- Getting Started with RPython
- Trying out the translator
- Where to start reading the sources
- Running PyPy’s unit tests
- Special Introspection Features of the Untranslated Python Interpreter
- Additional Tools for running (and hacking) PyPy
- Getting involved
Please read this FAQ entry first!
RPython is a subset of Python that can be statically compiled. The PyPy
interpreter is written mostly in RPython (with pieces in Python), while
the RPython compiler is written in Python. The hard to understand part
is that Python is a meta-programming language for RPython, that is,
“being valid RPython” is a question that only makes sense on the
live objects after the imports are done.
This might require more explanation. You start writing RPython from
entry_point, a good starting point is
rpython/translator/goal/targetnopstandalone.py. This does not do all that
much, but is a start. Now if code analyzed (in this case
calls some functions, those calls will be followed. Those followed calls
have to be RPython themselves (and everything they call etc.), however not
entire module files. To show how you can use metaprogramming, we can do
a silly example (note that closures are not RPython):
def generator(operation): if operation == 'add': def f(a, b): return a + b else: def f(a, b): return a - b return f add = generator('add') sub = generator('sub') def entry_point(argv): print add(sub(int(argv), 3) 4) return 0
In this example
entry_point is RPython,
sub are RPython,
generator is not.
The following introductory level articles are available:
- Laurence Tratt – Fast Enough VMs in Fast Enough Time.
- How to write interpreters in RPython and part 2 by Andrew Brown.
The translator is a tool based on the PyPy interpreter which can translate sufficiently static RPython programs into low-level code (in particular it can be used to translate the full Python interpreter). To be able to experiment with it you need to download and install the usual (CPython) version of:
To start the interactive translator shell do:
cd rpython python bin/translatorshell.py
Test snippets of translatable code are provided in the file
rpython/translator/test/snippet.py, which is imported under the name
snippet. For example:
>>> t = Translation(snippet.is_perfect_number, [int]) >>> t.view()
After that, the graph viewer pops up, that lets you interactively inspect the flow graph. To move around, click on something that you want to inspect. To get help about how to use it, press ‘H’. To close it again, press ‘Q’.
We have a type annotator that can completely infer types for functions like
is_perfect_number (as well as for much larger examples):
>>> t.annotate() >>> t.view()
Move the mouse over variable names (in red) to see their inferred types.
The graph can be turned into C code:
>>> t.rtype() >>> lib = t.compile_c()
The first command replaces the operations with other low level versions that
only use low level types that are available in C (e.g. int). The compiled
version is now in a
.so library. You can run it say using ctypes:
>>> f = get_c_function(lib, snippet.is_perfect_number) >>> f(5) 0 >>> f(6) 1
There is a small-to-medium demo showing the translator and the annotator:
python bin/rpython --view --annotate translator/goal/bpnn.py
bpnn.py to display itself as a call graph and class
hierarchy. Clicking on functions shows the flow graph of the particular
function. Clicking on a class shows the attributes of its instances. All
this information (call graph, local variables’ types, attributes of
instances) is computed by the annotator.
To turn this example to C code (compiled to the executable
python bin/rpython translator/goal/bpnn.py
To translate full RPython programs, there is the script
rpython/translator/goal. Examples for this are a slightly changed version of
python bin/rpython translator/goal/targetrpystonedalone
This will produce the executable “targetrpystonedalone-c”.
There are several environment variables you can find useful while playing with the RPython:
- RPython uses temporary session directories to store files that are generated during the
translation process(e.g., translated C files).
PYPY_USESSION_DIRserves as a base directory for these session dirs. The default value for this variable is the system’s temporary dir.
- By default RPython keeps only the last
PYPY_USESSION_KEEP(defaults to 3) session dirs inside
PYPY_USESSION_DIR. Increase this value if you want to preserve C files longer (useful when producing lots of lldebug builds).
PyPy is made from parts that are relatively independent of each other. You should start looking at the part that attracts you most (all paths are relative to the PyPy top level directory). You may look at our directory reference or start off at one of the following points:
- pypy/interpreter contains the bytecode interpreter: bytecode dispatcher in pypy/interpreter/pyopcode.py, frame and code objects in pypy/interpreter/eval.py and pypy/interpreter/pyframe.py, function objects and argument passing in pypy/interpreter/function.py and pypy/interpreter/argument.py, the object space interface definition in pypy/interpreter/baseobjspace.py, modules in pypy/interpreter/module.py and pypy/interpreter/mixedmodule.py. Core types supporting the bytecode interpreter are defined in pypy/interpreter/typedef.py.
