must have fixed most of the Admiral's bugs for now, as I don't find new bugs so often anymore. I still have couple of unsolved issues:
float to str routine still can't handle e format -> I have tried to fix this, but I have failed - some day I will try again
repr() gets into infinite recursion if objects reference each other -> I must decide what I want to do with this
serialization is bound to floppy only -> I will refactor whole serialization stuff once we get network specs from Notch
I will propably remove block scope (if, for, while), I haven't seen much benefit from it, but it has small performance impact
I will propably end Admiral's feature freeze in April and continue development with some new features. I haven't yet selected the next topics, but quite high in my list are things like inline assembler, SPED-3, integration with one of the DCPU operating systems, exception handling, screen library...
Posted Mar 30, 13
· Last edited Mar 30, 13by orlof
This site's owned by Enjin's owners - I assume that they got themselves an indefinite subscription and just forgot it after all. Pretty sure that Enjin is large enough to keep one site more or less up.