regression in f interpolator if interpolated value is an if/else

[scabug]

% scalac-hash 35209fc sandbox/test.scala
sandbox/test.scala:2: error: type mismatch;
 found   : Any
 required: Double
   f"${if (true) 2.5 else 2.5}%.2f"
       ^
one error found
% scalac-hash 35209fc~1 sandbox/test.scala

% cat sandbox/test.scala
object Test {
   f"${if (true) 2.5 else 2.5}%.2f"
}

Regressed in: scala/scala@35209fc

Might also be linked to the "boxity" of fast track macros.

I reckon changing:

def f(args: Any*): String = macro ???

to:

def f[A](args: A*): String = macro ???

Might alter behaviour. We experienced similar subtleties with quasiquotes.

[scabug]

Imported From: https://issues.scala-lang.org/browse/SI-8512?orig=1
Reporter: @retronym
Affected Versions: 2.11.0-M3

[scabug]

@gourlaysama said:
It actually happens in 2.10 too:

$ scalac-hash v2.10.0 sandbox/test.scala
sandbox/test.scala:2: error: type mismatch;
 found   : Any
 required: Double
   f"${if (true) 3.5 else 2.5}%.2f"
       ^
one error found
$ cat sandbox/test.scala
object Test {
   f"${if (true) 3.5 else 2.5}%.2f" // note the difference with the one in the description
}

[scabug]

@som-snytt said:
The case as described regressed in M3.

I think I still have my "I'm a lubber not a typer" t-shirt from back then.

The case with non-identical alternatives, and who ever needs that, is being unassigned and kicked down the road to foster work stealing...

I don't suppose we could change the title? I lose three months' life expectancy every time I see "regression in f interpolator."

[scabug]

@densh said (edited on Aug 2, 2014 4:35:42 PM UTC):
There was a very similar issue with quasiquotes.

The quick solution for this is to change f interpolator signature from current:

def f(args: Any*): String = macro ???

to:

def f[T](args: T*): String = macro ???

It will force typer not to be lazy and properly infer specific types to arguments of the interpolator. At least this is how quasiquotes implementation works around this issue.

[scabug]

@som-snytt said (edited on Aug 3, 2014 3:41:46 PM UTC):
I guess you guys know what you're talking about (because I don't know what I'm talking about), but I might like to supply hetero args:

scala> f"${3.14}%f rounds to ${3}%d"
<console>:8: error: type mismatch;
 found   : Double(3.0)
 required: Int
              f"${3.14}%f rounds to ${3}%d"
                                      ^

But FTR, I was prepared to apply whatever snake oil greases the wheel.

[scabug]

@densh said:
The way I expect it to work is following:

1. User calls f interpolator with heterogeneous arguments:

StringContext(...).f(1, 2.0, 'three)

2. Typer infers useless type for T

StringContext(...).f[Any](1, 2.0, 'three)

3. Macro comes and inspects .tpe of the arguments completely disregarding inferred T.

This way tpes should get attributed with proper concrete types rather than Any. I don't expect any problems wrt heterogeneous arguments.

[scabug]

@som-snytt said:
The 3 is promoted early to 3.0.

However, it seems to work by asking for a more usefully useless inferred type:

def f[A >: Any](args: A*): String = macro ???

scala> f"${3.14}%f rounds to ${3}%d"
res0: String = 3.140000 rounds to 3

[scabug]

@som-snytt said:
scala/scala#3909

Thanks again for the extra words.

[scabug]

@som-snytt said (edited on Aug 3, 2014 7:48:51 PM UTC):
Of course, -Xlint will warn you about the uselessly inferred type, even if it's useful.

You don't want to -Xlint:-warn-any,_ whenever you interpolate.

scala> f"${3.14}%f is ${"pi"}%s"
<console>:8: warning: a type was inferred to be `Any`; this may indicate a programming error.
              f"${3.14}%f is ${"pi"}%s"
                  ^
res0: String = 3.140000 is pi

scala> import reflect.runtime._
import reflect.runtime._

scala> import universe._
import universe._

scala> q"val i = 4"
res1: reflect.runtime.universe.ValDef = val i = 4

scala> val j = 5
j: Int = 5

scala> q"{ $res1 ; $j }"
<console>:16: warning: a type was inferred to be `Any`; this may indicate a programming error.
              q"{ $res1 ; $j }"
                   ^
res2: reflect.runtime.universe.Tree =
{
  val i = 4;
  5
}

[scabug]

@gkossakowski said:
The 2.11.2 is out so I'm rescheduling the issue for 2.11.3.