Hello Ketil, Wednesday, September 13, 2006, 10:41:13 AM, you wrote:
But a String is something like 8 or 12 bytes per character, a ByteString gets you down to 1.
12-16. Char itself, pointer to the next list element, and two boxes around them - this count for 16 bytes on 32-bit CPU. but cells with small Char are preallocated at program startup, so it will be 12 bytes for ascii-only strings but that is not the whole story :) copying GC makes program's memory usage 3 times larger, on average, than it really allocates while compacting GC has only 2x overhead ByteStrings are allocated in _unmovable_ part of GHC heap, so they don't suffer from this problem. of course, it is not free - program that creates and destroys large number of ByteStrings will suffer from memory holes, which is right the problem solved by ghc's GC so, for program that only allocates the difference may be 24/36 times on average while for create-use-destroy-loop scenarios i can't make any detailed prognoses also, traversing those lengthy lists on GCs is very time-consuming -- Best regards, Bulat mailto:Bulat.Ziganshin@gmail.com