Scoping and use of the 'global' keyword in Python
Here’s something I knocked up for one of my co-workers who’s an experienced developer, but a Python neophyte. It may be useful to others.
As in PHP, the ‘global’ keyword in Python declares that any reference to a given variable name within the scope of a function or method refers to a variable in module/global scope. However, the scoping behaviour of the two languages differs subtly.
In PHP, variables within a function or method are local unless declared as global. With Python, the behaviour is more subtle. Firstly, let’s take a simple case:
foo = 0 def bar(): print "In bar(), foo is", foo bar() print "And in module scope, foo is", foo
This will result in:
In bar(), foo is 0 And in module scope, foo is 0
However, if we try to modify foo without declaring it global:
foo = 0 def bar(): print "Entering bar(), foo is", foo foo += 1 print "Leaving bar(), foo is", foo bar() print "And in module scope, foo is", foo
You’ll get an error:
Entering bar(), foo is Traceback (most recent call last): File "globaltest.py", line 8, in
bar() File "globaltest.py", line 4, in bar print "Entering bar(), foo is", foo UnboundLocalError: local variable 'foo' referenced before assignment
This is because module/global scope is a read-only shadow within the local function/method scope: you can access it, providing there’s nothing declared with the same name in local scope, but you can’t change it. There are very good, though non-trivial1, reasons for this behaviour.
If you specifically want to be able to change the value a given variable in the global scope contains, then use the ‘global’ keyword.
foo = 0 def bar(): global foo print "Entering bar(), foo is", foo foo += 1 print "Leaving bar(), foo is", foo bar() print "And in module scope, foo is", foo
Entering bar(), foo is 0 Leaving bar(), foo is 1 And in module scope, foo is 1
However, if all you want to do is access methods on the object the variable contains a reference to, ‘global’ isn’t needed, and should be avoided.
In a nutshell, functions types and modules are just regular objects, no different from any other object, and when you declare a function or class, or import a module, you’re just using some syntactic sugar around a variable assignment. If Python behaved like PHP, you’d need to do stuff like this:
def boogie(): print "I'm dancing, I'm dancing!" def claptrap(): # 'boogie' is a variable that contains a reference to the function # object declared above. global boogie print "Come on everybody check me out." boogie() claptrap()
Also, Python are simply pigeonholes for holding references to objects; variables are not the objects they contain and everything, even number and strings, are objects. Variables containing immutable objects like numbers and strings may appear to act otherwise, but this is simply because those types of object are immutable. ↩