Mirror of forum.nim-lang.org

27 :: a silly question

• adrianveith (orginal) [2012-05-29T10:04:51+02:00] view original

  Hi,

  I am missing something obvious: How do I get a safe ref of a tuple or an object:

  type
    TObject = object
       data: int
  
  var o: TObject
  o.data = 123
  var po: ref TObject
  
  po = ??? # I want a ref to o
  Run

  I used cast[ref TObject] addr(o) - but this is unsafe. What is the safe way ? Or this this something I should not do ? Sorry to ask, but I did not find any information about this.

---

• Araq (orginal) [2012-05-29T12:41:42+02:00] view original

  This is something you should not do. A 'ref' is a traced pointer to an object that is guaranteed to be on the GC'ed heap. In your example you don't want the GC to trace this pointer at all; you want to create an alias. Use a 'ptr T' or a 'var T' instead.

---

• adrianveith (orginal) [2012-05-30T08:48:48+02:00] view original

  I am not sure if I understand this correct. Is the following safe to do ? Sorry I am confused and have the feeling, that I am missing something obvious here.

  
  type
  � TData = object
  � � s: string
  
  � TDataContainer = object
  � � data: array[0..10, TData]
  
  � PDataContainer = ref TDataContainer
  ��
  � PData = ptr TData
  
  var cont: PDataContainer
  new(cont)
  for i in 0..cont.data.high :
  � cont.data[i].s = $i
  
  # somewhere in the program
  var pEl: PData = addr(cont.data[2])
  
  # even more later I need new data
  new(cont)
  
  # can I access pEl safely here, or is the memory pEl points to collected ?
  echo pEl.s
  Run

---

• Araq (orginal) [2012-05-30T09:13:46+02:00] view original

  You can't access pEl safely here, the GC may have collected (btw GCs don't like interior pointers). addr is unsafe for this reason. What you can do, is to call GC_ref and GC_unref to keep the old cont alive. However, idiomatic Nimrod code uses indexes instead of interior pointers which don't have this problem.

---

• adrianveith (orginal) [2012-05-30T11:02:18+02:00] view original

  Hi Araq, thanks - I guess I have used the wrong language for too long ;-)

  BTW the results I get from my tests make me dizzy. I have increase in speed by factor 3 and more against the same code in Delphi or fpc and factor 1.5 - 2 against D (LDC not DMD). For example: my compression library is now on the same level as snappy (compiled with gcc) - which claims to be the fastest - but mine has a better compression ratio for my needs. I know its the C compiler backend (almost no difference if I use vcc or gcc, had no luck so far running clang on windows), but using C was never an option for me.

  I was on the jump to D, but now I am considering Nimrod for the next generation of our database server.

  And - It is amazing how clean the code looks, if you transfer it to Nimrod!

  Cheers, Adrian.

---

• Araq (orginal) [2012-05-31T00:26:24+02:00] view original

  ... but now I am considering Nimrod for the next generation of our database server.

  Excellent news. I hope Nimrod is up for the task. I'll try my best to quickly fix any occurring issues.

---