PyPy

pypy-logo
PyPy - How to not write Virtual Machines for
Dynamic Languages
Armin Rigo
Institut für Informatik
Heinrich-Heine-Universität Düsseldorf
ESUG 2007
Armin Rigo PyPy - How to not write Virtual Machines for Dynamic Languages

pypy-logo
Scope
This talk is about:
implementing dynamic languages
(with a focus on complicated ones)
in a context of limited resources
(academic, open source, or domain-speciﬁc)

pypy-logo
Scope
This talk is about:
Complicated = requiring a large VM
Smalltalk (etc...): typically small core VM
Python (etc...): the VM contains quite a lot

pypy-logo
Scope
This talk is about:
Complicated = requiring a large VM
Smalltalk (etc...): typically small core VM
Python (etc...): the VM contains quite a lot
Limited resources
Only near-complete implementations are really useful
Minimize implementer’s duplication of efforts

pypy-logo
Our point
Our point:
Do not write virtual machines “by hand”
Instead, write interpreters in high-level languages
Meta-programming is your friend

pypy-logo
Common Approaches to VM construction
Using C directly (or C disguised as another language)
CPython
Ruby
Spidermonkey (Mozilla’s JavaScript VM)
but also: Squeak, Scheme48
Building on top of a general-purpose OO VM
Jython, IronPython
JRuby, IronRuby

pypy-logo
Implementing VMs in C
When writing a VM in C it is hard to reconcile:
ﬂexibility, maintainability
simplicity of the VM
performance (needs dynamic compilation techniques)

pypy-logo
Implementing VMs in C
When writing a VM in C it is hard to reconcile:
ﬂexibility, maintainability
simplicity of the VM
performance (needs dynamic compilation techniques)
Python Case
CPython is a very simple bytecode VM, performance not
great
Psyco is a just-in-time-specializer, very complex, hard to
maintain, but good performance
Stackless is a fork of CPython adding microthreads. It was
never incorporated into CPython for complexity reasons

pypy-logo
Compilers are a bad encoding of Semantics
to reach good performance levels, dynamic compilation is
often needed
a dynamic compiler needs to encode language semantics
this encoding is often obscure and hard to change

pypy-logo
Compilers are a bad encoding of Semantics
to reach good performance levels, dynamic compilation is
often needed
a dynamic compiler needs to encode language semantics
this encoding is often obscure and hard to change
Python Case
Psyco is a dynamic compiler for Python
synchronizing with CPython’s rapid development is a lot of
effort
many of CPython’s new features not supported well

pypy-logo
Fixing of Early Design Decisions
when starting a VM in C, many design decisions need to
be made upfront
examples: memory management technique, threading
model
the decision is manifested throughout the VM source
very hard to change later

pypy-logo
Fixing of Early Design Decisions
when starting a VM in C, many design decisions need to
be made upfront
examples: memory management technique, threading
model
the decision is manifested throughout the VM source
very hard to change later
Python Case
CPython uses reference counting, increfs and decrefs
everywhere
CPython uses OS threads with one global lock, hard to
change to lightweight threads or ﬁner locking

pypy-logo
Implementation Proliferation
restrictions of the original implementation lead to
re-implementations, forks
all implementations need to be synchronized with
language evolution
lots of duplicate effort

pypy-logo
Implementation Proliferation
restrictions of the original implementation lead to
re-implementations, forks
all implementations need to be synchronized with
language evolution
lots of duplicate effort
Python Case
several serious implementations: CPython, Stackless,
Psyco, Jython, IronPython, PyPy
the implementations have various grades of compliance

pypy-logo
Implementing Languages on Top of General-Purpose
OO VMs
users wish to have easy interoperation with the
general-purpose OO VMs used by the industry (JVM, CLR)
therefore re-implementations of the language on the OO
VMs are started
even more implementation proliferation
implementing on top of an OO VM has its own set of
problems

pypy-logo
Implementing Languages on Top of General-Purpose
OO VMs
users wish to have easy interoperation with the
general-purpose OO VMs used by the industry (JVM, CLR)
therefore re-implementations of the language on the OO
VMs are started
even more implementation proliferation
implementing on top of an OO VM has its own set of
problems
Python Case
Jython is a Python-to-Java-bytecode compiler
IronPython is a Python-to-CLR-bytecode compiler
both are slightly incompatible with the newest CPython
version (especially Jython)

pypy-logo
Beneﬁts of implementing on top of OO VMs
higher level of implementation
the VM supplies a GC and mostly a JIT
better interoperability than what the C level provides
some proponents believe that eventually one single VM
should be enough

pypy-logo
The problems of OO VMs
some of the beneﬁts of OO VMs don’t work out in practice
most immediate problem: it can be hard to map concepts
of the dynamic lang to the host OO VM
performance is often not improved, and can be very bad,
because of the semantic mismatch between the dynamic
language and the host VM
poor interoperability with everything outside the OO VM
in practice, one OO VM is not enough

