`hasCustomPragma` does not work on generic types

[etan-status]

Description

import std/macros

template p {.pragma.}

type
  X {.p.} = object

  Y[T] = object
    t: T

echo Y[X].T
echo Y[X].T.hasCustomPragma(p)

Nim Version

% nim -v Nim Compiler Version 1.6.20 [MacOSX: amd64] Compiled at 2024-06-08 Copyright (c) 2006-2023 by Andreas Rumpf

git hash: 19fdbfc173bfccb64cb64e0a963e69f52f71fc73 active boot switches: -d:release

Current Output

/Nim/lib/core/macros.nim(1550, 18) Error: cannot get child of node kind: nnkSym

Expected Output

true

Possible Solution

Use typeof(default(Y[X]).t).hasCustomPragma(p) instead. But this is not always feasible as the inner t may not be exported in all situations.

Additional Information

Possibly related to https://github.com/nim-lang/Nim/issues/23564

[metagn]

The type of Y[X].T is typedesc[X], but it also stores the info of the symbol T of the original unresolved generic param and this is what getTypeInst gives. This symbol is given to the type here and here (with linkTo).

From what I understand this shouldn't happen, a typedesc type should not have a bound symbol unless it is literally typedesc. But I don't know which makes the most sense to do:

1. Remove the linkTo lines in readTypeParameter and just return an nkType node; this might lose information or break stuff
2. To the and t.sym != nil condition in mapTypeToAstX, add and t.kind != tyTypeDesc; this seems too specific
3. To fix #23564 as well we can add and t.kind notin {tyTypeDesc, tyAlias} instead. That is, the only types where an "instantiation symbol" is ignored in favor of the implementation become typedescs or primitive alias types like type T = X. This has the caveat of defining new behavior for getTypeInst that might be incompatible with old behavior. We could also combine the first option with and t.kind != tyAlias instead but this seems more independent of compiler internals.

Edit: Actually I think hasCustomPragma should deal with tyAlias, getTypeInst shouldn't skip it.