Redhawk

Redhawk is a code navigation system built on the idea of a language agnostic parse tree. It currently supports C & Python.

Code navigation systems are few and far. They are either too tied to a language, or are very heuristic in nature --- using regex based parsers. Redhawk attempts to acheive the best of both worlds. It uses standard, robust, parsers each of the languages, and converts the resulting AST to a language agnostic AST, or LAST.

The resulting LAST can be queried by using either Selectors (similar to JQuery), or an xpath like syntax. A Typical use of Redhawk is as shown below:

$ redhawk query '*/DefineFunction' file1.py file2.c

Redhawk is currently under heavy development. The code can be found on github.

Redhawk currently requires python 2.6 or 2.7.

Project News

An introductory set of videos, have been uploaded to Youtube.

A Vim plugin released in version 1.1.5, for query, and replace (using an editable quickfix list).

From Version 1.1.2 onwards, Redhawk supports parallel querying using the parallel-python (pp) module. This speeds up Redhawk's querying on large codebases. Querying for closures anywhere in Django (~2200 files) can now be done in ~20 seconds on a celeron netbook.

Project Objectives

(or what's coming up)

1. Allow users to effectively find and thereby navigate code in an editor-independent manner.

2. Better documentation for API usage, and a long list of examples, with examples scripts using the Selector API.

3. Allow cross-language analysis in the future, thereby benefitting projects in multiple languages.

4. Expose the LAST in a simple manner via the Redhawk API for other tools. These tools could involve indenting code, suggesting completions, or static analysis.

5. Eventually allow editing of the LAST, and thereby powerful refactoring.

Dependencies

Runtime Dependencies:

• pycparser is required to parse C code into ASTs. This in-turn depends on Python-PLY (python-ply on debian-ubuntu).

Optional but highly recommended Dependencies:

• pp - Parallel Python is required for running queries in parallel. This speeds up queries by more than 2x. This is highly recommended if you are going to query large projects. The whole of Django can be queried in less than 20 seconds, by using parallel python (passing -p to the query command).
• Python Graphviz is required for generating pretty AST graphs. This package is an optional dependency, but highly recommended. This package goes by the name python-pygraphviz on Ubuntu, and depends on graphviz, and dot. (Pip seems to have a hard time install pygraphviz. Either easy_install or installing from your distribution's package manager should work).

Development (Compile-time) Dependencies:

• Python YAML is required for generating the AST classes in node.py form a simple configuration file. This goes by the name python-yaml on debian/ubuntu.
• nosetests is required for running the test suite.

Development Cycle:

• Use bin/start_simple_bash_with_redhawk_in_pythonpath.sh
• Run nosetests from redhawk/ as root. Ensure tests pass.
• Make awesome changes!
• Submit a pull request to the project on github.

Installing

pip is the recommended tool to install Redhawk. It goes by python-pip on debian/ubuntu and pip on the Python Package Index. The command:

$ sudo pip install redhawk

should install redhawk, along with its dependency - pycparser.

It is however recommended that you install the other packages also:

$ sudo easy_install pygraphviz
$ sudo pip install nose 'PyYAML>=3.09' 'nose>=0.11'

or by using your distribution's package manager. On Ubuntu/Debian (Ubuntu Lucid seems to have new enough packages):

$ sudo aptitude install python-pygraphviz python-yaml python-nose

Notes

1. Run _build_tables.py in the pycparser directory, to pre-generate the lex and yacc tables. This will enable quicker parsing of C files. If pycparser was installed for all users, then

• Root priviliges may be required to run _build_tables.py
• Permissions for the resulting lextab.py and yacctab.py must be changed to allow all users to read (755).

You can find more about this here.

Using Redhawk

Redhawk can be used from either the redhawk executable, or via the redhawk API.

1. Using the redhawk program.
2. Using the Redhawk API via import redhawk

Using the redhawk executable

The redhawk program supports eight commands:

Command	Purpose
add	Add files to an AST index.
init	Create an EMPTY AST index.
listfiles	List all the files in the AST index.
prompt	Drop into a python prompt with helpful functions for exploring the parse tree.
query	Query for a pattern in a list of files, or in the index.
remove	Remove files from the AST index.
show	Show (visualize) a file either as text, or as an image.
where	Print the location of the current redhawk index (if there is one).

