如何证明参数评估在Python中“从左到右”？

Question

For example, in JavaScript we could write a program like this:

var a = 1;
testFunction(++a, ++a, a);
function testFunction(x, y, z){
      document.writeln("<br />x = " + x);
      document.writeln("<br />y = " + y);
      document.writeln("<br />z = " + z);
}

and we would get an output:

x = 2
y = 3
z = 3

This implies that parameters are truly evaluated from left to right in JavaScript. In C we would get output

x = 3
y = 3
z = 3

I was wondering if we could do the same in Python or is it impossible since it's a pass by value reference language?

I've made a simple program but I don't think that proves anything:

x = 2
def f(x, y, z):
    print(x, y, z)

f(x*2, x*2, x**2)
print(x)

4 4 4
2

Python won't let me do any new assignment within the function parameter when I call it (for example f(x=4, x, x) or something like this).

Answer 1

>>> def f(x, y): pass
...
>>> f(print(1), print(2))
1
2

Answer 2

Disassemble the function call.

>>> def foo():
...   bar(x+1, x+2, x+3)
... 
>>> dis.dis(foo)
  2           0 LOAD_GLOBAL              0 (bar)
              3 LOAD_GLOBAL              1 (x)
              6 LOAD_CONST               1 (1)
              9 BINARY_ADD          
             10 LOAD_GLOBAL              1 (x)
             13 LOAD_CONST               2 (2)
             16 BINARY_ADD          
             17 LOAD_GLOBAL              1 (x)
             20 LOAD_CONST               3 (3)
             23 BINARY_ADD          
             24 CALL_FUNCTION            3
             27 POP_TOP             
             28 LOAD_CONST               0 (None)
             31 RETURN_VALUE        

Answer 3

Using Python 3:

>>> a = []
>>> f = print(
    a.append(1), a[:],
    a.append(2), a[:],
    a.append(3), a[:]
)
None [1] None [1, 2] None [1, 2, 3]

---

Archive:

>>> a = []
>>> f = print(a.append(1), a, a.append(2), a, a.append(3), a)

Curiously enough (at first), this code produces:

None [1, 2, 3] None [1, 2, 3] None [1, 2, 3]

However, dis(f) makes this clearer:

>>> dis(f)

  1           0 LOAD_NAME                0 (print) #Loads the value of 'print' into memory. Precisely, the value is pushed to the TOS (Top of Stack)
    -->       3 LOAD_NAME                1 (a) #Loads the value of object 'a' 
              6 LOAD_ATTR                2 (append) #Loads the append attribute (in this case method)
              9 LOAD_CONST               0 (1) #Loads the constant 1
             12 CALL_FUNCTION            1 #a.append(1) is called
             15 LOAD_NAME                1 (a) #for print(...,a,...)
             18 LOAD_NAME                1 (a) #for the next a.append()
             21 LOAD_ATTR                2 (append) 
             24 LOAD_CONST               1 (2) 
             27 CALL_FUNCTION            1 #a.append(2)
             30 LOAD_NAME                1 (a) 
             33 LOAD_NAME                1 (a) 
             36 LOAD_ATTR                2 (append) 
             39 LOAD_CONST               2 (3) 
             42 CALL_FUNCTION            1 #a.append(3)
             45 LOAD_NAME                1 (a) #loads a to be used thrice by print
             48 CALL_FUNCTION            6 #calls print
             51 PRINT_EXPR                 #prints TOS and clears it
             52 LOAD_CONST               3 (None) #Loads None
             55 RETURN_VALUE             #Returns None

The output of dis(f) is what we expected - L-to-R evaluation. Essentially, this "discrepancy" is a consequence of print() being evaluated last. By then, the value of a has changed to [1, 2, 3] and the same final object is printed thrice.

If we replace a with a[:] , we get the expected result.

Answer 4

This shows it as well IMHO:

>>> '{} {} {}'.format(x,x+1,x+2)
'1 2 3'

Edit:

>>> def f(t):   return time.time()-t 
...         

>>> t1=time.time(); '{:.4} {:.4} {:.4}'.format(f(t1),f(t1),f(t1))
'5.007e-06 7.868e-06 9.06e-06'

Answer 5

A custom class can help here:

class Tester(object):
    "test object to reveal left to right evaluation"
    def __init__(self, value):
        self.value = value
    def __add__(self, value):
        print("adding ", value)
        return Tester(self.value + value)
    def __repr__(self):
        return repr(self.value)

and when run:

--> t = Tester(7)
--> t
7
--> t = t + 7
adding  7
--> t
14
--> def blah(a, b, c):
...   print(a, b, c)
... 
--> blah(t+1, t+2, t+3)
adding  1
adding  2
adding  3
15 16 17

Answer 6

Short answer: left to right

Example: Since this is a question that always pops in my head when I am doing arithmetic operations (should I convert to float and which number), an example from that aspect is presented:

>>> a = 1/2/3/4/5/4/3
>>> a
0

When we divide integers, not surprisingly it gets lower rounded.

>>> a = 1/2/3/4/5/4/float(3)
>>> a
0.0

If we typecast the last integer to float, we will still get zero, since by the time our number gets divided by the float has already become 0 because of the integer division.

>>> a = 1/2/3/float(4)/5/4/3
>>> a
0.0

Same scenario as above but shifting the float typecast a little closer to the left side.

>>> a = float(1)/2/3/4/5/4/3
>>> a
0.0006944444444444445

Finally, when we typecast the first integer to float, the result is the desired one, since beginning from the first division, ie the leftmost one, we use floats.

Extra 1: If you are trying to answer that to improve arithmetic evaluation, you should check this

Extra 2: Please be careful of the following scenario:

>>> a = float(1/2/3/4/5/4/3)
>>> a
0.0