bugs with var result

[timotheecour]

while working on this fix #13959 I identified 4 other issues with var result

Example 1

• iterator returning tuple of var in VM gives CT error

when true: # D20200413T140233
  template fun2() =
    block:
      var m = 1
      var m2 = 1
      iterator test3(o: var int): (var int, var int) =
        yield (o, m2)
      for ti in test3(m):
        ti[0]+=3
        ti[1]+=4
      doAssert (m, m2) == (4, 5)
  fun2() # ok
  static: fun2() # BUG: Error: attempt to access a nil address kind: rkInt

Current Output

Error: attempt to access a nil address kind: rkInt

Expected Output

works

Example 2

• proc returning tuple of var gives SIGSEGV or cgen static error, but should work, just like it does for iterator or for non-tuple

when true: # D20200413T140527
  # nim c: SIGSEGV
  # nim cpp: error: call to implicitly-deleted default constructor of 'tyTuple__oq0LVLj6AbX8hE12f64tQQ'
  template fun2() =
    block:
      var m = 1
      var m2 = 1
      proc test3(o: var int): (var int, var int) =
        (o, m2)
      test3(m)[0] += 3
      echo (m, m2)
  fun2()

Example 3

• proc returning tuple of var gives CT error in VM

when true: # D20200413T140826
  # Error: attempt to access a nil address kind: rkInt
  template fun2() =
    block:
      var m = 1
      var m2 = 1
      proc test3(o: var int): (var int, var int) =
        (o, m2)
      test3(m)[0] += 3 # BUG here
  static: fun2()

Example 4: accept invalid (un-initialized var result)

• proc returning var result that isn't initialized should give a semcheck CT error at least for simple cases

when true: # D20200413T141159
  # nim c: SIGSEGV
  # nim cpp: error: declaration of reference variable 'result' requires an initializer
  proc identity(): var int = discard
  let a = identity()

EDIT: at least gives a Warning: Cannot prove that 'result' is initialized. This will become a compile time error in the future. [ProveInit]

Example 5

when true: # D20200711T174301 gives SIGSEGV, should give CT error (1 is not an lvalue)
  proc main(): (var int, int) =
    result[0] = 1
  main()[0].inc

Additional Information

• devel 1.3.1 c269964

[krux02]

I don't think Example 4 would be least important. It should be a proper compilation error.

[timotheecour]

maybe it's hard to do so in the general case (hence ProveInit being a warning only?) but I'm not sure which cases cause the compiler not to be sure; but yes, when compiler knows for sure that it's not initialized, it should give CT error instead of ProveInit warning

EDIT: maybe something like this could be hard to prove for compiler:

    proc fun(a: var int): var int =
      {.emit: "`result` = `a`;".}