Mirror of forum.nim-lang.org

890 :: Why is my type declaration overridden?

• stevenrbrandt (orginal) [2015-02-18T16:36:03+01:00] view original

  type

  Cplx = tuple[r:float, i:float]

  var i : Cplx = (0,1) echo(repr(i)) i.r = 3.4'f32

  This produces a compilation on the last line, because I'm trying to store an int in a float. However, I declared Cplx to be of type (float,float). Why is my type declaration of float overridden by int?

---

• Jehan (orginal) [2015-02-18T16:58:34+01:00] view original

  The issue is that while ints are implicitly convertible to float (int, int) is not implicitly convertible to (float, float). While mathematically the Gaussian integers are a subring of the complex numbers, you have to tell the compiler how that embedding works. After all, as far as the compiler knows, you may be trying to represent Z[sqrt(-5)] instead with pairs of ints. Something like the following should work:

  type
    Complex = tuple[re, im: float]
  
  converter toComplex(z: tuple[re, im: int]): Complex =
    (z.re.float, z.im.float)
  
  var i: Complex = (0, 1)
  echo(repr(i))
  Run

---

• stevenrbrandt (orginal) [2015-02-18T18:21:56+01:00] view original

  That's good, but what I don't understand is why the type is (int,int). I declared Cplx to be of type (float,float), and declared var i to be of type Cplx. In the program below, nim doesn't change the type of j to an int, even though I initialized it with one. Why does it change the type if I use a tuple?

  var j : float64 = 1

  j = 3.2

  echo($j)

---

• Araq (orginal) [2015-02-18T19:28:28+01:00] view original

  Because the convertible relation is not lifted over to tuple types. So just because int literal can be converted to float for you this doesn't extend to tuples.

---

• swerling (orginal) [2015-02-18T23:07:30+01:00] view original

  @stevenrbrandt, to put it another way, if you change...

  var i : Cplx = (0,1)
  Run

  ...to...

  var i : Cplx = (0.0,1.0)
  Run

  ...then it compiles.

  You were passing ints into a type that expects floats.

---

• OderWat (orginal) [2015-02-18T23:35:12+01:00] view original

  I think stevenrbrandt just misinterpreted the error message of the compiler and thought that the ints got assigned and the error arises because the "suddenly" int type gets assigned a float.

  type
    Cplx = tuple[r:float, i:float]
  var i: Cplx = (0,1)
  echo repr(i)
  i.r = 3.4'f32
  Run

  gives:

  
  typemad.nim(3, 14) Error: type mismatch: got (tuple[int, int]) but expected 'Cplx'
  
  Run

  stevenrbrandt The error is in line 3 there the compiler got your (0:int , 1:int) tuple and can't store it because he needs a Cplx aka (0:float, 1:float). The i never gets assigned the integer types. jehan showed you hot wo implement an automatic converter for int tuples to complex tuples.

  P.S.: As he shows the "repr(i)" line and tells us that the type was suddenly int. Was there a compiler bug which made that possible int the past?

---