The simplest way to run redhawk is to simply use a query command on a file (or directory). The query command as described above takes an xpath-like query, and a list of files (or directories), and searches for matches.

In the case that the set of files is large and is to be repeatedly queried, a redhawk Language Agnostic Tree (LAST) database can be created using the redhawk init command. Files in the project can be added to the database using the redhawk add command.

The show command helps visualise the internal LAST structure used. The command:

$ redhawk show file.c

will show the LAST of file.c in a lisp/scheme like (sexp) syntax. A more descriptive helpful visualisation can be obtained using the -i (or -e) flags, which show graphs (generated using graphviz using the python-graphviz module). This requires the pygraphviz module, an optional though recommended, dependency. The command:

$ redhawk show file.c -i

shows a graph using the default image python libraries.

The prompt command drops you into a prompt for exploring and querying the LAST. This enables the use of selectors, a very powerful method for finding what you want. For more information on selectors, see:

$ pydoc redhawk.common.selector

for detailed documentation.

Introduction to the Query Language

The query command supports an XPATH-like language for querying. We describe examples below. In querying for a particular construct, the name of that Node in the LAST has to be known. (Thorough documentation about this is coming up. For now, one can refer to the node and types yaml configuration files on github.)¹

For the examples below, we shall use the counter.py file. It is to be noted that the same queries will work with other languages also (only C is supported for now).:

1  def CounterClosure(init=0):
2    value = [init]
3    def Inc():
4      value[0] += 1
5      return value[0]
6    return Inc
7  
8  class CounterClass:
9    def __init__(self, init=0):

10 self.value = init 11 12 def Bump(self): 13 self.value += 1 14 return self.value 15 16 def CounterIter(init = 0): 17 while True: 18 init += 1 19 yield init 20 21 if __name__ == '__main__': 22 c1 = CounterClosure() 23 c2 = CounterClass() 24 c3 = CounterIter() 25 assert(c1() == c2.Bump() == c3.next()) 26 assert(c1() == c2.Bump() == c3.next()) 27 assert(c1() == c2.Bump() == c3.next()) 28

Try redhawk show on the above file, to get a feel of its structure. You can view the graphviz generated graph at imgur.

Example 1: Let us find all functions at the module level in `counter.py`:

$ redhawk query 'DefineFunction' counter.py

This gives us:

counter.py:16:def CounterIter(init = 0):
counter.py:1:def CounterClosure(init=0):

NOTE:

1. The results are not necessarily in a sorted order, with respect to line number. This does not hamper the use of Redhawk for searching and navigation. (The results will always be guaranteed to be sorted with respect to the files). On the plus side, this makes Redhawk a little bit faster. If order is required, a simple invocation of the unix sort program should fix this.

2. The above query would work on a C program as well. Running the same query on stats.c gives us:

   stats.c:17:float Variance(float *p, int len)
   stats.c:5:float Mean(float *p, int len)
   stats.c:34:int main()

Example 2: Let us find all functions one level below the module level in `counter.py`:

$ redhawk query '*/DefineFunction' counter.py

This gives us:

counter.py:9:def __init__(self, init=0):
counter.py:3:def Inc():
counter.py:12:def Bump(self):

Example 3: Let us find all functions anywhere in the program.:

$ redhawk query '**/DefineFunction' counter.py

This gives us:

counter.py:9:def __init__(self, init=0):
counter.py:16:def CounterIter(init = 0):
counter.py:3:def Inc():
counter.py:1:def CounterClosure(init=0):
counter.py:12:def Bump(self):

Example 4: Suppose we wanted to find all closures in the file. We could do this via:

$ redhawk query '**/DefineFunction/**/DefineFunction' counter.py

This gives us:

counter.py:3:def Inc():

Example 5: Let us find all functions whose name starts with 'Counter'. Looking at the node yaml configuration tells us that DefineFunction has an argument called name. Now we simply need to test whether the first 7 letters of the name are "Counter":

$ redhawk query '**/DefineFunction@{n.name[:7] == "Counter"}' counter.py

This gives us:

counter.py:16:def CounterIter(init = 0): counter.py:1:def CounterClosure(init=0):

