regression(1.2 => 1.4 onwards): Block-local ref types with the same name lead to bad codegen: no member named

[timotheecour]

basically same as fixed #5170 but for ref types

Example

when true: # D20201008T141857
  block:
    type Foo = ref object
      x1: int
    let f1 = Foo(x1: 1)
  block:
    type Foo = ref object
      x2: int
    let f2 = Foo(x2: 2)

Current Output

@mt11116.nim.c:422:10: error: no member named 'x2' in 'struct tyObject_FoocolonObjectType___ktEbVfDBqcSUCx9b6syN8LA'
                (*T4_).x2 = ((NI) 2);

Expected Output

works

Additional Information

basically same as fixed #5170 (which was fixed) but for ref types

• nim devel 1.3.7 3eaacac and prior
• EDIT: very similar to regression(1.2=>1.4 onwards): enum with same name in different scope gives wrong RT results #17162

[timotheecour]

further info: it's a regression, it used to work in 1.2.0 (through 1.2.8)

with ref object

works in vm, js
fails in RT

when true:
  # D20201008T141857
  template fn() =
    block:
      type Foo = ref object
        x1: int
      let f1 = Foo(x1: 1)
      echo f1.x1
    block:
      type Foo = ref object
        x2: int
      let f2 = Foo(x2: 2)
      echo f2.x2
  static: fn()
  fn()

with object

works in vm, js, RT

when defined case5:
  template fn() =
    block:
      type Foo = object
        x1: int
      let f1 = Foo(x1: 1)
      echo f1.x1
    block:
      type Foo = object
        x2: int
      let f2 = Foo(x2: 2)
      echo f2.x2
  static: fn()
  fn()

[timotheecour]

git bisect shows it's caused by 2629d61 #15091; so this is in fact the same root cause as #17162 (we could close 1 of them as duplicate, or better yet, fix the regression)

[ringabout]

@timotheecour
not sure whether it is a regression, please double check the first example on Nim playground.

[timotheecour]

@xflywind indeed, #15526 (comment) was a regression but top example is not a regression, so there's a partial regression here; in any case i'm fixing all these here: #17710