Avoid spurious suboptimal code errors for code produced by implicit arg inference

[julien-truffaut]

I am getting the following error when I try to define a scalacheck property that generates an object with miniboxed annotation:

Error:(42, 10) Using the type argument "Any" for the miniboxed type parameter T1 of class AbstractMiniboxedFunction1 is not specific enough, as it could mean either a primitive or a reference type. Although class AbstractMiniboxedFunction1 is miniboxed, it won't benefit from specialization:
    check((as: ConsList[Int]) =>
         ^

link to the code: https://github.com/julien-truffaut/brique/blob/miniboxing/tests/src/test/scala/brique/tests/ConsListSpec.scala#L42

[VladUreche]

That's a warning that somehow ends up being seen as an error -- I will see what's happening. Thanks for the report!

[julien-truffaut]

I think it is my project configuration that transforms all warning on errors

[VladUreche]

In that case, maybe you can run the compilation with -P:minibox:warn-off, so miniboxing stops issuing warnings for every little optimization that could be done.

[VladUreche]

This one took all night to track down, literally:

$ cat gh-bug-183.scala 
package miniboxing.tests.bug183

import scala.language.implicitConversions

class ConsListSpec {

  class Pretty
  implicit def f(a: Any): Pretty = ???
  def check[P](implicit p: P => Pretty): Unit = ???

  check[String]
}

$ mb-scalac gh-bug-183.scala
gh-bug-183.scala:11: warning: Using the type argument "Any" for the miniboxed type parameter T1 of class AbstractMiniboxedFunction1 is not specific enough, as it could mean either a primitive or a reference type. Although class AbstractMiniboxedFunction1 is miniboxed, it won't benefit from specialization:
  check[String]
       ^
one warning found

[VladUreche]

Since it's quite late, I will leave this bug open, since -P:minibox:warn-off fixes it.

[VladUreche]

Quick update on this bug: In the morning I tried the simple idea I had, but it's not smart enough to discriminate well between real and spurious errors -- it hides real errors and leaves some spurious ones :-1:

[VladUreche]

To explain the problem: the warning is legit, the code it warns about is indeed suboptimal. But the code is not written by you, it's the result of implicit argument inference, so I don't even know what to do about it:

• it's legit, the code is really suboptimal
• it's not the programmer's code, so there should be a special warning saying it's inferred
• by the time the miniboxing plugin sees this code, there's no trace that it was introduced by implicit lookup anymore, so it looks exactly like programmer code

Still don't know what to do about this...

[julien-truffaut]

would it be possible to disable warning on locally? I mean it is very useful to get the warning but at the moment they are cases where you cannot do anything about it. Basically something similar to @unchecked for pattern match.