retain postfix node for typed variable/routine AST; save field syms

[metagn]

fixes #13828, refs #23101

Postfix nodes in var/let/const/routine nodes are now retained in typed AST, previously they would just become either an nkSym node or an nkPragmaExpr node containing the nkSym node. This was done for type sections in #23101 which also handled doc generation of postfix nodes. The newly generated postfix nodes get transformed back into just the symbol nodes in transf.

On top of this, object fields now store their symbol node in the AST. The old output of:

import macros

type Foo = object
  a: int

static:
  echo treeRepr getImpl bindSym"Foo"

was:

TypeDef
  Sym "Foo"
  Empty
  ObjectTy
    Empty
    Empty
    RecList
      IdentDefs
        Ident "a"
        Sym "int"
        Empty

Now the Ident "a" becomes Sym "a", storing the symbol of the field.

[metagn]

It seems like no packages in the CI were affected by the AST changes in this PR, but I added it to the changelog anyway, at least for the type case which did require package updates.

[Araq]

Sorry for the late reply. This is a step in the right direction but I dream of a more radical change: Introduce an nkIdentDecl which is always the tuple (name, export, pragma, type, value).

[metagn]

That would be nice, I don't know how it would interact with types and procs though, they don't use nkIdentDefs. At the very least we can combine nkPragmaExpr and nkPostfix.

We would also have to either gate it into an experimental switch or have it for internal use only and always produce the old AST to pass to macros and then turn it back after receiving it (presumably it's present in both untyped and typed AST).

[Araq]

... or have it for internal use only and always produce the old AST to pass to macros and then turn it back after receiving it (presumably it's present in both untyped and typed AST).

That's the solution I have in mind. As for "not used in parameters", maybe it should be nkLetDecl(5 fields), nkVarDecl(5 fields), nkParamDecl(5 fields), etc.

[metagn]

Rebased and added test for #13828

[metagn]

Rebased to fix merge conflicts but the diff for ast2nir is gone since it was removed, just a heads up in case it was moved somewhere else. Maybe we can have this before 2.2?

[Araq]

It was not moved to anywhere and the whole NIR thing got replaced by NIF which lives in its own repository.