Henning Thielemann wrote:
On Sat, 22 Nov 2008, Janis Voigtlaender wrote:
Definitely. And that surfaces even in quite innocently looking programs and statements about them. The introductory example of the following technical report may be amusing in that respect:
In example 1, I don't see on the one hand, why 'takeWhile (null.tail)' could fail with "tail: empty list",
Because the (GHC) semantics described in the following paper: http://doi.acm.org/10.1145/301631.301637 says so. You can also check it by calculating with the definitions from Section 3 and Figure 8 of the above technical report. Or see it demonstrated on slide pages 32-38 of: http://wwwtcs.inf.tu-dresden.de/~voigt/nwpt2008-slides.pdf
since all lists in '[[i] | i <- [1..(div 1 0)]]' are non-empty (namely singletons).
This does not really have anything to do with the above, but I may just as well say that all lists in '[[i] | i <- [1..(div 1 0)]]' are empty. Or that they are all of length 17. Because there are no lists in '[[i] | i <- [1..(div 1 0)]]'. (Note that the "supply" [1..(div 1 0)] is itself erroneous.) The "tail: empty list" failure mentioned above really has nothing at all to do with the concrete expression '[[i] | i <- [1..(div 1 0)]]'. Any other erroneous expression, such as just 'error "div-by-0"' would lead to the same result.
On the other hand, aren't those imprecise error problems not just proofs that mixing up errors and exceptions (here treating errors as exceptions) is a bad thing? 'error' is only a candy version of 'undefined' for simplifying debugging. If all 'error's are replaced by 'undefined' (plain bottom) then 'takeWhile p (map h l)' and 'map h (takeWhile (p.h) l)' behave also visually identical, don't they?
Yes, if all 'error's are replaced by 'undefined', then the two expressions are semantically equivalent. But I don't buy this as a decisive argument in the "errors vs. exceptions" debate. Ciao, Janis. -- Dr. Janis Voigtlaender http://wwwtcs.inf.tu-dresden.de/~voigt/ mailto:voigt@tcs.inf.tu-dresden.de