[urgent] [CI] broke recently for all PRs: tests/assign/tassign.nim broke

[timotheecour]

eg:
https://travis-ci.org/nim-lang/Nim/jobs/440787636 for a PR #9336 that only touches doc/manual.rst

testament/tester '--nim:compiler/nim ' cat assign 
FAIL: tassign.nim C
Test "tests/assign/tassign.nim" in category "assign"
Failure: reOutputsDiffer
Expected:
TEMP=C:\Programs\xyz\bin
8 5 0 0
pre test a:test b:1 c:2 haha:3
assignment test a:test b:1 c:2 haha:3
Concrete '='
Concrete '='
Concrete '='
Concrete '='
Concrete '='
GenericT[T] '=' int
GenericT[T] '=' float
GenericT[T] '=' float
GenericT[T] '=' float
GenericT[T] '=' string
GenericT[T] '=' int8
GenericT[T] '=' bool
GenericT[T] '=' bool
GenericT[T] '=' bool
GenericT[T] '=' bool
Gotten:
TEMP=C:\Programs\xyz\bin
8 5 0 0
pre test a:test b:1 c:2 haha:3
assignment test a:test b:1 c:2 haha:3

even if we fix this, how could it have passed through CI's testing ?

EDIT
I'm suspecting #9318 since it last touched tests/assign/tassign.nim but how come the CI was green for that PR? we shouldn't close this issue until we figure out root cause
/cc @narimiran @Araq

[LemonBoy]

It was @Araq's last couple of commits that broke the CI and were directly pushed to devel

[narimiran]

I'm suspecting #9318 since it last touched tests/assign/tassign.nim but how come the CI was green for that PR?

After that file was modified, the tests were run for six times in that PR: all six times there were no problems with that file.

---

EDIT: Building the latest Nim devel and running tests/assign/tassign.nim now produces this unexpected result.

This part:

import typetraits
block toverload_asgn:
  type
    Concrete = object
      a, b: string

  proc `=`(d: var Concrete; src: Concrete) =
    shallowCopy(d.a, src.a)
    shallowCopy(d.b, src.b)
    echo "Concrete '='"

  var x, y: array[0..2, Concrete]
  var cA, cB: Concrete

  var cATup, cBTup: tuple[x: int, ha: Concrete]

  x = y
  cA = cB
  cATup = cBTup

  type
    GenericT[T] = object
      a, b: T

  proc `=`[T](d: var GenericT[T]; src: GenericT[T]) =
    shallowCopy(d.a, src.a)
    shallowCopy(d.b, src.b)
    echo "GenericT[T] '=' ", type(T).name

  var ag: GenericT[int]
  var bg: GenericT[int]

  ag = bg

  var xg, yg: array[0..2, GenericT[float]]
  var cAg, cBg: GenericT[string]

  var cATupg, cBTupg: tuple[x: int, ha: GenericT[int8]]

  xg = yg
  cAg = cBg
  cATupg = cBTupg

  var caSeqg, cbSeqg: seq[GenericT[bool]]
  newSeq(cbSeqg, 4)
  caSeqg = cbSeqg

  when false:
    type
      Foo = object
        case b: bool
        of false: xx: GenericT[int]
        of true: yy: bool

    var
      a, b: Foo
    a = b

used to produce:

Concrete '='
Concrete '='
Concrete '='
Concrete '='
Concrete '='
GenericT[T] '=' int
GenericT[T] '=' float
GenericT[T] '=' float
GenericT[T] '=' float
GenericT[T] '=' string
GenericT[T] '=' int8
GenericT[T] '=' bool
GenericT[T] '=' bool
GenericT[T] '=' bool
GenericT[T] '=' bool

but now there's no output.

[Araq]

but now there's no output.

Sorry, my move optimizations now optimize away all the assignments...

[timotheecour]

ref: 1475697 but unclear to me whether it's a proper fix or temporary workaround (in which case should we re-open?)

I'm disabling this section of the test... it's hard to fix, the compiler is way too smart/aggressive (gitter)