为什么在使用 >>= 运算符时会出现语法错误？

Question

Note: I am importing the parsec library , which overloads the >>= (ie - __irshift()__ ) operator .

The following Python code:

#! /usr/bin/env python

# irshift_debug.py

from parsec import *

def id(x):
    """Identity parser."""
    return string("").result(x)

testp = digit() >>= id

print(testp.parse("5"))

yields:

$ ./irshift_debug.py
  File "./irshift_debug.py", line 9
    testp = digit() >>= id
                    ^
SyntaxError: invalid syntax

But, if I change the offending line to:

testp = digit().bind(id)

then I get:

$ ./irshift_debug.py
5

as expected.

I thought that >>= was the infix equivalent of bind() ; is that incorrect?

Answer 1

Because >>= is a statement, not an expression. It doesn't evaluate to a result that can be assigned to testp .

Perhaps you were looking for this?

testp = digit >> id

Answer 2

I have used neither Parsec nor Haskell, so I can't speak to the intended use in the context of this library. However, I believe the main root of your confusion here is that augmented assignment works differently than ordinary operators, and therefore reimplementing them works slightly differently as well.

So you've seen this code:

def __irshift__(self, other):
    '''Implements the `(>>=)` operator, means `bind`.'''
    return self.bind(other)

What isn't immediately obvious is that, because this is overriding augmented assignment, not an operator, the returned value is used to reassign the original item on the left. In other words, this code:

a >>= b

is (essentially) equivalent to:

a = a.bind(b)

In either case, as said in Alexander's answer , the resulting assignment is not actually an expression, and ultimately returns no value (not even None ). The right half of an assignment has to be an expression; it can't be another assignment ¹ . Just as in Python you can't do things like:

a = b = c

So you also can't do:

a = b >>= c

...even when you've reimplemented >>= .

---

¹ The only partial exception is the recently added assignment expression , but it is (intentionally) limited in use, and won't help you here.

Answer 3

It is a design mistake.

I have turned to use >= as the operator for bind() and deprecated >>= . A new release of parsec.py has been upload to pypi.