skip to navigation
skip to content

Python Wiki

Python Insider Blog

Python 2 or 3?

Help Fund Python

[Python resources in languages other than English]

Non-English Resources

Add an event to this calendar.

Times are shown in UTC/GMT.

Add an event to this calendar.

PEP:289
Title:Generator Expressions
Version:d3493191e8c8
Last-Modified:2009-01-18 09:47:54 +0000 (Sun, 18 Jan 2009)
Author:python at rcn.com (Raymond Hettinger)
Status:Final
Type:Standards Track
Content-Type:text/x-rst
Created:30-Jan-2002
Python-Version:2.4
Post-History:22-Oct-2003

Abstract

This PEP introduces generator expressions as a high performance, memory efficient generalization of list comprehensions [1] and generators [2].

Rationale

Experience with list comprehensions has shown their wide-spread utility throughout Python. However, many of the use cases do not need to have a full list created in memory. Instead, they only need to iterate over the elements one at a time.

For instance, the following summation code will build a full list of squares in memory, iterate over those values, and, when the reference is no longer needed, delete the list:

sum([x*x for x in range(10)])

Memory is conserved by using a generator expression instead:

sum(x*x for x in range(10))

Similar benefits are conferred on constructors for container objects:

s = Set(word  for line in page  for word in line.split())
d = dict( (k, func(k)) for k in keylist)

Generator expressions are especially useful with functions like sum(), min(), and max() that reduce an iterable input to a single value:

max(len(line)  for line in file  if line.strip())

Generator expressions also address some examples of functionals coded with lambda:

reduce(lambda s, a: s + a.myattr, data, 0)
reduce(lambda s, a: s + a[3], data, 0)

These simplify to:

sum(a.myattr for a in data)
sum(a[3] for a in data)

List comprehensions greatly reduced the need for filter() and map(). Likewise, generator expressions are expected to minimize the need for itertools.ifilter() and itertools.imap(). In contrast, the utility of other itertools will be enhanced by generator expressions:

dotproduct = sum(x*y for x,y in itertools.izip(x_vector, y_vector))

Having a syntax similar to list comprehensions also makes it easy to convert existing code into an generator expression when scaling up application.

Early timings showed that generators had a significant performance advantage over list comprehensions. However, the latter were highly optimized for Py2.4 and now the performance is roughly comparable for small to mid-sized data sets. As the data volumes grow larger, generator expressions tend to perform better because they do not exhaust cache memory and they allow Python to re-use objects between iterations.

BDFL Pronouncements

This PEP is ACCEPTED for Py2.4.

The Details

(None of this is exact enough in the eye of a reader from Mars, but I hope the examples convey the intention well enough for a discussion in c.l.py. The Python Reference Manual should contain a 100% exact semantic and syntactic specification.)

  1. The semantics of a generator expression are equivalent to creating an anonymous generator function and calling it. For example:

    g = (x**2 for x in range(10))
    print g.next()
    

    is equivalent to:

    def __gen(exp):
        for x in exp:
            yield x**2
    g = __gen(iter(range(10)))
    print g.next()
    

    Only the outermost for-expression is evaluated immediately, the other expressions are deferred until the generator is run:

    g = (tgtexp  for var1 in exp1 if exp2 for var2 in exp3 if exp4)
    

    is equivalent to:

    def __gen(bound_exp):
        for var1 in bound_exp:
            if exp2:
                for var2 in exp3:
                    if exp4:
                        yield tgtexp
    g = __gen(iter(exp1))
    del __gen
    
  2. The syntax requires that a generator expression always needs to be directly inside a set of parentheses and cannot have a comma on either side. With reference to the file Grammar/Grammar in CVS, two rules change:

    1. The rule:

      atom: '(' [testlist] ')'
      

      changes to:

      atom: '(' [testlist_gexp] ')'
      

      where testlist_gexp is almost the same as listmaker, but only allows a single test after 'for' ... 'in':

      testlist_gexp: test ( gen_for | (',' test)* [','] )
      
    2. The rule for arglist needs similar changes.

    This means that you can write:

    sum(x**2 for x in range(10))
    

    but you would have to write:

    reduce(operator.add, (x**2 for x in range(10)))
    

    and also:

    g = (x**2 for x in range(10))
    

    i.e. if a function call has a single positional argument, it can be a generator expression without extra parentheses, but in all other cases you have to parenthesize it.

    The exact details were checked in to Grammar/Grammar version 1.49.

