CODEWITHSUNDEEP
pythonc++parsingclangabstract-syntax-tree
slaughter98
slaughter98

Reputation: 1883

How to get class method definitions using clang python bindings?

Given the following C++ file:

class Foo
{
public:
  Foo();

  void bar(int input);

  void another(int input, double & output);
};

void
Foo::bar(int input)
{
  input += 1;
}

void
Foo::another(int input, double & output)
{
  input += 1;
  output = input * 1.2345;
}

How can I utilize clang python bindings to extract the definitions of the two methods. I am able to get the class declarations using the python script below, but I can't seem the figure out how to extract the complete methods. For example, I want this information:

void
Foo::another(int input, double & output)
{
  input += 1;
  output = input * 1.2345;
}

Python script:

#!/usr/bin/env python
import clang.cindex
clang.cindex.Config.set_library_path('/opt/moose/llvm-3.7.0/lib')

def getCursors(cursor, output, kind):
    """
    Recursively extract all the cursors of the given kind.
    """
    for c in cursor.get_children():
        if c.kind == kind:
            output.append(c)
        getCursors(c, output, kind)

if __name__ == '__main__':

    # Parse the test file
    index = clang.cindex.Index.create()
    tu = index.parse('method.C', ['-x', 'c++'])

    # Extract the parsers
    output = []
    getCursors(tu.cursor, output, clang.cindex.CursorKind.CXX_METHOD)

    # Print the method declarations (How to I get the definitions?)
    for c in output:
        defn = c.get_definition() # Gives nothing
        print defn 
        print c.extent.start.file, c.extent.start.line, c.extent.end.line # Gives decleration

FOLLOW UP:

The following function was suggested as a solution, but it doesn't work for clang 3.7. I am unable to update to 3.9 until it is released and I need to support the latest two versions of clang (3.7 and 3.8). If you add a print statement the results indicate that a definition is not located.

def method_definitions(cursor):
    for i in cursor.walk_preorder():
        print i.kind, i.is_definition() # Added this
        if i.kind != CursorKind.CXX_METHOD:
            continue
        if not i.is_definition():
            continue
        yield i

Running the method yields the following, anyone know what has changed between clang 3.7 and 3.9?

CursorKind.TRANSLATION_UNIT False
CursorKind.CLASS_DECL True
CursorKind.CXX_ACCESS_SPEC_DECL True
CursorKind.CONSTRUCTOR False
CursorKind.CXX_METHOD False
CursorKind.PARM_DECL True
CursorKind.CXX_METHOD False
CursorKind.PARM_DECL True
CursorKind.PARM_DECL True

Upvotes: 5

Views: 4151

Answers (1)

Andrew Walker
Andrew Walker

Reputation: 42490

You're pretty close - the trick is to use the cursor offsets rather than line / columns, and that AFAIK, libclang doesn't expose the bytes of the source, so you need to read the text out yourself.

The following is tested on:

  • Clang provided by Xcode 7.3(7D175) - OSX Apple LLVM version 7.3.0 (clang-703.0.29) and pip installed bindings from pypi (pip install 'clang==3.7')
  • Ubuntu clang version 3.7.1-svn253742-1~exp1 (branches/release_37) (based on LLVM 3.7.1)
  • Ubuntu clang version 3.6.2-svn240577-1~exp1 (branches/release_36) (based on LLVM 3.6.2)
  • clang version 3.9.0-svn267343-1~exp1 (trunk)

The only caveats are that for more complex cases with macros / headers YMMV.

import clang.cindex
from clang.cindex import *

def method_definitions(cursor):
    for i in cursor.walk_preorder():
        if i.kind != CursorKind.CXX_METHOD:
            continue
        if not i.is_definition():
            continue
        yield i

def extract_definition(cursor):
    filename = cursor.location.file.name
    with open(filename, 'r') as fh:
        contents = fh.read()
    return contents[cursor.extent.start.offset: cursor.extent.end.offset]

idx = Index.create()
tu = idx.parse('method.C', ['-x', 'c++'])
defns = method_definitions(tu.cursor)
for defn in defns:
    print extract_definition(defn)

Which gives:

void
Foo::bar(int input)
{
      input += 1;
}
void
Foo::another(int input, double & output)
{
      input += 1;
        output = input * 1.2345;
}

Upvotes: 1

Related Questions