namespace._copy trips on dict values that are types

[aaaaalbert]

A function declared to be available inside the RepyV2 namespace via one of namespace.py's *_WRAPPER_INFO wrapper dicts may list Dict() as (one of) its args processor(s). A call from within the sandbox into such function causes the argument(s) to be _copyed. Now certain dictionary contents cause problems during copying, particularly those that use types as values.

For instance, virtualnamespace.evaluate takes a dict or SafeDict argument. Supplying a dict with a problematic value like so,

v = createvirtualnamespace("", "")
v.evaluate({"abc": str}) # Ouch

breaks as namespace._copy raises Exception (with class 'namespace.NamespaceInternalError'): _copy failed on {'abc': <type 'str'>} with message _copy failed on <type 'str'> with message _copy is not implemented for objects of type <function safe_type at 0x10d0021b8>. The issue is that str is a type, or more accurately a safe.safe_type, and there is no handling for that in _copy.

Funnily enough, the problem is avoided by providing a SafeDict with the same contents, because the SafeDict makes a different clause in _copy match. The difference is that SafeDicts are instances / types.InstanceTypes.

v = createvirtualnamespace("", "")
v.evaluate(SafeDict({"abc": str})) # Success!

---

The superficial fix for this corner case of a corner case of a bit of funtionality I've never seen anyone use is to add safe.safe_type to the list of types that namespace._copy checks the type of its argument, obj, against. Whether or not dicts should be treated all that differently from SafeDicts as they traverse the namespace boundary is a different question though.

---

Historical note: I found this problem via the original virtualnamespace context safety unit test, although that test was not supposed the above problem: https://github.com/SeattleTestbed/repy_v2/blob/4db9ecd48a4577d646e4c38cf1af634dd8600810/testsV2/ut_repyv2api_virtualnamespacecontextsafety.py .

[aaaaalbert]

The rough patch thus is

--- ../namespace.py
+++ ./namespace.py
@@ -814,5 +814,6 @@
     if _is_in(type(obj), [str, unicode, int, long, float, complex, bool, frozenset,
                           types.NoneType, types.FunctionType, types.LambdaType,
-                          types.MethodType, types.InstanceType]):
+                          types.MethodType, types.InstanceType,
+                          safe.safe_type]):
       return obj

[aaaaalbert]

Side note while constructing the unit test: besides str, object also triggers the bug. The workaround is the same.

[aaaaalbert]

...and digging some more (with help(types) as a handy reference) reveals a few more:

unusual_cases = [
    # thing        # type of thing
    DummyClass,    # classobj
    chr,           # builtin_function_or_method (like abs, bool, chr, etc.)
    str, Exception, RepyException, # safe_type (like int, float, set, etc.)
    xrange(1),     # xrange
    slice(1),      # slice
    Ellipsis,      # ellipsis
    ]

As already demonstrated in the current namespace.py code, the types module provides appropriate objects for comparison for these types as well.

[aaaaalbert]

The corresponding patch, including some clean-up:

814,816c814,827
<     if _is_in(type(obj), [str, unicode, int, long, float, complex, bool, frozenset,
<                           types.NoneType, types.FunctionType, types.LambdaType,
<                           types.MethodType, types.InstanceType]):
---
>     if _is_in(type(obj), [
>         str, unicode, int, long, float, complex, bool, frozenset,
>         types.NoneType,     # for None
>         types.FunctionType, # for functions
>         types.BuiltinFunctionType, # for builtin functions like chr
>         types.LambdaType,   # XXX should not happen
>         types.MethodType,   # for methods of objects
>         types.InstanceType, # for instantiated objects
>         types.ClassType,    # for classes of objects
>         types.SliceType,    # for slices
>         types.XRangeType,   # for xrange
>         types.EllipsisType, # for Ellipsis
>         safe.safe_type,     # for wrapped types like str, RepyExcpetion
>         ]):

I don't understand why exactly we allow lambda here when safe.py actually forbids it (and there is no way to construct a lambda expression in RepyV2 thus).