Mirror of forum.nim-lang.org

2032 :: tables, mget() and mgetOrPut()

• Stefan_Salewski (orginal) [2016-02-11T00:08:21+01:00] view original

  I just tried to use mgetOrPut() of tables module. I was a bit surprised that modifying the returned result was possible as said in the docs, but the table itself was not modified, So I tested with mget(), because that is similar but simpler:

  import tables
  
  type
    
    Y = tuple
      i: int
  
  proc main() =
    var
      t = newTable[string, Y]()
      y: Y
    y.i = 1
    t["one"] = y
    echo t["one"].i
    t.mget("one").i = 3
    echo t["one"].i
    var x = t.mget("one")
    x.i = 5
    echo t["one"].i # one may expect 5, but it is 3
  
  main()
  Run

  1
  3
  3

  From the behaviour of mitems for arrays iterators one may expect 5 as last result. After some thinking I assume that this is not a bug but intended -- modifying the table entry itself would only work with hidden pointers, which may be dangerous.

  I wonder -- mget() is deprecated, mgetOrPut() is not deprecated?

  [EDIT]

  OK, I am not the first who is surprised:

  http://forum.nim-lang.org/t/1477#9246

  I still wonder what with this behaviour the use cses for mgetOrPut() are...

---

• cblake (orginal) [2016-02-11T03:18:52+01:00] view original

  Well, any use case where the value type of the table has some simple update protocol from a default value that is the same as for an existing value (and you want to avoid repeating the hash probe search). E.g., inc(tab.mgetOrPut(key, 0)) would work to histogram keys or tab.mgetOrPut(key, @[]).add(itm) would attach lists of items associated with the key, etc. { Not 100% sure about that @[] syntax, but hopefully you get the idea. }

---