Bonsai

Bonsai is an attempt to provide a miniature and refined representation for the often cumbersome syntax trees and program models. This idea, of providing a smaller tree that is more or less the same thing, is where the name comes from.

This work started as part of an analysis tool that I am developing for my own research. I am interested in analysing ROS robotics applications, which are often written in C++. Since free C++ analysis tools are rather scarce, I tried to come up with my own, using the Python bindings of the clang compiler. At the moment of this writing, I am aware that these bindings are incomplete in terms of AST information they provide.

As this analysis tool developed, I realized that the C++ analysis features are independent of ROS or any other framework, and that this kind of tool might be useful for someone else, either as is, or as a starting point for something else.

Features

Bonsai provides an interface to represent, analyse or manipulate programs. The model it uses is abstract enough to serve as a basis for specific language implementations, although it focuses more on imperative/object-oriented languages for now.

What to expect from bonsai:

• classes for the different entities of a program (e.g. variables, functions, etc.);
• extended classes for specific programming languages (only C++ for now);
• parser implementations, able to take a file and produce a model (e.g. clang for C++);
• extensible interface to manipulate and query the resulting model (e.g. find calls for a function);
• a console script to use as a standalone application.

Installation

Here are some instructions to help you get bonsai. Bonsai has been tested with Linux Ubuntu and Python 2.7, but the platform should not make much of a difference. Dependencies are minimal, and depend on what you want to analyse.

Since at the moment there is only a single implementation for C++ using clang 3.8, you will need to install libclang and the clang.cindex bindings (pip install clang) to parse C++ files. Skip this if you want to use the library in any other way.

Method 1: Running Without Installation

Open a terminal, and move to a directory where you want to clone this repository.

git clone https://github.com/git-afsantos/bonsai.git

There is an executable script in the root of this repository to help you get started. It allows you to run bonsai without installing it. Make sure that your terminal is at the root of the repository.

cd bonsai
python bonsai-runner.py <args>

You can also run it with the executable package syntax.

python -m bonsai <args>

Method 2: Installing Bonsai on Your Machine

Bonsai is now available on PyPi. You can install it from source or from a wheel.

[sudo] pip install bonsai-code

The above command will install bonsai for you. Alternatively, download and extract its source, move to the project's root directory, and then execute the following.

python setup.py install

After installation, you should be able to run the command bonsai in your terminal from anywhere.

Examples

The cpp_example.py script at the root of this repository is a small example on how to parse a C++ file and then find all references to a variable a in that file. In it, you can see parser creation

parser = CppAstParser(workspace = "examples/cpp")

access to the global (top level, or root) scope of the program, and obtaining a pretty string representation of everything that goes in it

parser.global_scope.pretty_str()

getting a list of all references to variable a, starting the search from the top of the program (global scope)

CodeQuery(parser.global_scope).all_references.where_name("a").get()

and accessing diverse properties from the returned CodeReference objects, such as file line and column (cppobj.line, cppobj.column), the type of the object (cppobj.result), what is it a reference of (cppobj.reference, in this case a CodeVariable) and an attempt to interpret the program and resolve the reference to a concrete value (resolve_reference(cppobj)).

Do note that resolving expressions and references is still experimental, and more often that not will not be able to produce anything useful.

This is the pretty string output for a program that defines a class C and a couple of functions.

class C:
  C():
    [declaration]

  void m(int a):
    [declaration]

  int x_ = None

C():
  x_ = 0

void m(int a):
  a = (a + 2) * 3
  this.x_ = a

int main(int argc, char ** argv):
  C c = new C()
  c.m(42)
  C * c1 = new C()
  C * c2 = new C()
  new C()
  delete(c1)
  delete(c2)
  return 0

The pretty string representation, as seen, is a sort of pseudo-language, inspired in the Python syntax, even though the parsed program is originally in C++.

For more details on what you can get from the various program entities, check out the source for the abstract model and then the language-specific implementation of your choice.