The @{..} represents a python lambda function, with the default variable n. Thus, it is another way of providing arbitrary functions to match with.²

To remind the reader that all these queries are langauge agnostic, running the above command, but instead search for all functions that have the letter e in the them, in the stats.c file.:

$ redhawk query '**/DefineFunction@{n.name.find("e") != -1}' stats.c

gives us:

stats.c:17:float Variance(float *p, int len)
stats.c:5:float Mean(float *p, int len)

Example 7: Find all assignments where init is involved. Looking again at the node configuration file, we realise that we are looking for Assignment Nodes, which have a ReferVariable descendent, whose name is 'init':

$ redhawk query '**/Assignment/**/ReferVariable@[name="init"]' counter.py

This gives us:

counter.py:2:value = [init]
counter.py:18:init += 1
counter.py:10:self.value = init

Note the @[..] syntax similar to XPATH, for referring to an attribute.

Example 8: What if we wanted assignments were init was being set, and not referred to? We would use a code block to look at the lvalue of the Assignment.:

$ redhawk query '**/Assignment@{n.lvalue.name == "init"}' counter.py

This gives us:

counter.py:18:init += 1

Example 9: Let us find all Function calls that start with 'Counter'. Looking again at the node yaml configuration, we see that we want to find 'CallFunction's, where the function object has a name starting with "Counter".³ :

$ redhawk query '**/CallFunction@{n.function.name[:7] == "Counter"}' counter.py

This gives us:

counter.py:24:c3 = CounterIter()
counter.py:22:c1 = CounterClosure()
counter.py:23:c2 = CounterClass()

Example 10 Let us find all Function definitions whose first argument is self⁴:

$ redhawk query '**/DefineFunction/FunctionArguments/@[name="self"][0]' counter.py

This gives us:

counter.py:12:  def Bump(self):
counter.py:9:  def __init__(self, init=0):

The last [0] is square brackets, indicates the position of that node with respect to its parent.

Example 11 Let us find all Function definitions whose last argument is self. The following query is WRONG:

$ redhawk query '**/DefineFunction/FunctionArguments/@[name="self"][-1]' counter.py

The above query gives us no output. Why? Looking at the node configuration file, we see that, FunctionArguments has three children --- arguments, var_arguments, kwd_arguments, the latter two of which are None everywhere in the file as no variable or keyword arguments are used. Thus, the children of FunctionArguments everywhere in the counter.py file takes the form [[..], None, None].

What we really want, is the last element of the first element, the arguments list. This can be expressed as follows⁵:

$ redhawk query '**/DefineFunction/FunctionArguments/@[name="self"][0, -1]' counter.py

This gives us:

counter.py:12:  def Bump(self):

In hindsight, the query in the previous example could have also been expressed as:

$ redhawk query '**/DefineFunction/FunctionArguments/@[name="self"][0, 0]' counter.py

Note: For convenience's sake, even [0, -1, 0], or [0, -1, 0, 0, .. , 0] is defined to return the same result. Read the 'Position Syntax' section in the documentation of redhawk.common.xpath for more information.

An abstract grammar of the query language can be found via:

$ pydoc redhawk.common.xpath

Much more is possible, using the Selector API.

Using the API

The redhawk package can also be used as an API by importing redhawk.common.selector and related packages. Some of the useful packages are already imported for the user in redhawk prompt and are a good place to start things at.

Example 1: Suppose in the above file we wanted to find all generators, i.e, function definitions, which had a yield as a descendent. We shall see how easy, and logical this query becomes using selectors.

We first go into a redhawk prompt:

$ redhawk prompt counter.py

We are greeted with a help banner:

Built in Variables:
    trees - contains the parse trees of the files passed in the command line

Built in Functions:
    ConvertFileToAst - Converts a file into a language agnostic AST.
    ConvertCodeToAst - Converts a code snippet into a language agnostic AST.
    Help             - Displays this prompt.
    ShowASTAsImage   - Shows the AST as a graph using dot.

Built in Modules:
    S - redhawk.common.selector
    X - redhawk.common.xpath
    F - redhawk.common.format_position

To view this again, use the Help() function.