- pypy/interpreter/pyparser contains a recursive descent parser, and grammar files that allow it to parse the syntax of various Python versions. Once the grammar has been processed, the parser can be translated by the above machinery into efficient code.
- pypy/interpreter/astcompiler contains the compiler. This contains a modified version of the compiler package from CPython that fixes some bugs and is translatable.
- pypy/objspace/std contains the Standard object space. The main file
is pypy/objspace/std/objspace.py. For each type, the files
xxxobject.pycontain respectively the definition of the type and its (default) implementation.
- rpython/translator contains the code analysis and generation stuff. Start reading from translator.py, from which it should be easy to follow the pieces of code involved in the various translation phases.
- rpython/annotator contains the data model for the type annotation that can be inferred about a graph. The graph “walker” that uses this is in rpython/annotator/annrpython.py.
- rpython/rtyper contains the code of the RPython typer. The typer transforms annotated flow graphs in a way that makes them very similar to C code so that they can be easy translated. The graph transformations are controlled by the code in rpython/rtyper/rtyper.py. The object model that is used can be found in rpython/rtyper/lltypesystem/lltype.py. For each RPython type there is a file rxxxx.py that contains the low level functions needed for this type.
- rpython/rlib contains the RPython standard library, things that you can use from rpython.
The PyPy source tree comes with an inlined version of
which you can invoke by typing:
python pytest.py -h
This is usually equivalent to using an installed version:
If you encounter problems with the installed version
make sure you have the correct version installed which
you can find out with the
Now on to running some tests. PyPy has many different test directories and you can use shell completion to point at directories or files:
py.test pypy/interpreter/test/test_pyframe.py # or for running tests of a whole subdirectory py.test pypy/interpreter/
See py.test usage and invocations for some more generic info on how you can run tests.
Beware trying to run “all” pypy tests by pointing to the root
directory or even the top level subdirectory
pypy. It takes
hours and uses huge amounts of RAM and is not recommended.
To run CPython regression tests you can point to the
This will usually take a long time because this will run the PyPy Python interpreter on top of CPython. On the plus side, it’s usually still faster than doing a full translation and running the regression test with the translated PyPy Python interpreter.
If you are interested in the inner workings of the PyPy Python interpreter, there are some features of the untranslated Python interpreter that allow you to introspect its internals.
If you start an untranslated Python interpreter via:
If you press
<Ctrl-C> on the console you enter the interpreter-level console, a
usual CPython console. You can then access internal objects of PyPy
(e.g. the object space) and any variables you have created on the PyPy
prompt with the prefix
>>>> a = 123 >>>> <Ctrl-C> *** Entering interpreter-level console *** >>> w_a W_IntObject(123)
The mechanism works in both directions. If you define a variable with the
w_ prefix on the interpreter-level, you will see it on the app-level:
>>> w_l = space.newlist([space.wrap(1), space.wrap("abc")]) >>> <Ctrl-D> *** Leaving interpreter-level console *** KeyboardInterrupt >>>> l [1, 'abc']
Note that the prompt of the interpreter-level console is only ‘>>>’ since it runs on CPython level. If you want to return to PyPy, press <Ctrl-D> (under Linux) or <Ctrl-Z>, <Enter> (under Windows).
You may be interested in reading more about the distinction between interpreter-level and app-level.
You can use a simple tracing mode to monitor the interpretation of
bytecodes. To enable it, set
__pytrace__ = 1 on the interactive
>>>> __pytrace__ = 1 Tracing enabled >>>> x = 5 <module>: LOAD_CONST 0 (5) <module>: STORE_NAME 0 (x) <module>: LOAD_CONST 1 (None) <module>: RETURN_VALUE 0 >>>> x <module>: LOAD_NAME 0 (x) <module>: PRINT_EXPR 0 5 <module>: LOAD_CONST 0 (None) <module>: RETURN_VALUE 0 >>>>
The example-interpreter repository contains an example interpreter written using the RPython translation toolchain.
We use some optional tools for developing PyPy. They are not required to run the basic tests or to get an interactive PyPy prompt but they help to understand and debug PyPy especially for the translation process.
graphviz and pygame are both necessary if you want to look at generated flow graphs:
The py.test testing tool drives all our testing needs.
We use the py library for filesystem path manipulations, terminal writing, logging and some other support functionality.
You don’t necessarily need to install these two libraries because we also ship them inlined in the PyPy source tree.
PyPy employs an open development process. You are invited to join our pypy-dev mailing list or look at the other contact possibilities. Usually we give out commit rights fairly liberally, so if you want to do something with PyPy, you can become a committer. We are also doing coding Sprints which are separately announced and often happen around Python conferences such as EuroPython or Pycon. Upcoming events are usually announced on the blog.