[Bf-committers] "Security" gets in the way

[Benjamin Tolputt]

horace grant wrote:
> no need for lua. python is the much nicer language. :p there is pypy
> which supports sandboxing and which also gets cpython api compatible
> at the moment.
>
> http://morepypy.blogspot.com/2010/04/using-cpython-extension-modules-with.html
>
> in 2 years or so (once pypy is more mature and python 3 compatible) it
> should be no big problem to replace cpython with pypy. as another
> benefit pypy will be much faster than cpython due to its jit compiler.
>   

Whether Python is or isn't a nicer language depends on your point of
view, so I won't debate that.

However, the "sand-boxing" as presented in PyPy is very crude and will
do nothing to fix the issues with Python in Blender. The major problem
with Python in Blender is not that it can access files "in general" (as
that is a REQUIREMENT of import / export scripts for example) but that I
can access EVERYTHING Python can from every execution context. That is,
I might only want Python to have access to other elements in the scene
(say for a rig or controlling a particle simulation) but, so long as
Python can access files (which, as I said, is *required*) everything
executing Python code can.

In Lua, AngelScript, Falcon, TinyScheme, etc it is possible to only
expose to the execution context that which you want it to have access
to. If you don't want it to read/write files - don't give it the
necessary modules/functions. This is not possible in Python (everything
is accessible everywhere) and the sand-boxing in PyPy is an "all or
nothing" affair. Either you can access the file system or you cannot. No
way to only restrict access in only some scripts (say those included in
the untrusted .blend file) and not others (those installed by the
end-user in the .blender/scripts directory). Not to mention the
performance issues with the method PyPy users (dual processes - with all
"sand-boxed" data needing to be marshalled between the Blender/Python
process and it's sand-boxed proxy).

Sorry, Python is designed in such a way as to make securing it an
unlikely scenario.