Motivation for asteval

The asteval module allows you to evaluate a large subset of the Python language from within a python program, without using eval(). It is, in effect, a restricted version of Python’s built-in eval(), forbidding several actions, and using using a simple dictionary as a flat namespace. A completely fair question is: Why is this desirable? That is, why not simply use eval(), or just use Python itself?

The short answer is that sometimes you want to allow evaluation of user input, or expose a simple calculator inside a larger application. For this, eval() is pretty scary, as it exposes all of Python, which makes user input difficult to trust. Since asteval does not support the import statement (or many other constructs), user code cannot access the os and sys modules or any functions or classes outside the provided symbol table.

Other missing features (modules, classes, lambda, yield, generators) are similarly motivated by a desire for a safer version of eval(). The idea for asteval is to make a simple procedural, mathematically oriented language that can be embedded into larger applications.

In fact, the asteval module grew out the the need for a simple expression evaluator for scientific applications such as the lmfit and xraylarch modules. A first attempt using the pyparsing module worked but was error-prone and difficult to maintain. It turned out that using the Python ast module is so easy that adding more complex programming constructs like conditionals, loops, exception handling, complex assignment and slicing, and even user-defined functions was fairly simple to implement. Importantly, because parsing is done by the ast module, whole classes of implementation errors disappear. Valid python expression will be parsed correctly and converted into an Abstract Syntax Tree. Furthermore, the resulting AST is easy to walk through, greatly simplifying evaluation over any other approach. What started as a desire for a simple expression evaluator grew into a quite useable procedural domain-specific language for mathematical applications.

Asteval makes no claims about speed. Obviously, evaluating the ast tree involves a lot of function calls, and will likely be slower than Python. In preliminary tests, it’s about 4x slower than Python. For certain use cases (see, use of asteval and numpy can approach the speed of eval and the numexpr modules.

How Safe is asteval?

Asteval avoids the known exploits that make eval() dangerous. For reference, see, Eval is really dangerous and the comments and links therein. From this discussion it is apparent that not only is eval() unsafe, but that it is a difficult prospect to make any program that takes user input perfectly safe. In particular, if a user can cause Python to crash with a segmentation fault, safety cannot be guaranteed. Asteval explicitly forbids the exploits described in the above link, and works hard to prevent malicious code from crashing Python or accessing the underlying operating system. That said, we cannot guarantee that asteval is completely safe from malicious code. We claim only that it is safer than the builtin eval(), and that you might find it useful.

Some of the things not allowed in the asteval interpreter for safety reasons include:

In addition (and following the discussion in the link above), the following attributes are blacklisted for all objects, and cannot be accessed:

__subclasses__, __bases__, __globals__, __code__, __closure__, __func__, __self__, __module__, __dict__, __class__, __call__, __get__, __getattribute__, __subclasshook__, __new__, __init__, func_globals, func_code, func_closure, im_class, im_func, im_self, gi_code, gi_frame f_locals, __mro__

This approach of making a blacklist cannot be guaranteed to be complete, but it does eliminate classes of attacks known to seg-fault the Python. On the other hand, asteval will typically expose numpy ufuncs from the numpy module, and several of these can seg-fault Python without too much trouble. If you’re paranoid about safe user input that can never cause a segmentation fault, you’ll want to disable the use of numpy.

There are important categories of safety that asteval does not even attempt to address. The most important of these is resource hogging, which might be used for a denial-of-service attack. There is no guaranteed timeout on any calculation, and so a reasonable looking calculation such as:

from asteval import Interpreter
aeval = Interpreter()
txt = """nmax = 1e8
a = sqrt(arange(nmax))

can take a noticeable amount of CPU time. It is not hard to come up with short program that would run for hundreds of years, which probably exceeds anyones threshold for an acceptable run-time. As a very simple example, it is very hard to predict how long the expression x**y**z will take to run without knowing the values of x, y, and z. In short, runtime cannot be determined lexically.

For a limited range of problems, you can try to avoid asteval taking too long. For example, you may try to limit the recursion limit when executing expressions, with a code like this:

import contextlib

def limited_recursion(recursion_limit):
    old_limit = sys.getrecursionlimit()

with limited_recursion(100):

You can also pass in a max_time (in seconds) when you create an asteval Interpreter, wich will try to limit the amount of time an expression will take. This is actually of limited utility, since the calculation must return to the asteval interpreter for the runtime to be checked at all. Many long-running calculations will be stuck deep inside C-code evaluated by the Python interpreter itself, and not return or allow other threads to run until that calculation is done. That is, from within a single process, there really is not a foolproof way to tell asteval to have a maximum runtime. The most reliable way to put a firm limit on runtime is to have a second process watching the execution time of the asteval process and interrupt or kill it.

In summary, while asteval attempts to be safe and is definitely safer than using eval(), there are many ways that asteval could be considered part of an un-safe programming environment. Recommendations for how to improve this situation would be greatly appreciated.