Skip to content

Simple Safe Sandboxed Extensible Expression Evaluator for Python

License

Notifications You must be signed in to change notification settings

VerdigrisTech/simpleeval

 
 

Repository files navigation

simpleeval (Simple Eval)

Build Status Coverage Status PyPI Version

A quick single file library for easily adding evaluatable expressions into python projects. Say you want to allow a user to set an alarm volume, which could depend on the time of day, alarm level, how many previous alarms had gone off, and if there is music playing at the time.

Or if you want to allow simple formulae in a web application, but don't want to give full eval() access, or don't want to run in javascript on the client side.

It's deliberately very simple, pull it in from PyPI (pip or easy_install), or even just a single file you can dump into a project.

Internally, it's using the amazing python ast module to parse the expression, which allows very fine control of what is and isn't allowed. It should be completely safe in terms of what operations can be performed by the expression.

The only issue I know to be aware of is that you can create an expression which takes a long time to evaluate, or which evaluating requires an awful lot of memory, which leaves the potential for DOS attacks. There is basic protection against this, and you can lock it down further if you desire. (see the Operators section below)

You should be aware of this when deploying in a public setting.

The defaults are pretty locked down and basic, and it's very easy to add whatever extra specific functionality you need (your own functions, variable/name lookup, etc).

Basic Usage

To get very simple evaluating:

from simpleeval import simple_eval

simple_eval("21 + 21")

returns 42.

Expressions can be as complex and convoluted as you want:

simple_eval("21 + 19 / 7 + (8 % 3) ** 9")

returns 535.714285714.

You can add your own functions in as well.

simple_eval("square(11)", functions={"square": lambda x: x*x})

returns 121.

For more details of working with functions, read further down.

Note:

all further examples use >>> to designate python code, as if you are using the python interactive prompt.

Operators

You can add operators yourself, using the operators argument, but these are the defaults:

+ add two things. x + y 1 + 1 -> 2
- subtract two things x - y 100 - 1 -> 99
/ divide one thing by another x / y 100/10 -> 10
* multiple one thing by another x * y 10 * 10 -> 100
** 'to the power of' x**y 2 ** 10 -> 1024
% modulus. (remainder) x % y 15 % 4 -> 3
== equals x == y 15 == 4 -> False
< Less than. x < y 1 < 4 -> True
> Greater than. x > y 1 > 4 -> False
<= Less than or Equal to. x <= y 1 < 4 -> True
>= Greater or Equal to x >= 21 1 >= 4 -> False
in is something contained within something else. "spam" in "my breakfast" -> False

The ^ operator is notably missing - not because it's hard, but because it is often mistaken for a exponent operator, not the bitwise operation that it is in python. It's trivial to add back in again if you wish (using the class based evaluator explained below):

>>> import ast
>>> import operator

>>> s = SimpleEval()
>>> s.operators[ast.BitXor] = operator.xor

>>> s.eval("2 ^ 10")
8

Limited Power

Also note, the ** operator has been locked down by default to have a maximum input value of 4000000, which makes it somewhat harder to make expressions which go on for ever. You can change this limit by changing the simpleeval.POWER_MAX module level value to whatever is an appropriate value for you (and the hardware that you're running on) or if you want to completely remove all limitations, you can set the s.operators[ast.Pow] = operator.pow or make your own function.

On my computer, 9**9**5 evaluates almost instantly, but 9**9**6 takes over 30 seconds. Since 9**7 is 4782969, and so over the POWER_MAX limit, it throws a NumberTooHigh exception for you. (Otherwise it would go on for hours, or until the computer runs out of memory)

Strings (and other Iterables) Safety

There are also limits on string length (100000 characters, MAX_STRING_LENGTH). This can be changed if you wish.

Related to this, if you try to create a silly long string/bytes/list, by doing 'i want to break free'.split() * 9999999999 for instance, it will block you.

If Expressions

You can use python style if x then y else z type expressions:

>>> simple_eval("'equal' if x == y else 'not equal'",
                names={"x": 1, "y": 2})
'not equal'

which, of course, can be nested:

>>> simple_eval("'a' if 1 == 2 else 'b' if 2 == 3 else 'c'")
'c'

Functions

You can define functions which you'd like the expresssions to have access to:

>>> simple_eval("double(21)", functions={"double": lambda x:x*2})
42

You can define "real" functions to pass in rather than lambdas, of course too, and even re-name them so that expressions can be shorter

>>> def double(x):
        return x * 2
>>> simple_eval("d(100) + double(1)", functions={"d": double, "double":double})
202

If you don't provide your own functions dict, then the the following defaults are provided in the DEFAULT_FUNCTIONS dict:

randint(x) Return a random int below x
rand() Return a random float between 0 and 1
int(x) Convert x to an int.
float(x) Convert x to a float.
str(x) Convert x to a str (unicode in py2)

If you want to provide a list of functions, but want to keep these as well, then you can do a normal python .copy() & .update:

>>> my_functions = simpleeval.DEFAULT_FUNCTIONS.copy()
>>> my_functions.update(
        square=(lambda x:x*x),
        double=(lambda x:x+x),
    )
>>> simple_eval('square(randint(100))', functions=my_functions)

Names

Sometimes it's useful to have variables available, which in python terminology are called 'names'.

