regression: {.pure.} pragma fails to hide enum

[geekrelief]

Compiling with 979148e breaks compilation of https://github.com/pragmagic/godot-nim

Current Output

C:\godot\gdnim\deps\godot\core\poolarrays.nim(128, 16) template/generic instantiation of definePoolArray from here
C:\godot\gdnim\deps\godot\core\poolarrays.nim(49, 26) Error: type expected, but got symbol 'Array' of kind 'enumField'

Expected Output

This compiled without error before.

Additional Information

979148e breaks.
Two modules are involved. https://github.com/pragmagic/godot-nim/blob/master/godot/core/poolarrays.nim and https://github.com/pragmagic/godot-nim/blob/master/godot/core/arrays.nim

poolarrays has a template that fails because Array is not found.

template definePoolArray(T, GodotT, DataT, fieldName, newProcName, initProcName;
                         noData = false) =

  proc newProcName*(arr: Array): T {.inline.} =
    #new(result, poolArrayFinalizer)
    new(result)
    initProcName(result.fieldName, arr.godotArray[])

poolarrays imports arrays online 125.

import arrays

definePoolArray(PoolByteArray, GodotPoolByteArray, uint8,
                godotPoolByteArray, newPoolByteArray,
                initGodotPoolByteArray)

[timotheecour]

@geekrelief can you reduce this to an example with just 2 (say mod1, mod2) modules and no imports besides mod1 importing mod2 or vice versa?
this would be needed anyways for regression tests

[geekrelief]

Yes, I'll make this my next priority. Thanks!

[geekrelief]

It's actually an issue with the {.pure.} pragma

#v.nim
type
  V* {.pure.} = enum
    A

#test.nim
import v

template defA() =
  proc onA(a:A) =
    echo A.val

type 
  A* = object
    val: int

defA()

nim c test produces

Error: type expected, but got symbol 'A' of kind 'enumField'

V.A should be hidden with the pure pragma per the manual. https://nim-lang.org/docs/manual.html#pragmas-pure-pragma

[metagn]

V.A should be hidden with the pure pragma per the manual.

There was discussion about this in #8066. For now {.pure.} not providing a full restriction is intended. Your code example should be fixed by changing template defA() = to template defA(): untyped =, or adding mixin A below. If you still think the issue is "{.pure.} doesn't work" then #12893 exists and this would be a duplicate, but I don't think that's the issue here and this is a more specific result (perhaps even intended).

[geekrelief]

Those issues you referenced are quite old. This was working until pretty recently.

I tried both of your suggestions. Adding untyped produces the same error. Adding mixin A produces this:

Error: type expected, but got: <<0th child missing for nkOpenSymChoice >>

[metagn]

Ok, my point is that something did change, but you shouldn't change the issue title to "{.pure.} doesn't work", because that's an old issue and not the problem here. The incremental compilation commit broke something with symbols as the mixin error shows.

[geekrelief]

I was trying to be more helpful with the title. Would you prefer I change it to something else?

[ghost]

#15935 also breaks https://github.com/ftsf/nico - ftsf/nico#49 (not sure if this is the same bug):

# a.nim
import sdl except Scancode
import other

let a = SCANCODE_LEFT

# other.nim
type
  Scancode* = enum
    SCANCODE_LEFT = 80

# sdl.nim
type
  Scancode* = enum
    SCANCODE_LEFT = 80

The error is:

Error: ambiguous identifier: 'SCANCODE_LEFT' -- use one of the following:
 Scancode.SCANCODE_LEFT: Scancode
 Scancode.SCANCODE_LEFT: Scancode

[jyapayne]

@geekrelief this can also be worked around by declaring the template underneath the type declaration like so:

#v.nim
type
  V* {.pure.} = enum
    A

#test.nim
import v

type 
  A* = object
    val: int

template defA() =
  proc onA(a:A) =
    echo A.val

defA()

[geekrelief]

@jyapayne Thanks for the suggestion. I saw the commit for godot-nim, and it helps for that case. But this issue popped up again in another place when I implemented signal declarations and got a name clash with VariantType.Object and type Object defined in godotapi/godottypes.nim. For now, I've renamed the enum identifiers following https://nim-lang.org/docs/nep1.html naming conventions.

[jyapayne]

Ah, I see. You're right that there's another clash down the line.

@Araq do you have any advice on where this could be fixed? I'm interested in attempting to fix the issue

[jyapayne]

Or @timotheecour :)

[Araq]

We should do this nim-lang/RFCs#373 (comment)

In general, the compiler must produce sym choices that are then collapsed taking into account more context. For example in param: Type we know Type is an skType. And then we should ignore the "pure" enum annotation altogether, there should only be enums in Nim, not "pure enums". But I suspect this to be highly controversial...

[ringabout]

Hello @Yardanico, the pure enum version of your example works in devel

# a.nim
import sdl except Scancode
import other

let a = SCANCODE_LEFT

# other.nim
type
  Scancode* {.pure.} = enum
    SCANCODE_LEFT = 80

# sdl.nim
type
  Scancode* {.pure.} = enum
    SCANCODE_LEFT = 80

see also: #19692