There are essentially three ways to use the with statement:
Use an existing context manager:
with manager:
pass
Create a contex
If the context manager is a class and only ever has a single instance, then you could find it on the heap:
import gc
class ConMan(object):
def __init__(self, name):
self.name = name
def __enter__(self):
print "enter %s" % self.name
def found(self):
print "You found %s!" % self.name
def __exit__(self, *args):
print "exit %s" % self.name
def find_single(typ):
single = None
for obj in gc.get_objects():
if isinstance(obj, typ):
if single is not None:
raise ValueError("Found more than one")
single = obj
return single
def foo():
conman = find_single(ConMan)
conman.found()
with ConMan('the-context-manager'):
foo()
(Disclaimer: Don't do this)
The difference between this case and similar-appearing cases like super
is that here there is no enclosing frame to look at. A with
statement is not a new scope. sys._getframe(0)
(or, if you're putting the code into a function, sys._getframe(1)
) will work just fine, but it'll return you the exact same frame you have before and after the with
statement.
The only way you could do it would be by inspecting the bytecode. But even that won't help. For example, try this:
from contextlib import contextmanager
@contextmanager
def silly():
yield
with silly():
fr = sys._getframe(0)
dis.dis(fr.f_code)
Obviously, as SETUP_WITH explains, the method does get looked up and pushed onto the stack for WITH_CLEANUP to use later. So, even after POP_TOP
removes the return value of silly()
, its __exit__
is still on the stack.
But there's no way to get at that from Python. Unless you want to start munging the bytecode, or digging apart the stack with ctypes
or something, it might as well not exist.
Unfortunately, as discussed in the comments, this is not possible in all cases. When a context manager is created, the following code is run (in cPython 2.7, at least. I can't comment on other implementations):
case SETUP_WITH:
{
static PyObject *exit, *enter;
w = TOP();
x = special_lookup(w, "__exit__", &exit);
if (!x)
break;
SET_TOP(x);
/* more code follows... */
}
The __exit__
method is pushed onto a stack with the SET_TOP
macro, which is defined as:
#define SET_TOP(v) (stack_pointer[-1] = (v))
The stack pointer, in turn, is set to the top of the frame's value stack at the start of frame eval:
stack_pointer = f->f_stacktop;
Where f is a frame object defined in frameobject.h. Unfortunately for us, this is where the trail stops. The python accessible frame object is defined with the following methods only:
static PyMemberDef frame_memberlist[] = {
{"f_back", T_OBJECT, OFF(f_back), RO},
{"f_code", T_OBJECT, OFF(f_code), RO},
{"f_builtins", T_OBJECT, OFF(f_builtins),RO},
{"f_globals", T_OBJECT, OFF(f_globals), RO},
{"f_lasti", T_INT, OFF(f_lasti), RO},
{NULL} /* Sentinel */
};
Which, unfortunaltey, does not include the f_valuestack
that we would need. This makes sense, since f_valuestack
is of the type PyObject **
, which would need to be wrapped in an object to be accessible from python any way.
TL;DR: The __exit__
method we're looking for is only located in one place, the value stack of a frame object, and cPython doesn't make the value stack accessible to python code.