Proposal: making inits and tails less strict

[Isaac Dupree]

On 03/18/11 04:06, Bas van Dijk wrote:
> On 17 March 2011 21:09, Don Stewart<dons00 at gmail.com>  wrote:
>> Could the strictness properties be stated in the specification (i.e.
>> include some QuickCheck properties that define the strictness)
>
> Good point. I will add the following strictness properties to the docs:
>
> inits ⊥ = [] : ⊥
> tails ⊥ = ⊥ : ⊥
>
> I'm not sure if we should add QC properties to the docs but we can have:
>
> prop_lazyInits xs = head (inits xs) == []
> ...

By the way, QuickCheck can't really check laziness properties.  It 
always provides finite, non-bottom input.  (Well unless you make a 
generator that is somewhat out of its usual mold.)  I think there's a 
library like "StrictCheck" that is meant to check strictness/laziness 
properties?

For the Report, I sort of like the idea of just writing things like you said
 > inits ⊥ = [] : ⊥
 > tails ⊥ = ⊥ : ⊥

Although consider, it's been interpreted that the Report already has 
binding strictness requirements (strictness must be equivalent to the 
Report's implementation).  We might want to change that 
means-of-specification if some other means is better.  I think that's a 
good, unambiguous means though, unless we *want* the Report to 
underspecify strictness.

Other thought about the proposal: Often, changing strictness has 
unintuitive effects on performance (I don't remember if that came up for 
this particular proposal sometime in the past, or for some different 
list function that libraries@ had been wishing were slightly lazier).