Mirror of forum.nim-lang.org

796 :: Weird compiler error

• andrea (orginal) [2015-01-26T17:06:01+01:00] view original

  I am trying to make a program where I will need a few numeric typeclasses, and in particular I will need rational arithmetic. I am encountering a weird behaviour when compiling, and I have been able to obtain a minimal example like this:

  type Quotient[A] = object
    a, b: A
  
  template zero(x: int): int = 0
  template zero[A](x: Quotient[A]): Quotient[A] = Quotient[A](a: zero(x.a), b: id(x.a))
  
  template id(x: int): int = 1
  template id[A](x: Quotient[A]): Quotient[A] = Quotient[A](a: id(x.a), b: id(x.a))
  
  proc quot[A](x: A): Quotient[A] = Quotient[A](a: x, b: id(x))
  
  echo zero(5.quot)
  Run

  The compilation fails with fail.nim(12, 9) Error: undeclared identifier: 'A'. I think this has something to do with the interaction between templates and generics, but I am not experienced enough to tell what is going wrong

---

• def (orginal) [2015-01-26T18:09:48+01:00] view original

  I don't think generics in templates work. In this case (inline) procs would be fine?:

  type Quotient[A] = object
    a, b: A
  
  template id(x: int): int = 1
  proc id[A](x: Quotient[A]): Quotient[A] {.inline.} =
    Quotient[A](a: id(x.a), b: id(x.a))
  
  template zero(x: int): int = 0
  proc zero[A](x: Quotient[A]): Quotient[A] {.inline.} =
    Quotient[A](a: zero(x.a), b: id(x.a))
  
  proc quot[A](x: A): Quotient[A] = Quotient[A](a: x, b: id(x))
  
  echo zero(5.quot)
  Run

---

• andrea (orginal) [2015-01-26T18:13:22+01:00] view original

  Yes, I think inline procs are the way to go. By the way, does the compiler make automatic inlining? I would venture to guess yes for one-expression procs, but I am not sure

---

• def (orginal) [2015-01-26T19:31:10+01:00] view original

  Even when you write {.inline.} the Nim compiler does not actually inline anything. It just tells the C compiler to inline the function with the inline keyword (or __inline for some compilers). Whether the C compiler adheres to that is its own decision. For example GCC doesn't inline with -O0 even when inline is set for the function. On the other hand the C compiler can inline automatically when optimizations are activated.

---

• LeuGim (orginal) [2015-01-26T19:49:03+01:00] view original

  Andrea, I wrote a rational arithmetic module some time ago - https://github.com/Leu-Gim/Nim-modules/blob/master/exact.nim, maybe it can be helpful for you. I don't remember now how much it's completed/tested, but it works for a simple test.

---

• def (orginal) [2015-01-26T22:11:28+01:00] view original

  Andrea, I wrote a rational arithmetic module some time ago

  That's (at least) the third one. Rationals should really go into the standard libray before even more people create their own:

  https://github.com/Araq/Nim/pull/1840

  https://github.com/Araq/Nim/pull/1825

---