Inconsistent generics behavior

[ghost]

Consider this:

a.nim:

proc new*[T](_: typedesc[seq[T]], len = 0.Natural): seq[T] {.inline.} = new_seq[T](len)
discard seq[tuple[k: string, v: string]].new()

Yields no error, a.new is called instead of system.new. I think it is correct behavior because T: typedesc of system.new is more general and _: typedesc[seq[T]] of a.new[T] is more specific thus more specific variant gets selected.

a.nim:

import b
discard seq[tuple[k: string, v: string]].new()

b.nim:

proc new*[T](_: typedesc[seq[T]], len = 0.Natural): seq[T] {.inline.} = new_seq[T](len)

nim c b.nim yields Error: ambiguous call; both system.new(T: typedesc) and a.new(_: typedesc[seq[T]], len: Natural) match for: (typedesc[seq[tuple[k: string, v: string]]])

I think proc with more specific type should be preferred instea of generating ambigious call error. Practical usecase as seen from exampleis binding new() constructor to a type while that constructor has no parameters. So far this bug mandates writing seq[string].new(0) instead of seq[string].new() completely negating convenience of default parameter value.

[Araq]

The compiler also uses scope to disambiguate so if you are in the same module your alternative 'new' gets picked. This needs to be documented at least but I really don't like this behaviour. It's of course required for some template hacking to work. sigh

[ghost]

So what behavior would/should be in the end?