Proposal: Infer return type of expression bodied local functions

[gulshan]

When local functions were first proposed, the return type was supposed to be inferred. But the idea was complex to implement and was postponed. Expression bodied local functions will be the simplest to infer the return type. Please do this in some upcoming 7.x release. Local functions with single return statements can also be considered for inferring. Example (not that good)-

int add(int a, int b) => a+b;
// can be written as
var add(int a, int b) => a+b;

ref: dotnet/roslyn#2930 (comment)

[gafter]

Now that local functions can be recursive and can forward-reference other local functions, supporting inference in this way isn't a cleanly defined problem. I don't believe we're ever likely to do it.

[HaloFour]

@gafter

Could the compiler be made to check for those situations and allow inference of the return value where the local function isn't recursive or invoke other local functions from which these complications would arise?

#249 (comment)

[jnm2]

var for return type feels so wrong but it can't be helped I guess. auto would make sense but var has always meant "create a local variable."

[chergey]

@jnm2 auto may be unavailable (i.e. break present codebase)

[gafter]

@HaloFour

Could the compiler be made to check for those situations and allow inference of the return value where the local function isn't recursive or invoke other local functions from which these complications would arise?

No, not without that being described in the language specification somehow. I don't think there is a clean way to do that.

[bondsbw]

For that matter, would it be possible to infer the entire signature, given that inference only needs to search locally?

var add(a, b) => a+b;
var x = add(1, 5);

[bondsbw]

And then why not allow in the same block:

var y = add(1.5, 2.25);

It would generate two overloads for add, one for (int, int) => int and another (double, double) => double.

[onionhammer]

how about let value (int x, int y) => x * y ;P

[lucasteles]

@onionhammer i like it, 'let' is being considered for local immutable references to, at #421 and #188