  3. The loop variable (if it is a simple variable or a tuple of simple variables) is not exposed to the surrounding function. This facilitates the implementation and makes typical use cases more reliable. In some future version of Python, list comprehensions will also hide the induction variable from the surrounding code (and, in Py2.4, warnings will be issued for code accessing the induction variable).

    For example:

    x = "hello"
    y = list(x for x in "abc")
    print x    # prints "hello", not "c"
    
  4. List comprehensions will remain unchanged. For example:

    [x for x in S]    # This is a list comprehension.
    [(x for x in S)]  # This is a list containing one generator
                      # expression.
    

    Unfortunately, there is currently a slight syntactic difference. The expression:

    [x for x in 1, 2, 3]
    

    is legal, meaning:

    [x for x in (1, 2, 3)]
    

    But generator expressions will not allow the former version:

    (x for x in 1, 2, 3)
    

    is illegal.

    The former list comprehension syntax will become illegal in Python 3.0, and should be deprecated in Python 2.4 and beyond.

    List comprehensions also "leak" their loop variable into the surrounding scope. This will also change in Python 3.0, so that the semantic definition of a list comprehension in Python 3.0 will be equivalent to list(<generator expression>). Python 2.4 and beyond should issue a deprecation warning if a list comprehension's loop variable has the same name as a variable used in the immediately surrounding scope.

Early Binding versus Late Binding

After much discussion, it was decided that the first (outermost) for-expression should be evaluated immediately and that the remaining expressions be evaluated when the generator is executed.

Asked to summarize the reasoning for binding the first expression, Guido offered [5]:

Consider sum(x for x in foo()). Now suppose there's a bug in foo()
that raises an exception, and a bug in sum() that raises an
exception before it starts iterating over its argument. Which
exception would you expect to see? I'd be surprised if the one in
sum() was raised rather the one in foo(), since the call to foo()
is part of the argument to sum(), and I expect arguments to be
processed before the function is called.

OTOH, in sum(bar(x) for x in foo()), where sum() and foo()
are bugfree, but bar() raises an exception, we have no choice but
to delay the call to bar() until sum() starts iterating -- that's
part of the contract of generators. (They do nothing until their
next() method is first called.)

Various use cases were proposed for binding all free variables when the generator is defined. And some proponents felt that the resulting expressions would be easier to understand and debug if bound immediately.

However, Python takes a late binding approach to lambda expressions and has no precedent for automatic, early binding. It was felt that introducing a new paradigm would unnecessarily introduce complexity.

After exploring many possibilities, a consensus emerged that binding issues were hard to understand and that users should be strongly encouraged to use generator expressions inside functions that consume their arguments immediately. For more complex applications, full generator definitions are always superior in terms of being obvious about scope, lifetime, and binding [6].

Reduction Functions

The utility of generator expressions is greatly enhanced when combined with reduction functions like sum(), min(), and max(). The heapq module in Python 2.4 includes two new reduction functions: nlargest() and nsmallest(). Both work well with generator expressions and keep no more than n items in memory at one time.

Acknowledgements

  • Raymond Hettinger first proposed the idea of "generator comprehensions" in January 2002.
  • Peter Norvig resurrected the discussion in his proposal for Accumulation Displays.
  • Alex Martelli provided critical measurements that proved the performance benefits of generator expressions. He also provided strong arguments that they were a desirable thing to have.
  • Phillip Eby suggested "iterator expressions" as the name.
  • Subsequently, Tim Peters suggested the name "generator expressions".
  • Armin Rigo, Tim Peters, Guido van Rossum, Samuele Pedroni, Hye-Shik Chang and Raymond Hettinger teased out the issues surrounding early versus late binding [5].
  • Jiwon Seo single handedly implemented various versions of the proposal including the final version loaded into CVS. Along the way, there were periodic code reviews by Hye-Shik Chang and Raymond Hettinger. Guido van Rossum made the key design decisions after comments from Armin Rigo and newsgroup discussions. Raymond Hettinger provided the test suite, documentation, tutorial, and examples [6].

References

[1]PEP 202 List Comprehensions http://www.python.org/dev/peps/pep-0202/
[2]PEP 255 Simple Generators http://www.python.org/dev/peps/pep-0255/
[3]Peter Norvig's Accumulation Display Proposal http://www.norvig.com/pyacc.html
[4]Jeff Epler had worked up a patch demonstrating the previously proposed bracket and yield syntax http://python.org/sf/795947
[5](1, 2) Discussion over the relative merits of early versus late binding http://mail.python.org/pipermail/python-dev/2004-April/044555.html
[6](1, 2) Patch discussion and alternative patches on Source Forge http://www.python.org/sf/872326