Question

In the following code, why doesn’t Python compile f2 to the same bytecode as f1?

Is there a reason not to?

>>> def f1(x):
    x*100

>>> dis.dis(f1)
  2           0 LOAD_FAST                0 (x)
              3 LOAD_CONST               1 (100)
              6 BINARY_MULTIPLY
              7 POP_TOP
              8 LOAD_CONST               0 (None)
             11 RETURN_VALUE
>>> def f2(x):
        x*10*10

>>> dis.dis(f2)
  2           0 LOAD_FAST                0 (x)
              3 LOAD_CONST               1 (10)
              6 BINARY_MULTIPLY
              7 LOAD_CONST               1 (10)
             10 BINARY_MULTIPLY
             11 POP_TOP
             12 LOAD_CONST               0 (None)
             15 RETURN_VALUE

Answer 1

This is because x could have a __mul__ method with side-effects. x * 10 * 10 calls __mul__ twice, while x * 100 only calls it once:

>>> class Foo(object):
...     def __init__ (self):
...             self.val = 5
...     def __mul__ (self, other):
...             print "Called __mul__: %s" % (other)
...             self.val = self.val * other
...             return self
... 
>>> a = Foo()
>>> a * 10 * 10
Called __mul__: 10
Called __mul__: 10
<__main__.Foo object at 0x1017c4990>

Automatically folding the constants and only calling __mul__ once could change behavior.

You can get the optimization you want by reordering the operation such that the constants are multiplied first (or, as mentioned in the comments, using parentheses to group them such that they are merely operated on together, regardless of position), thus making explicit your desire for the folding to happen:

>>> def f1(x):
...     return 10 * 10 * x
... 
>>> dis.dis(f1)
  2           0 LOAD_CONST               2 (100)
              3 LOAD_FAST                0 (x)
              6 BINARY_MULTIPLY     
              7 RETURN_VALUE