Is PartialFunction a SAM?

[DarkDimius]

In order to implement a PartialFunction one needs to define both

def isDefinedAt(x: T)
def apply(x: T)

so there are two abstract methods.

PartialFunction currently is treated specially,
https://github.com/lampepfl/dotty/blob/master/src/dotty/tools/dotc/core/Types.scala#L2723-L2730
forcing it to be treated as a SAM.

The comment says that

// In the future it would be better to merge
// Function and PartialFunction, have Function1 contain a isDefinedAt method
//     def isDefinedAt(x: T) = true
// and overwrite that method whenever the function body is a sequence of
// case clauses.

The comment means that Function1 instances which are PartialFunctions could not be SAMs as they need to override a method.

[odersky]

Yes, I think that's the consequence. Need to evaluate implementation
complexity of this approach.

On Tue, Apr 28, 2015 at 10:28 AM, Dmitry Petrashko <notifications@github.com

wrote:

In order to implement a PartialFunction one needs to define both

def isDefinedAt(x: T)
def apply(x: T)

so there are two abstract methods.

PartialFunction currently is treated specially,

https://github.com/lampepfl/dotty/blob/master/src/dotty/tools/dotc/core/Types.scala#L2723-L2730
forcing it to be treated as a SAM.

The comment says that

// In the future it would be better to merge
// Function and PartialFunction, have Function1 contain a isDefinedAt method
// def isDefinedAt(x: T) = true
// and overwrite that method whenever the function body is a sequence of
// case clauses.

The comment means that Function1 instances which are PartialFunctions
could not be SAMs as they need to override a method.

—
Reply to this email directly or view it on GitHub
#504.

Martin Odersky
EPFL

[odersky]

I'll take a look at this now. Will try to write a phase that converts partial functions (and some other "false" SAMs) to anonymous classes.

[retronym]

I think that's a good tool to have; we fall back to this translation in scalac when the delambdafy transform would lift the lambda body to a method that triggers a VerifyError. This is done for lambdas in super-contructor call position: scala/scala#3616

[odersky]

Jason,

I thought at first we could just restrict SAM lambdas not to use super
accessors (or fields, or outer pointers). But it would indeed be nicer to
the user and easier to specify to deal with it in the translations.

• Martin

On Wed, Apr 29, 2015 at 6:35 AM, Jason Zaugg notifications@github.com
wrote:

I think that's a good tool to have; we fall back to this translation in
indylambda when the delambdafy transform would be lift the lambda body
into a method in a way that leads to VerifyError. This is done for
lambdas in super-contructor call position: scala/scala#3616
scala/scala#3616

—
Reply to this email directly or view it on GitHub
#504 (comment).

Martin Odersky
EPFL

[retronym]

We envision adding support for Java's @FunctionalInterface annotation to help the programmer to opt in to those restrictions when defining a trait.

[DarkDimius]

#509. PartialFunction is not a Sam for JVM, we silently expand it to anonymous class.