Mirror of forum.nim-lang.org

13306 :: "Objects that contain pointers that point to the same object"

• nrk (orginal) [2025-08-10T19:40:31+02:00] view original

  https://nim-lang.org/docs/destructors.html#swap states:

  Objects that contain pointers that point to the same object are not supported by Nim's model. Otherwise swapped objects would end up in an inconsistent state.

  Does this mean:

  1. "self-references are completely broken in ORC, do not use them under any circumstances"
  2. "self-references are fine, as long as you don't use swap"
  3. or something completely different that I'm misinterpreting?

  FWIW, this seems to work (valgrind reports no leaks at least):

  
  type A = ref object
    x: A
    b: string
  
  proc main() =
    for i in 0 ..< 100:
      var a = A()
      a.x = a
      a.b = "hello world"
      a.b &= "test"
      swap(a, a.x)
      swap(a[], a.x[])
      echo a.b
      echo a.x.b
    GC_fullCollect()
  
  main()
  Run

---

• Araq (orginal) [2025-08-10T19:57:47+02:00] view original

  It is option (3). What I tried to say is that this does not work:

  
  
  type
    SSO = object
      len: int
      p: ptr UncheckedArray[char] # points to payload or into the heap
      payload: array[24, char]
  Run

  The reason for that is that the SSO objects are not always made aware of object movements as the compiler assumes a bitcopy + wasMoved is sufficient.

---

• planetis (orginal) [2025-08-13T17:56:04+02:00] view original

  btw clang's libc++ does not use self-references either https://tc-imba.github.io/posts/cpp-sso/, instead they have something akin to this translation:

  template isLong(s): bool = (s.cap and strLongFlag) == strLongFlag

  template data(s: String): untyped =

  if isLong(s): s.p else: castptr UncheckedArray[char]

  Which of course tanks the performance...

---

• Araq (orginal) [2025-08-13T20:44:48+02:00] view original

  Which of course tanks the performance...

  Maybe, or maybe they decided on this design so that their strings become memcpy'able.

---