Namespace pollution is the main argument against wildcard imports:

from foo import *

While explicit unqualified imports are considered fine:

from foo import bar

There is, however, a harmful side-effect of unqualified imports:

from sys import stdout

stdout = open('stdout.txt', 'w')
print('hello')
stdout.close()
assert open('stdout.txt').read() == 'hello\n'

The assertion fails, because the output from print does not go to stdout rather it goes to the value of sys.stdout . Unqualified imports in other languages do not work this way, that is, the qualified and unqualified names are exactly the same thing.

For qualified imports, the attribute’s value is retrieved on every use so the following example works as you would expect:

import sys

sys.stdout = open('stdout.txt', 'w')
print('hello')
sys.stdout.close()
assert open('stdout.txt').read() == 'hello\n'

Python imports values, not names. The interpreter makes a new variable in the importing module with the reference of the variable from the exporting module. When the first example assigns to stdout , it assigns the newly opened file to the importing module’s variable, which happens to be called stdout . The as keyword makes this behavior explicit.

Here’s the first example rewritten with as to show this effect:

from sys import stdout as new_stdout

new_stdout = open('stdout.txt', 'w')
print('hello')
new_stdout.close()
assert open('stdout.txt').read() == 'hello\n'

While this is not the typical form of using the as keyword, it shows more clearly that new_stdout is a name defined in the importing module. The name sys.stdout is easily seen to be unrelated by this atypical usage of the as keyword.

Python also creates a new variable for a module import. All of these names are all “the same”, that is, they are attributes, and you can do strange things like this:

import sys

sys = 1
assert sys == 1

The assertion passes, because the reference to the module sys is replaced by the reference to the number 1 by the assignment. The code on line three does not reassign the meaning of sys globally, that is, another module which imports sys will get a copy of the module reference, not a reference to the number 1 .

from impedes reload()

To reload a module, you have to pass the reference of the module to the reload built-in function. Reloading a module has to reuse the exact same reference, because every import makes a copy of the reference in the importing module.

The subtle thing about reloading is that any attributes of the module which are imported unqualified (using from ) do not change. The following complex example demonstrates this problem:

import os
import sys

def write_m1(val):
    f = open('m1.py', 'w')
    f.write('''
def f1():
    return {}
'''.format(val))
    f.close()
    os.unlink('m1.pyc')

write_m1('1')
from m1 import f1
assert f1() == 1

write_m1('2')
reload(sys.modules['m1'])
assert f1() == 2

The assertion fails on line 20, because f1() still returns one, not two. A qualified call m1.f1() would return two.

Reloading modules is very useful for large systems where reloading a single module is orders of magnitude faster than reloading the whole system. You can find references to this problem, but it’s not widely known. People argue against from for stylistic reasons (which I strongly agree with). I go further and say that from is harmful, and should be deprecated.