PEP 8 exegetics: conditional imports?
Albert Hopkins
marduk at letterboxes.org
Fri Aug 7 14:18:51 EDT 2009
On Fri, 2009-08-07 at 16:50 +0000, kj wrote:
>
> Conditional imports make sense to me, as in the following example:
>
> def foobar(filename):
> if os.path.splitext(filename)[1] == '.gz':
> import gzip
> f = gzip.open(filename)
> else:
> f = file(filename)
> # etc.
>
> And yet, quoth PEP 8:
>
> - Imports are always put at the top of the file, just after any module
> comments and docstrings, and before module globals and constants.
>
> ...which seems to condemn conditional imports unequivocally.
>
> Then again, the PEP 8 scriptures do not explicitly mention conditional
> imports at all, as far as I can tell, which leaves open the
> possibility that they are still righteous.
>
> In fact, venerable passages in the Python standard library source
> code, if properly interpreted, can be seen to carry out conditional
> imports, such as this fragment recovered from random.py:
>
> if a is None:
> try:
> a = long(_hexlify(_urandom(16)), 16)
> except NotImplementedError:
> import time
> a = long(time.time() * 256) # use fractional seconds
>
> Or even more clearly, this one from test/pystone.py:
>
> if __name__ == '__main__':
> import sys
>
>
I can't speak for others... but generally you can "tell" when an import
belongs at the top of a module and when you need to make exceptions. I
would say that, as a general rule, they go up top (easier to identify
what external dependencies a module have). There are, of course,
exceptions:
* Exceptions and other not-often-executed blocks
* Inside test functions (where the imported modules are only used
by tests
* Inside main() functions where the imported modules are only used
if the module is run as a script
* When startup-time is important. Often it's necessary to have you
module up and running in an instant and import "expensive"
modules as-needed.
* "Dynamic" import such as plugins, etc.
Of course, like everything else in PEP 8, it's meant as a "guide" and
not as an "order". YMMV.
-a
More information about the Python-list
mailing list