[Haskell-cafe] Is "take" behaving correctly?
ctm at cs.nott.ac.uk
Wed Sep 12 05:03:02 EDT 2007
On 12 Sep 2007, at 00:38, Brent Yorgey wrote:
> On 9/11/07, PR Stanley <prstanley at ntlworld.com> wrote: Hi
> take 1000 [1..3] still yields [1,2,3]
> I thought it was supposed to return an error.
> If for some reason you want a version that does return an error in
> that situation, you could do something like the following:
> take' n _ | (n <= 0) = 
> take' n  | (n > 0) = error "take': list too short"
> | otherwise = 
> take' n (x:xs) = x : take' (n-1) xs
> I'm not sure why you'd want that, though. The standard
> implementation gracefully handles all inputs, and usually turns out
> to be what you want.
There are two sides to this form of "grace", though. I'll grant you,
quite hard to pull off a successful fraud without a degree of aplomb.
I always hope that key invariants and hygiene conditions can be built
the static semantics of programs, but where that's impractical
prefer if the dynamic behaviour makes it as easy as possible to
blame for errors. In such circumstances, I'd like operations to complain
about bogus input, rather than producing bogus output.
These GIGO problems do bite in real life. I spent a long time finding
in somebody else's typechecker which boiled down to the silly mistake of
zipping the wrong lists together. The right lists were guaranteed to
in length, but there was no reason for the wrong lists to do so.
unless you were doing fairly wacky stuff, they both happened to have
zero. Once weirder things arrived, the discrepancies showed up and the
truncations started happening, causing well-formed but ill-typed
to be generated by and propagated around the kernel of the system, which
eventually choked in some essentially random place. We had many suspects
before we found the culprit. If the programmer had used a version of zip
which bombed in off-diagonal cases, we'd have been straight there.
So I might very well consider having more than one version of take
depending on my requirements for its use. I might even consider the more
informative function which also returns the length of the shortfall.
In a dependently typed setting, I wouldn't write take and drop: I'd
lists with a view incorporating both, guaranteeing the length invariants
and that the sublists actually append to the input. But where shame is
unattainable, blame is really quite important.
All the best
More information about the Haskell-Cafe