regression(1.04): reset broken in VM; incorrect VM handling of var params

[timotheecour]

This regression was introduced in the recent reset PR (system.reset is no longer magic (#12937)) (the 2nd regression i found for this PR although this one is more serious).
The root cause is probably not #12937 per se, but what may be a bug in VM regarding var parameters / moves.

Example

type Atom* = object
  bar: int

proc main() =
  var s: seq[Atom]
  var atom: Atom
  for i in 0..<2:
    atom.bar = 5
    s.add atom
    atom.reset # BUG
    # atom = default(type(atom)) # works
    echo s

static: main()
main()

Current Output

nim c -r -f --warnings:off t0855.nim

@[(bar: 5)]
@[(bar: 5), (bar: 0)] # BUG HERE
Hint: operation successful (52766 lines compiled; 0.386 sec total; 58.273MiB peakmem; Debug Build) [SuccessX]
@[(bar: 5)]
@[(bar: 5), (bar: 5)]

Expected Output

@[(bar: 5)]
@[(bar: 5), (bar: 5)]
Hint: operation successful (52766 lines compiled; 0.386 sec total; 58.273MiB peakmem; Debug Build) [SuccessX]
@[(bar: 5)]
@[(bar: 5), (bar: 5)]

Possible Solution

• instead of reverting system.reset is no longer magic #12937, the underlying root cause in VM should be fixed as this easily pops up in other contexts

Additional Information

D20191230T155433

• regression introduced in system.reset is no longer magic #12937

[Araq]

As a workaround, we can also specialize system.reset in the VM and map it to the old implementation, avoids the calling overhead in the VM.