- status report
       . compliancy with CPython
       . analysis/checking toolchain
       . easy hacks  -o thunk

   * ep2004-pypy

   - flow obj space
       . placeholder objects
       . records operations

* <=> translator.py: t.view()

   - staticness restrictions
       . do whatever you like at boot-time
       . then no more dynamic function/class creation, no generator
       . constant globals, constant class attributes/methods
       . can use exceptions
       . type coherency

   - type annotation
       . "type" as in "set of objects"
       . custom selection of types

* <=> t.annotate()

   - algorithm
       . "naive" forward propagation
       . also known as "abstract interpretation"
       . bottom-up fixpoint search

   - features
       . analyses the preinitialized program in-memory
       . full-program analysis
       . no user annotation
       . mostly language-independent

   - contrasting to
       . declaring types everywhere
       . exact restriction-less typing (local vars only)
       . full-program source code analysis (Starkiller)

   - (cont)
       . Python2C (supersceded)
       . Pyrex (analysis of Python-like source with types)
       . Psyco (100% run-time)

   - PyPy's type model
       . int, float, str, bool, function, class
       . tuple, list, dict, iter
       . prebuilt-constant
       . class, instance
           > mostly single inheritance
           > attributes lifted to base class on demand

* <=> t.annotate() with OO  (translate_pypy1.py?)

   - code generation
       . can generate Pyrex (at PyCon 2004 already)
       . can generate a slow C extension module (a la Python2C)
       . Common Lisp / Java / LLVM / C++ / Python...

   - next steps
       . good C code
       . Java, for object model
       . LLVM in development (Carl Friedrich Bolz)

   - status
       . rather CPython-compliant source base in PyPy
       . mostly annotatable
       . next step: use the type annotations in the code generator!