Why is "import" implemented this way?

Question

>>> import math
>>> math.pi
3.141592653589793
>>> math.pi = 3
>>> math.pi
3
>>> import math
>>> math.pi
3

Initial question: Why can't I get math.pi back?

I thought import would import all the defined variables and functions to the current scope. And if a variable name already exists in current scope, then it would replace it.

Yes, it does replace it:

>>> pi = 3
>>> from math import *
>>> pi
3.141592653589793

Then I thought maybe the math.pi = 3 assignment actually changed the property in the math class(or is it math module?), which the import math imported.

I was right:

>>> import math
>>> math.pi
3.141592653589793
>>> math.pi = 3
>>> from math import *
>>> pi
3

So, it seems that:
If you do import x, then it imports x as a class-like thing. And if you make changes to x.property, the change would persist in the module so that every time you import it again, it's a modified version.

Real question:

1. Why is import implemented this way? Why not let every import math import a fresh, unmodified copy of math? Why leave the imported math open to change?
2. Is there any workaround to get math.pi back after doing math.pi = 3 (except math.pi = 3.141592653589793, of course)?
3. Originally I thought import math is preferred over from math import *. But this behaviour leaves me worrying someone else might be modifying my imported module if I do it this way...How should I do the import?

Answer 1

Python only creates one copy of any given module. Importing a module repeatedly reuses the original. This is because if modules A and B imported C and D, which imported E and F, etc., C and D would get loaded twice, and E and F would get loaded 4 times, etc. With any but the most trivial of dependency graphs, you'd spend a few minutes loading redundant modules before running out of memory. Also, if A imported B and B imported A, you'd get stuck in a recursive loop and, again, run out of memory without doing anything useful.

The solution: Don't screw with the contents of other modules. If you do, that's an interpreter-wide change. There are occasionally situations where you'd want to do this, so Python lets you, but it's usually a bad idea.