Hello, On Aug 18, 2010, at 13:44 , Duncan Coutts wrote:

This is just an issue of specification vs implementation. The spec from the H98 report is

and = foldr (&&) True [snip] Note that 'equal' includes all partial and total lists, so you can rely on the above spec to reason about the behaviour on infinite lists and expect that reasoning to work for correct H98 implementations.

ok, thank you. It was precisely the reason for my question. I used the fold-based definition in my ramblings and then bothered to look into the Prelude code. ;)

That said, there are a couple functions where the obviously sensible, and standard implementations differ from the H98 spec for some partial values. Notably splitAt is specified badly in the H98 report (it is too lazy).

By the way, does some good reading on streams in Haskell exist? I am interested primarily in a theoretical foundation, which should however be somehow related to Haskell. My guess would be the relation of streams, represented as lazy, not completely evaluated lists, and the evaluation strategies. The practical part is easier, feeding [1..] to various functions in GHCi. I have found [Ennals2003optimistic], but it seems to go in wrong direction. [Ennals2003optimistic]: Robert Ennals, Simon Peyton Jones. "Optimistic evaluation: an adaptive evaluation strategy for non-strict programs". ICFP 2003. http://portal.acm.org/citation.cfm?id=944705.944731 Greetings, Oleg

On 8/24/10 1:55 PM, Jan-Willem Maessen wrote:

On Wed, Aug 18, 2010 at 9:56 PM, wren ng thornton wrote:

...

The thing I find puzzling is that the foldr is inlined. The (regular) clever optimizations for build/foldr seem like they should already handle this without the need for the extra rules. I wonder how much is gained by specializing to (&&)/True instead of relying on the regular deforestation?

The above code does not inline and / or, *unless they are fused using the RULES.* There's not really any benefit to inlining them otherwise, and it duplicates code.

Ah, that makes sense. Thanks. -- Live well, ~wren