"Lazy" and the spec

On Thursday, May 27, 2010 06:04:27 pm Seth Tisue wrote:
> I understand why the compiler complains here:
>
> scala> def f(n: Int) = { val fac: Int => Int = n => if(n == 0) 1 else n *
> fac(n - 1); fac(n) } :6: error: forward reference extends over
> definition of value fac def f(n: Int) = { val fac: Int => Int = n => if(n
> == 0) 1 else n * fac(n - 1); fac(n) } ^
>
> It's because of this language at the beginning of chapter 4 in the spec:
>
> "There is a restriction on forward references in blocks: In a statement
> sequence s1 . . . sn making up a block, if a simple name in si refers to
> an entity defined by sj where j >= i, then none of the definitions
> between and including si and sj may be a value or variable definition."
>
> However, I can make the error go away by adding "lazy":
>
> scala> def f(n: Int) = { lazy val fac: Int => Int = n => if(n == 0) 1 else
> n * fac(n - 1); fac(n) } f: (n: Int)Int
>
> scala> f(5)
> res0: Int = 120
>
> And that's a good thing, of course! But what language in the spec
> justifies it? Section 5.2 says how "lazy" behaves at runtime, but it
> doesn't say anything about there being an exception to the chapter 4
> rule at compile time. After all, "lazy val" is still technically a
> value definition.
>
> Is this a hole in the spec?

It seems likely. The lazy parts of the spec were added after the forward
reference section. I submitted a patch:
http://lampsvn.epfl.ch/trac/scala/ticket/3468

and a bug:
http://lampsvn.epfl.ch/trac/scala/ticket/3503

-Mark