Well, I think people use the functions that come with an implementation as the de facto spec. So we should update the spec to reflect reality. (BTW, I wouldn't be surprised if I'm to blame for some of the discrepancies, even though I tried to be careful when I implemented more efficient versions for the Prelude functions for hbc.) -- Lennart On Mar 22, 2007, at 00:58 , Duncan Coutts wrote:
On Wed, 2007-03-21 at 23:41 +0000, Lennart Augustsson wrote:
Of course it means that a function must have exactly the same strictness. If it doesn't have the same strictness it's not the same function.
Of course.
I agree the report should be fixed where the functions in the report do not have the appropriate strictness. But this is a change to the report.
It's interesting to note that the implementations in base for some functions in Data.List have been inconsistent with the report for years and nobody has noticed (or at least if they noticed they didn't complain very loudly).
For example the standard implementations of partition and splitAt have been inconsistent with the H98 spec forever. I say forever since the implementations were consistent with the Haskell 1.2, 1.3 and 1.4 specifications but the Haskell98 report changed the specification and the implementations were never updated.
Similarly, GHC has had an incorrect implementation of genericTake for some years and no-one has noticed. The current implementation is inconsistent with Haskell98 and looking back, GHC 3.02 was inconsistent with Haskell 1.4 which as far as I can tell was the first version of the spec to define genericTake.
Duncan