[Tutor] importing into a function

[Steven D'Aprano]

On 11/07/13 05:02, Jim Mooney wrote:

> The question is - does importing into a function
> ever make sense? It  occurred to me that if you use    from module
> import * , putting it in a function protects you from invisible
> namespace pollution by a big module with a lot of names. But importing
> has to be a big overhead. So does importing into a function make sense
> at any time, or is it a dead end?

It's not encouraged, but it's not prohibited either.

Imports are very expensive the first time you do them, but after that they are cached and are quite fast. There's a long-term cache on disk (the .pyc file) and a short-term cache in memory (inside sys.modules). "import module" looks something like this pseudo-code:

if "module" in sys.modules:
     module = sys.modules["module"]
     return
for directory in sys.path:
     if there are both "module.py" and "module.pyc" files in
     the directory, and "module.pyc" is newer than "module.py",
     or there is only "module.pyc":
         module = load("module.pyc")
         break
     elif there is only a "module.py" file:
         module = compile("module.py")
         try:
             write file "module.pyc"  # on-disk cache
         except:
             pass
         break
if no module found:
     raise ImportError
sys.modules["modules"] = module  # memory cache cache
return


only the real code is much, much more complex. But the end result is that importing is only slow the first time.

Why might you put an import inside a function? The main reason is to delay a circular import. Suppose you have a module X that relies on Y, and Y relies on X. That's tricky to do: X can't import until Y is imported, and Y can't import until X is imported. But if X can delay importing Y until X's functions are called, it will work out fine.



-- 
Steven