How to parse multi-dimensional arrays with python libclang bindings

I'm having a difficult time with what, in my mind, should be a fairly simple task: Using the python bindings to libclang I want to get the dimensions for a multi-dimensional array field of a POD C++ structure. When I traverse the AST, I can drill down the cursor containing the array field declaration, and I can even see that it has two child nodes containing, what I assume, are the sizes of each of its dimensions... but I've had no luck accessing the size as a value. Below is a minimal example of my attempt at accessing the size:

test.hpp

#pragma once

struct my_struct {
    int mdarr[10][20];
};

test.py

import clang.cindex as cl

def process(c):
    if c.kind in [cl.CursorKind.STRUCT_DECL, cl.CursorKind.CLASS_DECL]:
        print("Found struct: ", c.spelling)
        for field in c.type.get_fields():
           print("Found field: ", field.spelling) 
           # Returns size of first dimension but not the second dimension
           print("Array size: ", field.type.get_array_size())
           for child in field.get_children():
                # Prints an empty string
                print("Found child: ", child.spelling)
                # How do I extract the value from the `INTEGER_LITERAL`?
                print("Child cursor kind: ", child.kind)

        return

    for child in c.get_children():
        process(child)

idx = cl.Index.create()
tu = idx.parse("test.hpp")
process(tu.cursor)

Output

Found struct:  my_struct
Found field:  mdarr
Array size:  10
Found child:
Child cursor kind:  CursorKind.INTEGER_LITERAL
Found child:
Child cursor kind:  CursorKind.INTEGER_LITERAL

Is there an easy way to extract each dimension's size using libclang?

What is a multidimensional array?

C and C++ do not have multidimensional arrays per se. Instead, an array type can have another array type as its element type. In the declaration:

int mdarr[10][20];

this is parsed "inside out" (like all C/C++ declarators) as:

    mdarr
    ^^^^^               mdarr is ...
    mdarr[10]
         ^^^^           ... an array of 10 elements, each element being ...
    mdarr[10][20]
             ^^^^       ... an array of 20 elements, each element being ...
int mdarr[10][20]
^^^                     ... an integer.

Getting the dimensions in libclang

Clang's representation follows this structure, representing the type as an array(10) of an array(20) of int.

The type of a Python libclang Cursor is obtained with Cursor.type.

From a Type one may invoke:

• Type.kind to determine what kind of type (primitive, array, etc.) this is,

• Type.get_array_element_type to get the element type for an array, and

• Type.get_array_size to get the array size, among other methods.

Adjusting the code to print all of the dimensions

The following is a modified (added between BEGIN ADDED and END ADDED) version of the code in the question that prints all of the array dimensions:

import clang.cindex as cl

def process(c):
    if c.kind in [cl.CursorKind.STRUCT_DECL, cl.CursorKind.CLASS_DECL]:
        print("Found struct: ", c.spelling)
        for field in c.type.get_fields():
           print("Found field: ", field.spelling) 
           # Returns size of first dimension but not the second dimension
           print("Array size: ", field.type.get_array_size())

           # BEGIN ADDED
           t = field.type
           while t.kind == cl.TypeKind.CONSTANTARRAY:
               print("  ADDED: Array size:", t.get_array_size())
               t = t.get_array_element_type()
               print("  ADDED: Element type:", t.spelling)
           # END ADDED

           for child in field.get_children():
                # Prints an empty string
                print("Found child: ", child.spelling)
                # How do I extract the value from the `INTEGER_LITERAL`?
                print("Child cursor kind: ", child.kind)

        return

    for child in c.get_children():
        process(child)

idx = cl.Index.create()
tu = idx.parse("test.hpp")
process(tu.cursor)

On the original input, this script prints:

Found struct:  my_struct
Found field:  mdarr
Array size:  10
  ADDED: Array size: 10
  ADDED: Element type: int[20]
  ADDED: Array size: 20
  ADDED: Element type: int
Found child:  
Child cursor kind:  CursorKind.INTEGER_LITERAL
Found child:  
Child cursor kind:  CursorKind.INTEGER_LITERAL

Getting the value of an INTEGER_LITERAL

Relatedly, you ask (in a comment in the question code) how to get the value of an INTEGER_LITERAL node. To do that, retrieve the tokens that make up the literal, get the first one, and get its "spelling".

See the Q+A How to retrieve function call argument values using libclang for more details about that.