In the prompt, we define our selectors. (See pydoc redhawk.common.selector for what selectors are, and how they can be composed):

In [1]: function_def = S.S(node_type='DefineFunction')
In [2]: yield_stmt = S.S(node_type='Yield')
In [3]: reqd_selector = function_def.HasDescendant(yield_stmt)

We then apply the selector on the file. The asts of the files passed are in the trees argument. Since this file was the first, it is in `trees[0]`:

In [4]: results = list(reqd_selector(trees))
In [5]: results[0]

gives us:

Out[5]: DefineFunction

This is indeed the function we wanted. Just to be sure, we use the F.PrintContextInFile function to print the context of the tree.:

In [6]: F.PrintContextInFile(results[0], context=6)
counter.py:10:       self.value = init
counter.py:11:   
counter.py:12:     def Bump(self):
counter.py:13:       self.value += 1
counter.py:14:       return self.value
counter.py:15:   
counter.py:16: > def CounterIter(init = 0):
counter.py:17:     while True:
counter.py:18:       init += 1
counter.py:19:       yield init
counter.py:20:   
counter.py:21:   if __name__ == '__main__':
counter.py:22:     c1 = CounterClosure()

It is easy to see from this example that selectors are highly composable, and thus are very powerful. It is hoped that using selectors becomes a natural way to write powerful custom scripts, for querying code.

License

Redhawk is distributed under the terms of the 2-clause BSD license. You are free to use it for commercial or non-commercial projects with little or no restriction. For a complete text of the license see the LICENSE.txt file in the source distribution.

Change List

v1.2.3

* Previous release introduced a regression in C where a top level node was returning an empty AST. This version contains this small but important fix.

v1.2.2

* Support for running Redhawk via Python 3. Great thanks to bkabrda@ and ncoghlan@ for their changes!

• Compatibility for new PyCParser changes.

v1.2.1

• The Lua source code written in ANSI-compliant C, can now be Redhawk-ed!

v1.2.0

• Added new position functionality to xpath.py! (See Examples 10, and 11 aboev for example usage).
• Added to the default imports in prompt.py: redhawk.common.nodes, redhawk.common.types, redhawk.common.xpath
• Added a --show-parsed-query option to redhawk query.
• Made only critical messages appear in the default verbose level.

v1.1.6

• Major internal refactoring involving get_ast.py
• Prompt command accepts directories, and can be told not to use IPython.
• A new selector function called Apply to make prompt usage easier.
• Bug fixes wrt IPython shell and error handling.

v1.1.5

• Vim plugin released.
• Patch to FormatPosition to not strip lines when context = 0.

v1.1.4

• Bugs fixed in xpath.py and pickling of NodeMatchQuery class for Parallel Python.

v1.1.3

Bugs in the README's RST syntax fixed.

v1.1.2

• Redhawk can now use parallel python (on the same machine), to perform queries on codebases. This speeds up Redhawk (almost) proportionally to the number of cores you have on your computer. Redhawk can now query for closures in Django in just ~20 seconds.
• Friendlier usage strings and help messages.

v1.1.1

• Python2.7 compatibility: ast.parse (Thanks to Nafai77)
• Profiled, performance improvements by 15% by shifting to deque, and caching flattened children.
• Provided a bin/start_simple_bash_with_redhawk_in_pythonpath.sh to enter a temporary shell with redhawk in PYTHONPATH (for devs).

v1.1.0

• Fast enough to work on Django - Querying DefineClass anywhere in the codebase (~2300 python files), takes just 45 seconds on a celeron netbook. Thats 19ms per file!
• Uses the shelve module instead of the pickle module, to decrease read and write times for the redhawk database.
• Redhawk supports three new commands - listfiles, remove, where
• The query, and show, commands take an extra argument -s, to decide if new trees should be added to the database.
• Skip a file if there is a parser error.

---

1. ast_gen.py generates node.py and types.py using these YAML configuration files.↩

2. In fact the portion inside the @{..} is just appended to a 'lambda n:' and eval-ed to get a function.↩

3. Note that 'CallFunction's do not directly have a name. This is because the function object, unlike that of a function definition, can be a value. It is possible to do (f.g[x])(y), and such.↩

4. These queries actually finds us the argument, and not the function itself. But this shouldn't matter when we have the definition on the same line.↩

5. These queries actually finds us the argument, and not the function itself. But this shouldn't matter when we have the definition on the same line.↩