converter fails to work with multiple generic types

[jxy]

With the following code

type
  G[N:static[int]] = object
    v: int
  F[N:static[int]] = object
    v: int
static:
  echo(G[1] is G[2])
  echo(G[1] is F[1])
converter G2int[N:static[int]](x:G[N]):int = x.v
converter F2int[N:static[int]](x:F[N]):int = x.v
proc p(x,y:int) = echo "p ",x," ",y
var
  g1 = G[1](v:1)
  g2 = G[2](v:20)
  f1 = F[1](v:1000)
  f2 = F[2](v:200)
p(g1,g2)    # Error: type mismatch: got (G[1], G[2])
p(f1,f2)    # Error: type mismatch: got (F[1], F[2])
p(g1,g1)    # compiles
p(f1,f1)    # compiles
p(g1,f1)    # compiles
p(f1,g1)    # compiles
p(g1,f2)    # compiles
p(f1,g2)    # compiles

the first two calls, p(g1,g2) and p(f1,f2), fail with type mismatch complained by the compiler. However, the other calls compile and run fine.

[jxy]

The compiler believes that G[1] is G[2], so after changing the converter definition, the following code actually works

type
  G[N:static[int]] = object
    v: int
  F[N:static[int]] = object
    v: int
static:
  echo(G[1] is G[2])
  echo(G[1] is F[1])
converter G2int(x:G[0]):int = x.v
converter F2int(x:F[0]):int = x.v
proc p(x,y:int) = echo "p ",x," ",y
var
  g1 = G[1](v:1)
  g2 = G[2](v:20)
  f1 = F[1](v:1000)
  f2 = F[2](v:200)
p(g1,g2)
p(f1,f2)
p(g1,g1)
p(f1,f1)
p(g1,f1)
p(f1,g1)
p(g1,f2)
p(f1,g2)

Somehow G[0] and F[0], or defined with any int, make the compiler correctly call the converter. Bug or feature?

[jxy]

Now with the devel branch, Nim no long thinks G[1] is G[2] nor G[1] is F[1]. Both of the examples above do not compile.

[n0bra1n3r]

I was about to file a bug but saw this issue. I think I have the same issue, but with normal generic parameters and not static ones:

type Obj[T] = object
  b: T

converter test1[T](a: Obj[T]): T = a.b

proc doStuff(a: int, b: float) = discard

doStuff(Obj[int](b: 1), Obj[float](b: 1.2)) # Error: type mismatch: got <Obj[system.int], Obj[system.float]>

As far as I'm aware this has been an issue forever, just hit it again today. I try very hard not to use converters much because of the bugs and magic, but when I do use them they provide some very nifty functionality.

There was a patch awhile ago that seemed related but didn't fix this specific issue: #9698

[metagn]

The compiler reuses the inferred generic parameters for a converter when matching another argument in the same overload match here, likely other open issues are because of this. These bindings should probably be reverted/isolated and matched again when instantiating the generic converter.