Filed. Bug #8952. On Wed, Apr 2, 2014 at 3:41 PM, wren romano wrote:

On Tue, Apr 1, 2014 at 3:02 PM, Dan Doel wrote:

...

Specifically, consider:

case Nothing of !(~(Just x)) -> 5 Nothing -> 12

Now, the way I'd expect this to work, and how I think the spec says it works, is that my Nothing is evaluated, and then the irrefutable ~(Just x) matches Nothing, giving a result of 5. In fact, GHC warns about overlapping patterns for this.

It's sensible to give an overlap warning --that is, assuming we don't want overlap to be an error-- since the irrefutable pattern matches everything, and adding bangs doesn't change what values are matched (it only changes whether we diverge or not).

However, I have no idea how top-level bang in case-expressions is supposed to be interpreted. If anything, it should be ignored since we must already force the scrutinee to WHNF before matching *any* of the clauses of a case-expression. However, I thought bangs were restricted to (1) immediately before variables, and (2) for top-level use in let/where clauses...

In any case, following the standard desugaring of the specs:

case Nothing of !(~(Just x)) -> 5 ; Nothing -> 12

...

,

=== { next to last box of < http://www.haskell.org/ghc/docs/latest/html/users_guide/bang-patterns.html the proposed clause (t) for section 3.17.3, figure 4 }

Nothing `seq` case Nothing of ~(Just x) -> 5 ; Nothing -> 12

=== { Haskell Report, section 3.17.3, figure 3, clause (d) }

Nothing `seq` (\x -> 5) (case Nothing of Just x -> x)

Which most definitely does not evaluate to 12. Either the specs are wrong (dubious) or the implementation is. File a bug report.

-- Live well, ~wren