Do not retain unmanaged allocations in generated ctors

[dcodeIO]

potentially fixes #761

[jtenner]

I can try this later. Does the example fail in #761

[jtenner]

I built the compiler, tested it against my example and I still get a rogue release problem. Something is wrong and I can't figure it out. Will investigate tomorrow.

[dcodeIO]

The example works for me

 (func $empty/ExampleClass#constructor (; 12 ;) (type $FUNCSIG$ii) (param $0 i32) (result i32)
  local.get $0
  i32.eqz
  if
   i32.const 0
   i32.const 0
   call $~lib/rt/tlsf/__alloc
   (; retain in question was here ;)
   local.set $0
  end
  local.get $0
 )
 (func $empty/returnExample (; 13 ;) (type $FUNCSIG$i) (result i32)
  i32.const 0
  call $empty/ExampleClass#constructor
 )
 (func $start:empty (; 14 ;) (type $FUNCSIG$v)
  call $empty/returnExample
  global.set $empty/t
 )

According to rtrace t is leaking there, which is correct since it hasn't been __freed manually and is ignored by GC.

[jtenner]

In this case, Yeah, it looks like the example is good but I've identified another unmanaged reference problem. My testing suite passes each test but crashes without explanation. Just need some time to get settled today and I promise to try and help.

[jtenner]

Great news. I tracked down the problem. It's actually within my testing software now.

https://github.com/jtenner/as-pect/blob/master/packages/assembly/assembly/internal/report/Actual.ts#L111

This is because the reporting mechanism does not account for @unmanaged class references. Thanks for your help in this.