Inconsistent null equality check for primitive types

[scabug]

scala> null.asInstanceOf[Double]
res0: Double = 0.0

scala> null.asInstanceOf[Double] == null
<console>:11: warning: comparing values of types Double and Null using `==' will always yield false
       null.asInstanceOf[Double] == null
                                 ^
res1: Boolean = true

scala> 0.0 == null
<console>:11: warning: comparing values of types Double and Null using `==' will always yield false
       0.0 == null
           ^
res2: Boolean = false

scala> null.asInstanceOf[Double] == 0.0
res6: Boolean = true

scala> val a = null.asInstanceOf[Double]
a: Double = 0.0

scala> a == null
<console>:12: warning: comparing values of types Double and Null using `==' will always yield false
       a == null
         ^
res7: Boolean = false

Same for Int and Long.

At least, the warning message seems incorrect, regardless that the boxing/unboxing logic might be fine (?)

[scabug]

Imported From: https://issues.scala-lang.org/browse/SI-9671?orig=1
Reporter: Dmytro Kondratiuk (dk14)
Affected Versions: 2.11.7
See #602

[scabug]

Dmytro Kondratiuk (dk14) said:
null.asInstanceOf[Double] == null compiles to the:

aconst_null
ifnonnull
The val-version compiles to the:

aconst_null
invokestatic unboxToDouble
putfield
aload_0
invokevirtual а
invokestatic boxToDouble
ifnonnull

So compiler just forgets to add unbox/box in the first case

[scabug]

@szeiger said:
This looks like more than just a problem with the warning message to me. I don't see why res1 and res7 should be different. Assigning to M5 because changing this has the potential to break code.

[scabug]

@lrytz said:
regressed here: scala/scala@60f1b4b

[scabug]

@lrytz said:
in short

def f1(a: Any) = "" + a
f1(null.asInstanceOf[Int])

first replaces the cast by an unbox. then it introduces a box to pass the argument to f1. the box(unbox(t)) is replaced by t, but should not

[scabug]

@lrytz said:
scala/scala#5176