>>> simple_eval("a + b", names={"a": 11, "b": 100})
111

You can also hand the handling of names over to a function, if you prefer:

>>> def name_handler(node):
        return ord(node.id[0].lower(a))-96

>>> simple_eval('a + b', names=name_handler)
3

That was a bit of a silly example, but you could use this for pulling values from a database or file, say, or doing some kind of caching system.

The two default names that are provided are True and False. So if you want to provide your own names, but want True and False to keep working, either provide them yourself, or .copy() and .update the DEFAULT_NAMES. (See functions example above).

Creating an Evaluator Class

Rather than creating a new evaluator each time, if you are doing a lot of evaluations, you can create a SimpleEval object, and pass it expressions each time (which should be a bit quicker, and certainly more convenient for some use cases):

>>> s = SimpleEval()

>>> s.eval("1 + 1")
2

>>> s.eval('100 * 10')
1000

# and so on...

You can assign / edit the various options of the SimpleEval object if you want to. Either assign them during creation (like the simple_eval function)

def boo():
    return 'Boo!'

s = SimpleEval(functions={"boo": boo})

or edit them after creation:

s.names['fortytwo'] = 42

this actually means you can modify names (or functions) with functions, if you really feel so inclined:

s = SimpleEval()
def set_val(name, value):
    s.names[name.value] = value.value
    return value.value

s.functions = {'set': set_val}

s.eval("set('age', 111)")

Say. This would allow a certain level of 'scriptyness' if you had these evaluations happening as callbacks in a program. Although you really are reaching the end of what this library is intended for at this stage.

Compound Types

Compound types (dict, tuple, list, set) in general just work if you pass them in as named objects. If you want to allow creation of these, the EvalWithCompoundTypes class works. Just replace any use of SimpleEval with that.

The EvalWithCompoundTypes class also contains support for simple comprehensions. eg: [x + 1 for x in [1,2,3]]. There's a safety MAX_COMPREHENSION_LENGTH to control how many items it'll allow before bailing too. This also takes into account nested comprehensions.

Since the primary intention of this library is short expressions - an extra 'sweetener' is enabled by default. You can access a dict (or similar's) keys using the .attr syntax:

>>>  simple_eval("foo.bar", names={"foo": {"bar": 42}})
42

for instance. You can turn this off either by setting the module global ATTR_INDEX_FALLBACK to False, or on the SimpleEval instance itself. e.g. evaller.ATTR_INDEX_FALLBACK=False.

Extending

The SimpleEval class is pretty easy to extend. For instance, to create a version that disallows method invocation on objects:

import ast
import simpleeval

class EvalNoMethods(simpleeval.SimpleEval):
    def _eval_call(self, node):
        if isinstance(node.func, ast.Attribute):
            raise simpleeval.FeatureNotAvailable("No methods please, we're British")
        return super(EvalNoMethods, self)._eval_call(node)

and then use EvalNoMethods instead of the SimpleEval class.

Other...

The library supports both python 2 and 3.

Object attributes that start with _ or func_ are disallowed by default. If you really need that (BE CAREFUL!), then modify the module global simpleeval.DISALLOW_PREFIXES.

A few builtin functions are listed in simpleeval.DISALLOW_FUNCTIONS. type, open, etc. If you need to give access to this kind of functionality to your expressions, then be very careful. You'd be better wrapping the functions in your own safe wrappers.

The initial idea came from J.F. Sebastian on Stack Overflow ( http://stackoverflow.com/a/9558001/1973500 ) with modifications and many improvements, see the head of the main file for contributors list.

Please read the test_simpleeval.py file for other potential gotchas or details. I'm very happy to accept pull requests, suggestions, or other issues. Enjoy!

Developing

Run tests:

$ make test

Or to set the tests running on every file change:

$ make autotest

(requires entr)

BEWARE

I've done the best I can with this library - but there's no warrenty, no guarentee, nada. A lot of very clever people think the whole idea of trying to sandbox CPython is impossible. Read the code yourself, and use it at your own risk.

About

Simple Safe Sandboxed Extensible Expression Evaluator for Python

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages

  • Python 98.7%
  • Makefile 1.3%