Why do directly imported variables behave differently than thoseattached to imported module?

[Dun Peal]

Hi!

Here's the demonstrating code:

# module foo.py
var = 0

def set():
global var
var = 1

Script using this module:

import foo
from foo import *

print var, foo.var
set()
print var, foo.var

Script output:

0 0
0 1

Apparently, the `var` we imported from `foo` never got set, but
`foo.var` on the imported `foo` - did. Why?

Thanks, D.

 

[Daniel Kluev]

Apparently, the `var` we imported from `foo` never got set, but
`foo.var` on the imported `foo` - did. Why?

Because all names are references to some values, not other names (in
CPython, it means all names are PyObject*, and point directly to the
objects, not other pointers)

When you do `from foo import bar` it assigns globals()['bar'] of
current module to reference same value as `foo.bar`. Its now local
namespace name, not `foo` namespace, and therefore functions in `foo`
cannot modify this namespace.

Since ints are immutable, when you do `var = 1` you create new object
of type int, and re-assign `var` name to point to new object.

`foo.var`, on other hand, is a way to access `foo`'s own namespace, so
its exactly same name as globals()['var'] of `foo`.

 

[Chris Rebert]

Hi!

Here's the demonstrating code:

   # module foo.py
   var = 0

   def set():
       global var
       var = 1

Script using this module:

   import foo
   from foo import *

   print var, foo.var
   set()
   print var, foo.var

Script output:

   0 0
   0 1

Apparently, the `var` we imported from `foo` never got set, but
`foo.var` on the imported `foo` - did. Why?

Because imports (and assignments generally) bind names to values, they
don't alias names to other names.

from foo import *

can be thought of as essentially doing:

import foo
set = foo.set
var = foo.var
del foo

So the new, separate name __main__.var gets the current value of
foo.var at import-time, which is the integer 0.
You then call foo.set(), which re-binds foo.var to a new value (i.e.
1) rather than mutating the existing value (which would be impossible
anyway since integers are immutable). This has absolutely no effect on
__main__.var, which is an entirely separate binding.

The behavior is comparable to that of function arguments. Values can
be mutated, but re-binding names has only local effect:.... b = 1 # rebinds local name b
....

incr(a)
a # outside name unaffected, just like in your example 0
c = [7]
def mutate_then_rebind(b):

.... b.append(99) # mutates passed-in value
.... b = [42] # rebinds local name; has no outside effect
....[7, 99]

Cheers,
Chris

 

[Mel]

Hi!

Here's the demonstrating code:

# module foo.py
var = 0

def set():
global var
var = 1

Script using this module:

import foo
from foo import *

print var, foo.var
set()
print var, foo.var

Script output:

0 0
0 1

Apparently, the `var` we imported from `foo` never got set, but
`foo.var` on the imported `foo` - did. Why?

They're different because -- they're different. `foo.var` is defined in the
namespace of the foo module. Introspectively, you would access it as
`foo.__dict__['var']` .

Plain `var` is in your script's namespace so you could access it as
`globals()['var']` . The values given to the vars are immutable integers,
so assignment works by rebinding. The two different bindings in
foo.__dict__ and globals() get bound to different integer objects.

Note too the possible use of `globals()['foo'].__dict__['var'] . (Hope
there are no typos in this post.)

Mel.

 

[Terry Reedy]

Your problem is reveal in the subject line. As discussed in many other
threads, including a current one, Python does not have 'variables' in
the way that many understand the term. Python binds names to objects.
Binding statements (assignment, augmented assignment, import, def,
class, and others with 'as' clauses) all bind names to objects.

Suppose we both have address books. You copy all entries from my book to
yours. Afterwards, I update an entry in my book. That does not change
the now obsolete entry in your book.

# module foo.py
var = 0

def set():
global var
var = 1

Module foo's dict now has two entries for 'var' and 'set' bound to int 0
and a function. Note that 'set' is a builtin class name so not the best
choice.

Script using this module:

Script creates module '__main__'

import foo

This binds 'foo' to module foo in __main_'s dict.

from foo import *

This copies foo's dict to __main__'s dict.

print var, foo.var
set()

This changes foo's dict, rebinding 'var' to int 1. It does not change
__main__'s dict. How could it? In the same way, 'var = 2' would leave
foo's dict unchanged. __main__.var.

print var, foo.var

Script output:

0 0
0 1

If instead of 'from foo import *', you had written 'from foo import var
as rav', then perhaps you would not have been so surprised that
__main__.rav and foo.var have independent fates.

You have discovered that using from x import * is a bad idea when module
x has global names that get rebound.

 

[harrismh777]

# module foo.py
var = 0

def set():
global var
var = 1

My two cents to add in addition to the correct accounts of [ Daniel,
Chris, Mel] is that you might have a misunderstanding of var, generally.

Python has no variables--- in the way you think of them in other
languages. Python only has references.

Var is not 'set' in the classical sense of the word...

var is a 'reference' (think C pointer, sort-of) to an object... '0'
and '1' are objects of type int and

var=0 means create a reference to object int('0') called var.

a=b=0 a is a reference to object '0'
b is a reference to object '0'

a does not reference b

This rest of the discussion has to do with name spaces:

See:

http://docs.python.org/release/3.0.1/reference/datamodel.html?highlight=namespaces

 

[Dun Peal]

OK, I understand now.

`from foo import var` means "create a module-global name `var` inside
the current module, and have it point at the object `foo.var` is
pointing at (following its evaluation)".

Naturally, regardless of whether `foo.var` ever changes, the global
`var` of the current module still points at the original object
`foo.var` was pointing to at the time of the `var = foo.var`
assignment.

Thanks, D.

 

[Dun Peal]

P.S. now I have to ask: is there a symbolic reference in Python, i.e.
a name foo that points to "whatever bar.baz is pointing at"?

Thanks, D.

 

[Daniel Kluev]

P.S. now I have to ask: is there a symbolic reference in Python, i.e.
a name foo that points to "whatever bar.baz is pointing at"?

Well, you could easily simulate that with proxy object,
class SymbolicReference(object):
def __init__(self, ns, name):
self.__ns = ns
self.__name = name

def __get(self):
return self.__ns[self.__name]

def __getattribute__(self, attr):
try:
return object.__getattribute__(self, attr)
except:
return self.__get().__getattribute__(attr)

def __repr__(self):
return self.__get().__repr__()

def __str__(self):
return self.__get().__str__()
10

 

[Chris Angelico]

from foo import *

can be thought of as essentially doing:

import foo
set = foo.set
var = foo.var
del foo

Here's a side point. What types will hold a reference to the enclosing
module (or at least its dictionary)? Would it be possible to use a
from import to load a module, then "lose" the module even though
you're using objects from it?

I am guessing that a function or class will hold such a reference,
because otherwise it would be a bit awkward for them to use any sort
of module-level state. Or do they not, and instead go look for their
module in sys.modules?

Chris Angelico

 

[Chris Rebert]

Here's a side point. What types will hold a reference to the enclosing
module (or at least its dictionary)? Would it be possible to use a
from import to load a module, then "lose" the module even though
you're using objects from it?

I am guessing that a function or class will hold such a reference,
because otherwise it would be a bit awkward for them to use any sort
of module-level state. Or do they not, and instead go look for their
module in sys.modules?

some_python_function.__globals__ is the exact reference you postulated.
Classes don't seem to hold such references; their methods do though
(by virtue of being function objects), which makes sense.

Cheers,
Chris