On 8/17/06, Chris Capel <pdf23ds@gmail.com> wrote:
For instance, wherever you have x <- y !z you should be able to add x <- y !z / x-err ERR x-err <- y without ever changing the result on valid input.
I don't think that will work in general. Consider this: a <- x / b x <- y !z b <- y !c If you change x as you propose, you might mark something which is not an error as an error. You will read something as an erroneous x instead of as a correct b. mu'o mi'e xorxes