Re: [Haskell-cafe] Are bottoms ever natural?

Siddharth, how would you deal with functions that terminate for some arguments/inputs but do not terminate for the others ? Alexey. On Tue, 2017-12-19 at 07:20 +0000, (IIIT) Siddharth Bhat wrote:

...

Is that really true, though? Usually when you have an infinite loop, you have progress of some sort. Infinite loops with no side effects can be removed from the program according to the C standard, for example. So, in general, we should allow programmers to express termination / progress, right? At that point, no computation ever "bottoms out"? Shouldn't a hypothetical purely functional programming language better control this (by eg. Forcing totality?) It seems like we lose much of the benefits of purity by muddying the waters with divergence. From an optimising compiler perspective, Haskell is on some weird lose-lose space, where you lose out on traditional compiler techniques that work on strict code, but it also does not allow the awesome stuff you could do with "pure" computation because bottom lurks everywhere. What neat optimisations can be done on Haskell that can't be done in a traditional imperative language? I genuinely want to know. What are your thoughts on this? Cheers Siddharth

On Tue 19 Dec, 2017, 08:09 Brandon Allbery, wrote:

...

...

Define "natural".

You might want to look into the concept of Turing completeness. One could define a subset of Haskell in which bottoms cannot occur... but it turns out there's a lot of useful things you can't do in such a language. (In strict languages, these often are expressed as infinite loops of one kind or another. Note also that any dependency on external input is an infinite loop from the perspective of the language, since it can only be broken by the external entity providing the input.)

On Tue, Dec 19, 2017 at 1:47 AM, (IIIT) Siddharth Bhat wrote:

...

...

I've been thinking about the issue of purity and speculation lately, and from what little I have read, it looks like the presence of bottom hiding inside a lazy value is a huge issue.

How "natural" is it for bottoms to exist? If one were to change Haskell and declare that any haskell value can be speculated upon, what ramifications does this have?

Is it totally broken? Is it "correct" but makes programming unpleasant? What's the catch?

Thanks, Siddharth