Next step after pychecker

[Steven Bethard]

Philippe Fremy wrote:
> I would like to develop a tool that goes one step further than pychecker 
> to ensure python program validity. The idea would be to get close to 
> what people get on ocaml: a static verification of all types of the 
> program, without any kind of variable declaration. This would definitely 
> brings a lot of power to python.

It's a very cool idea, and there was talk of including such a tool in 
Python 3.0, though as yet, no one has volunteered to solve this problem. 
=)  Note that you can't actually really ensure program validity, but you 
can make some educated guesses like pychecker does.  An example:

def f1(a):
     return a + 1

def f2():
     return f1([])

Your suggestion was that this code should be deemed invalid.  But if 
it's followed with:

f1 = ''.join
f2()

where f1 is rebound, then it's perfectly valid code.  That said, I think 
providing "suggestions" like pychecker does is definitely worth pursuing.

> What is in your opinion the best tool to achieve this ? I had an 
> extensive look at pychecker, and it could certainly be extended to do 
> the job. Things that still concern me are that it works on the bytecode, 
> which prevents it from working with jython and the new .NET python.

I don't know much about what pychecker does, but if it works with the 
bytecode, shouldn't it be fine for jython and IronPython?  I thought the 
bytecode was part of the language spec, and what was CPython specific 
was how the bytecodes were actually implemented...

> I am currently reading the documentation on AST and visitor, but I am 
> not sure that this will be the best tool either. The AST seems quite 
> deep and I am afraid that it will make the analysis quite slow and 
> complicated.

You should read up on the recent python-dev discussions about merging 
the AST branch.  I don't know too much about it, but apparently there's 
some work being done currently on how the AST is used in Python.  Sorry 
I don't know more about this...


Steve