Mirror of forum.nim-lang.org

13529 :: Help with macro error

• treeRepr (orginal) [2025-11-30T16:40:24+01:00] view original

  I am learning macro system and trying to solve this issue (nim 2.2.6)

  
  macro class(name: untyped, body: untyped): untyped =
      result = quote("@") do:
          type `@name` = ref object of system.RootObj
              `@body`
  
  class TestClass:
      b = 10
  Run

  This produces error - Error: ':' or '=' expected, but got 'class'

  if i change indentation like this

  
  macro class(name: untyped, body: untyped): untyped =
      result = quote("@") do:
          type `@name` = ref object of system.RootObj
          `@body`
  
  class TestClass:
      b = 10
  Run

  I get - Error: undeclared identifier: 'b'

  What is happening? (I tried searching but the documentation for quote do, how body works is scarce)

---

• PMunch (orginal) [2025-12-01T10:48:43+01:00] view original

  Macros can be a bit confusing sometimes. Part of this is caused by similar looking code generating quite different abstract syntax trees (AST, the input and output from a macro).

  As I mention in my article about metaprogramming the best way to create a macro is to first write what you want your output to be, then use dumpTree to see what AST you need to create. For your code we expect it to generate:

  
  type
    TestClass = ref object of RootObj # The system part isn't necessary
      b = 10
  Run

  Looking at the output from dumpTree, it's long and a bit confusing, but don't worry:

  
  StmtList
    TypeSection
      TypeDef
        Ident "TestClass"
        Empty
        RefTy
          ObjectTy
            Empty
            OfInherit
              DotExpr
                Ident "system"
                Ident "RootObj"
            RecList
              IdentDefs
                Ident "b"
                Empty
                IntLit 10
  Run

  We see that everything is wrapped in a StmtList but this is just an implementation detail of dumpTree. The interesting bits are the TypeSection, particularly where the Ident "TestClass" and IdentDefs(Ident "b", Empty, IntLit 10) appear. If we now take a look at your inputs:

  name:

  
  Ident "TestClass"
  Run

  body:

  
  StmtList
    Asgn
      Ident "b"
      IntLit 10
  Run

  We can see that the name input is Ident "TestClass" which is what we need for the output TypeSection, but the body is a StmtList(Asgn(Ident "b", IntLit 10)) which isn't what we need. In order to fix this you need to write some code that traverses the body, then generate a RecList of IdentDefs instead and patch them into the tree generated by quote do.

  It's a bit confusing, and I wish the macros module had more tools to reshape trees like this, but at least for now it's a bit of a manual process.

---

• dlesnoff (orginal) [2025-12-01T11:55:52+01:00] view original

  I have been tinkering with the AST to graphViz code and I am able to produce with a simple code:

  
  astToImage("type_forum_question", "Type with Root Obj", "jpg", "pictures"):
    type
       TestClass = ref object of RootObj
          b = 10
  Run

  the following image (sorry for the double StmtList and unuseful nnks):

  

  Distinguishing leaves from structural nodes in this kind of output may help @treeRepr understanding the AST better than only from the raw output of dumpTree.

  Note that a class macro example doing this tree traversal is given in this macro tutorial: https://nim-by-example.github.io/macros/

  The difference with your macro is it does not force to inherit of RootObj but requires the user to specify the inherited class themself.

---

• PMunch (orginal) [2025-12-01T15:24:41+01:00] view original

  Ooh, that's neat! I have tried something similar in the past but the generated trees for large chunks of code quickly gets complicated. I wonder if we could somehow add a label with the code generating each sub part, would be interesting. The positional nature of NimNodes isn't great for discoverability or robustness either. Like the nnkEmpty nodes in this tree. What do they mean?! The context is the only way to figure it out, which isn't great.

---