On 9/24/12 8:53 AM, George Giorgidze wrote:
We will release GHC patches for both approaches, meanwhile the feedback from the community on the approaches that we took would be very much appreciated. Which one those would you prefer? or would you suggest a different one.
The first one is much cleaner, and more closely mirrors the other overloaded literals. It seems that in most cases the intermediate list should be eliminated via build/foldr fusion. Did you do any testing to figure out why that fusion wasn't happening? (I.e., *why* is the generic approach faster?)
The only other thing I'll mention is that for overloadable strings, part of the reason why they're so fast is that string literals are stored a la C, and so the conversion to ByteString and Text requires minimal work. I wonder if you might be able to leverage a similar technique for representing list literals as vectors, which are then inflated to lists/vectors/sets/whatever at runtime.
-- Live well, ~wren _______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
