Synthesize a PartialFunction from function literal

[dwijnand]

Allow val pf: PartialFunction[Int, Int] = { _ => 3 }

[sjrd]

Hum, can I ask why?

[dwijnand]

Sure! Because it's safe and the workaround is annoying (ironically the reverse, being able to define a function from case statements, is unsafe but already possible).

[adriaanm]

Looking good. Suggested a few improvements/tweaks.

[adriaanm]

Nice! Few more suggestions.

[lrytz]

Looks good!

[dwijnand]

(Added some overloading and eta-expanding tests, which came up in conversation.)

[dwijnand]

Anyone else want to review this?

[som-snytt]

It's not too late to add language in 6.23.1 of the spec. There's already a slot for PF because of x => x match shape case, just add "if the expected type is a PF", etc.

[dwijnand]

Good thinking. I actually dropped the "is of the shape", seeing as it's no longer true in 2.13, right?

Welcome to Scala 2.13.0 (OpenJDK GraalVM CE 19.0.0, Java 1.8.0_212).
Type in expressions for evaluation. Or try :help.

scala> def takeFun(f: Int => String) = f.getClass
takeFun: (f: Int => String)Class[_ <: Int => String]

scala> takeFun(x => x match { case 1 => "one"; case _ => "other" })
res0: Class[_ <: Int => String] = class $$Lambda$2053/217123986

scala> def takeFun(f: Int => String) = f.isInstanceOf[PartialFunction[Int, String]]
takeFun: (f: Int => String)Boolean

scala> takeFun(x => x match { case 1 => "one"; case _ => "other" })
res1: Boolean = false

[som-snytt]

It's the other case you preserve in your patch:

scala> def f(pf: PartialFunction[Int, String]) = pf.isDefinedAt(27)
f: (pf: PartialFunction[Int,String])Boolean

scala> f { case 42 => "" }
res0: Boolean = false

scala> f(x => x match { case 42 => "" })
res1: Boolean = false

So the last is still false not true with this change.

A couple of paragraphs later in the spec says how the PF is constructed. I want to suggest it's the same as 8.5 for "pattern matching anonymous functions", the bare case block. The piece to add is how a default case is supplied with the function body, so that isDefinedAt is true. (Edit: maybe don't use those words.)

The list of "expected types" can now just say PF without the conditional; you were right to drop it there, but add it back in the subsequent paragraph: If it has this shape, isDefinedAt is derived from the cases; otherwise, isDefinedAt is true.

(I think of the translation as if it were written as for 8.5, which is how it's implemented, but the spec is only in terms of adding the isDefinedAt member.)

[dwijnand]

Thanks for your feedback, @som-snytt. Let me know if I captured everything you were indicating or if I missed or misunderstood anything - I'm happy to make further changes.

[som-snytt]

Sorry, I intended to come up with some draft words this weekend. But it was the weekend.

Here is my simplification, which says that your PR is the usual case, but in this special case where the body looks like a "pattern matching anonymous function", it's handled in a similar way. I wasn't gutsy enough to say that directly, because it would require linking to that section. (Maybe this text should add, See [link to anchor for PMAFs].)

When a PartialFunction is required, an additional member isDefinedAt
is synthesized, which simply returns true.
However, if the function literal has the shape x => x match { $\ldots$ },
then isDefinedAt is derived from the pattern match in the following way:
each case from the match expression evaluates to true,
and if there is no default case,
a default case is added that evalutes to false.

[lrytz]

Still LGTM :-) I think we can merge.

[nafg]

In some places it's misspelled Funtion1

[dwijnand]

#8430 (with an unintentional typo in the commit message. English is hard.)