Null is a subclass of AnyRef... isn't it?

2009/6/15 Andrew Forrest :
> Hi all,
>
> I'm seeing strange (to me) behaviour with type bounds and null.
>
> def doStuff[T <: AnyRef]() {
>        var t: T = null // compile error here
>        println(t)
> }
>
> ...doesn't compile (2.7.4)— “type mismatch; found   : Null(null) required:
> T”. Whereas, as I understand it, I'm saying that the array must be an array
> of reference type, and any reference type value may be null... The following
> DOES compile:
>
> def doStuff[T >: Null]() {
>        val t: T = null
>        println(t)
> }
>
> Now that's actually 'better' code in that the type bound is more explicit
> about what I require of the type, but I thought that the first one should
> work.
>
> What say you?
>
> Is it a consequence of the fact that trait NotNull extends AnyRef too…?

Null is a subtype of AnyRef, but it doesn't follow that it's a subtype
of everything which is a subtype of AnyRef (example: String is a
subtype of AnyRef, but it doesn't follow from that that String is a
Subtype of Map).

NotNull is one reason for this, but a more fundamental one is that
Nothing <: AnyRef but Null is not a subtype of Nothing (Nothing is the
type which has no values - the only way a method can have return type
Nothing is to never return normally. If Null <: Nothing then null
would be a value of type Nothing, which is not permitted)

On Mon, Jun 15, 2009 at 12:59 PM, Andrew Forrest wrote:
> Hi all,
>
> I'm seeing strange (to me) behaviour with type bounds and null.
>
> def doStuff[T <: AnyRef]() {
>        var t: T = null // compile error here
>        println(t)
> }
>
> ...doesn't compile (2.7.4)— “type mismatch; found   : Null(null) required:
> T”. Whereas, as I understand it, I'm saying that the array must be an array
> of reference type, and any reference type value may be null... The following
> DOES compile:
>
> def doStuff[T >: Null]() {
>        val t: T = null
>        println(t)
> }
>
> Now that's actually 'better' code in that the type bound is more explicit
> about what I require of the type, but I thought that the first one should
> work.
>
> What say you?
>
Null is a subtype of any class or trait that inherits of AnyRef. But
it's not automatically a subtype of a type parameter that's bounded by
an AnyRef. So the behavior you are seeing is as expected.

Cheers

On Mon, Jun 15, 2009 at 3:59 AM, Andrew Forrest <andrew [at] dysphoria [dot] net> wrote:

Hi all,

I'm seeing strange (to me) behaviour with type bounds and null.

def doStuff[T <: AnyRef]() {
       var t: T = null // compile error here
       println(t)
}

Null is not a subtype of Nothing.

Has anybody else noticed that Scala makes you speak in double negatives?

All other answers are correct, and you couldn't find more knowledgeable persons to do it. But since I found the answers a bit too indirect, I'll give you mine, which is imprecise (or, strictly speaking, incorrect :), but might be easier to understand on a practical level.   Nothing is a subtype of everything, and null is not a valid value o Nothing. So if you get AnyRef, you could assign null to it, but if you get Nothing, you can't.   By the way, you can pass "null" as value o "Any".     On Mon, Jun 15, 2009 at 7:59 AM, Andrew Forrest <andrew [at] dysphoria [dot] net> wrote:

Hi all,

I'm seeing strange (to me) behaviour with type bounds and null.

def doStuff[T <: AnyRef]() {
       var t: T = null // compile error here
       println(t)
}

...doesn't compile (2.7.4)— “type mismatch; found   : Null(null) required: T”. Whereas, as I understand it, I'm saying that the array must be an array of reference type, and any reference type value may be null... The following DOES compile:

def doStuff[T >: Null]() {
       val t: T = null
       println(t)
}

Now that's actually 'better' code in that the type bound is more explicit about what I require of the type, but I thought that the first one should work.

What say you?

Is it a consequence of the fact that trait NotNull extends AnyRef too…?


--Andrew

--
Daniel C. Sobral

Something I learned in academia: there are three kinds of academic reviews: review by name, review by reference and review by value.

Cool, thanks everybody.

On 15 Jun 2009, at 15:58, Daniel Sobral wrote:

All other answers are correct, and you couldn't find more knowledgeable persons to do it. But since I found the answers a bit too indirect, I'll give you mine, which is imprecise (or, strictly speaking, incorrect :), but might be easier to understand on a practical level.   Nothing is a subtype of everything, and null is not a valid value o Nothing. So if you get AnyRef, you could assign null to it, but if you get Nothing, you can't.   By the way, you can pass "null" as value o "Any".     On Mon, Jun 15, 2009 at 7:59 AM, Andrew Forrest <andrew [at] dysphoria [dot] net> wrote:

Hi all,

I'm seeing strange (to me) behaviour with type bounds and null.

def doStuff[T <: AnyRef]() {
       var t: T = null // compile error here
       println(t)
}

...doesn't compile (2.7.4)— “type mismatch; found   : Null(null) required: T”. Whereas, as I understand it, I'm saying that the array must be an array of reference type, and any reference type value may be null... The following DOES compile:

def doStuff[T >: Null]() {
       val t: T = null
       println(t)
}

Now that's actually 'better' code in that the type bound is more explicit about what I require of the type, but I thought that the first one should work.

What say you?

Is it a consequence of the fact that trait NotNull extends AnyRef too…?


--Andrew

--
Daniel C. Sobral

Something I learned in academia: there are three kinds of academic reviews: review by name, review by reference and review by value.