Mirror of forum.nim-lang.org

13524 :: Partial inference on generics (in templates)

• acmt (orginal) [2025-11-27T13:27:19+01:00] view original

  
  # MWE: Input: some type and an array. Output: length of array
  template foo[T; N: static int](shape: static[array[N, int]]): untyped  = N
  const Z = foo[float, 2]([2,3]) # it works
  const Z = foo[float]([2,3]) # Error: missing generic parameter: N
  
  template foo2[N: static int](shape: static[array[N, int]]): untyped  = N
  const Z = foo2([2,3]) # it works
  
  template foo3[N: static int](T: typedesc, shape: static[array[N, int]]): untyped  = N
  const Z = foo3(float, [2,3]) # it works
  Run

  Can't some generic parameters be inferred from arguments when we manually define others (float in the example)? If I remove T and float, it works. I also works if I move T and float to the argument side. It looks like it's all (infer everything) or nothing (manually input everything), there is no "partial inference".

  Thinking twice, if partial inference would be provided, how could we use it? In an complicated case

  
  foo[T1; N; T2](shape: static[array[N, int]]): untyped = N
  Run

  Could it be foo[float; T2:4]([2,3])? Would we follow some ordering (e.g., inferable first)?

---