Hello Donald, Thursday, April 20, 2006, 1:06:38 PM, you wrote:
this lib should be slower than your on small strings due to ForeignPtr inefficiency
Do you have some sketches of ByteArray# represenations? I'm very happy to change the representation if it makes things more time or space efficient.
you should see at the jhc's PackedString implementation that uses UArray (which is internally ByteArray#). his implementation also don't use additional Ints - array size is stored as part of UArray structure and O(1) substr operation is just not exists (instead, it is O(n) with copying) as i said, this can be extended to support List-like interface for ANY arrays, that is the very useful thing. if you will implement this, then we will only need utf-8 wrapper around UArray Int Word8 for implementing packed Unicode strings (and Latin1/UCS-2/UCS-4 strings will work just as unboxed arrays of Word8/Word16/Word32) you can compare speed and memory requirements of ForeignPtr and ByteArray# solutions by comparing jhc's module with Simon's utf-8 implementation - both are not such greatly optimized as your lib, both uses utf-8. please use large number of typical strings (with size of 10-100) for real-world comparison, comparing on 20mb strings is not indicative i can write disk traversing algorithm (like ls-lR) for some sort of real app, to make testing more realistic -- Best regards, Bulat mailto:Bulat.Ziganshin@gmail.com