pypy-logo
The problems of OO VMs
some of the beneﬁts of OO VMs don’t work out in practice
most immediate problem: it can be hard to map concepts
of the dynamic lang to the host OO VM
performance is often not improved, and can be very bad,
because of the semantic mismatch between the dynamic
language and the host VM
poor interoperability with everything outside the OO VM
in practice, one OO VM is not enough
Python Case
Jython about 5 times slower than CPython
IronPython is about as fast as CPython (but some
introspection features missing)

pypy-logo
PyPy’s Approach to VM Construction
Goal: achieve ﬂexibility, simplicity and performance together
Approach: auto-generate VMs from high-level descriptions
of the language
... using meta-programming techniques and aspects
high-level description: an interpreter written in a high-level
language
... which we translate (i.e. compile) to VMs running on top
of various targets, like C/Posix, CLR, JVM

pypy-logo
PyPy
PyPy = Python interpreter written in RPython + translation
toolchain for RPython

pypy-logo
PyPy
PyPy = Python interpreter written in RPython + translation
toolchain for RPython
What is RPython
RPython is a subset of Python
subset chosen in such a way that type-inference can be
performed
still a high-level language (unlike SLang or Prescheme)
...really a subset, can’t give a small example of code that
doesn’t just look like Python :-)

pypy-logo
Auto-generating VMs
high-level source: early design decisions not necessary
we need a custom translation toolchain to compile the
interpreter to a full VM
many aspects of the ﬁnal VM are orthogonal to the
interpreter source: they are inserted during translation
translation aspect ∼= monads, with more ad-hoc control

pypy-logo
Auto-generating VMs
high-level source: early design decisions not necessary
we need a custom translation toolchain to compile the
interpreter to a full VM
many aspects of the ﬁnal VM are orthogonal to the
interpreter source: they are inserted during translation
translation aspect ∼= monads, with more ad-hoc control
Examples
Garbage Collection strategy
Threading models (e.g. coroutines with CPS...)
non-trivial translation aspect: auto-generating a dynamic
compiler from the interpreter

pypy-logo
Good Points of the Approach
Simplicity:
dynamic languages can be implemented in a high level
language
separation of concerns from low-level details
a potential single-source-ﬁts-all interpreter – less
duplication of efforts
runs everywhere with the same semantics – no outdated
implementations, no ties to any standard platform

pypy-logo
Simplicity:
dynamic languages can be implemented in a high level
language
separation of concerns from low-level details
a potential single-source-ﬁts-all interpreter – less
duplication of efforts
runs everywhere with the same semantics – no outdated
implementations, no ties to any standard platform
PyPy
arguably the most readable Python implementation so far

pypy-logo
Flexibility at all levels:
when writing the interpreter (high-level languages rule!)
when adapting the translation toolchain as necessary
to break abstraction barriers when necessary

pypy-logo
Flexibility at all levels:
when writing the interpreter (high-level languages rule!)
when adapting the translation toolchain as necessary
to break abstraction barriers when necessary
Example
boxed integer objects, represented as tagged pointers
manual system-level RPython code

pypy-logo
Performance:
“reasonable” performance
can generate a dynamic compiler from the interpreter
(work in progress, 60x faster on very simple Python code)

pypy-logo
Performance:
“reasonable” performance
can generate a dynamic compiler from the interpreter
(work in progress, 60x faster on very simple Python code)
JIT compiler generator
almost orthogonal from the interpreter source - applicable
to many languages, follows language evolution “for free”
based on Partial Evaluation
beneﬁts from a high-level interpreter and a tweakable
translation toolchain
generating a dynamic compiler is easier than generating a
static one!

pypy-logo
Open Issues / Drawbacks / Further Work
writing the translation toolchain in the ﬁrst place takes lots
of effort (but it can be reused)
writing a good GC is still necessary. But: maybe we can
reuse existing good GCs (e.g. from the Jikes RVM)?
conceptually simple approach but many abstraction layers
dynamic compiler generation seems to work, but needs
more efforts. Also: can we layer it on top of the JIT of a
general purpose OO VM?

pypy-logo
Conclusion / Meta-Points
high-level languages are suitable to implement dynamic
languages
doing so has many beneﬁts
VMs shouldn’t be written by hand
PyPy’s concrete approach is not so important
diversity is good
let’s write more meta-programming toolchains!

pypy-logo
For more information
PyPy
http://codespeak.net/pypy/
“Sprints”
Main way we develop PyPy
They are programming camps, a few days to one week
long
We may have one in Bern soon (PyPy+Squeak) and/or in
Germany (JIT and other topics)
See also
Google for the full paper corresponding to these slides that was
submitted at Dyla’2007

PyPy

More Related Content

What's hot

Similar to PyPy

More from ESUG

Recently uploaded

PyPy