Task Isolation in Python (was Re: How do you develop in Python?)
Glyph Lefkowitz
glyph at twistedmatrix.com
Sun Jun 10 18:37:29 EDT 2001
On Sun, 10 Jun 2001, Nick Perkins wrote:
> I want an IDE to provide at least the following:
> 1. Each run takes place in a fresh, clean environment.
> 2. The IDE stays responsive while a program is run.
> 3. The IDE can kill a running program.
One idea which I've been kicking around (and finally even started a little
implementation of last night) is a Python interpreter in Python. I
believe that the existing compiler is relatively robust (at least, I
haven't been able to generate any source code that can crash or hang it
yet), but the bytecode interpreter can't isolate tasks, which makes it
impossible to kill a process or even slow it down to see what it's doing.
Stackless seems like it might make this possible, but it's still using
essentially the same partially-recursive interpreter. I'm also interested
in running completely untrusted code from python bytecode. However, the
tricky parts are in making sure that this meta-interpreter isn't
recursive. (In particular, I am getting stuck on operator overloading,
since it's possible to implement __getattr__...)
Some advantages that this approach would have:
* .pyc bytecode interpreter for Jython.
* untrusted code execution
* resource control
* mobile code
* defined API for communication between isolated tasks
* ability to run "blocking" code without threads in a singlethreaded app
* educational value; easier to read than ceval.c
* task-scheduling in python without continuations
* IDEs could safely execute/debug development code w/o a separate process
* python code would run 2 orders of magnitude slower, making it even more
high-level
Has anyone attempted such a thing before, or have any advice to give me?
Also, is it easy/possible to crash the "marshal" module with bad data?
______ __ __ _____ _ _
| ____ | \_/ |_____] |_____|
|_____| |_____ | | | |
@ t w i s t e d m a t r i x . c o m
http://twistedmatrix.com/users/glyph
More information about the Python-list
mailing list