Backports the Python 3.7 range class as a replacement for python 2.X range
functions (and the Python pre-3.7 range class).
The range class is similar to xrange in that its values are computed on demand -
however, the range class is also a lazy sequence:
it supports indexing, membership testing and other sequence features.
Thus, it can be used to replace both range and xrange.
Table of Contents
This implementation provides all features introduced and documented in
python 3.6, and implements the collections.abc.Sequence interface of Python 3.5 [1] and later.
The range class is available as backports.range.range -
you can import it to a separate name or replace the builtin range and/or xrange.
# import for explicit usage
from backports.range import range as range3
values = range3(499, 501)
# as a lazy, reusable iterable
print(', '.join(values), 'and', ', '.join(values))
# as a lazy sequence
print(500 in values, 501 in values)
print(values[2], len(values))
All objects are available by default in a pure python implementation. In addition, an optional, optimized implementation is available using Cython.
The backports.range is adequate in performance for most applications.
It is comparable in iteration speed to builtins when using PyPy or Cython.
For small ranges of less than 1000 elements, there is some small overhead.
This should not be noticeable in all but the most high-performance applications.
When using CPython (aka plain python), pure python mode is considerably slower than the builtins.
Again, this should not matter for most applications, but the use of Cython is strongly advised.
| Interpreter | vs Builtin range | vs Builtin xrange |
|---|---|---|
| Py2 | 50 - 100 | 20 - 50 |
| Py3 | 25 - 30 | --- |
| Py2 + Cython | 3 - 6 | 1.0 - 3 |
| Py3 + Cython | 1.0 - 1.2 | --- |
| PyPy2 | 1.4 - 1.6 | 1.6 - 2.3 |
| PyPy3 | 1.0 - 1.1 | --- |
The backport features a Cython implementation.
It is transparently used when creating a range class.
It optimises operations which are purely in the C long long range. [2]
All Cython optimizations are optional. They are automatically made available if Cython is installed.
- Features are tested against the Python 3.6 unittests for
range. - The following python versions are tested explicitly:
- Some additional features are available for compatibility:
- Instances of
backports.range.rangecompare equal to equivalentbuiltin.rangeinstances (new in Python 3.3) - The
indexmethod is compliant with the Python 3.5+ specification ofcollections.abc.Sequence. [1]
- Instances of
- Some features depending on language features or other modules may not be
available or work differently:
- Comparing
rangeagainst other types does not throwTypeErrorin python 2.X. - Python implementations use custom optimisations at compile time when provably encountering a builtin range. These cannot be replicated by the backport. Note that this is only noticeable in highly optimised code, which should be hand-optimised anyway.
- Comparing
For all practical purposes, the backport is complete.
The range class satisfies all unittest of Python 3.7, as well as several tests for compatibility.
It is a fully featured replacement suitable even for recent versions of Python.
Performance should be more than adequate for any use case. The backport is comparable or even equal in performance to builtins.
If you encounter any problems, please head over to the issue tracker. Feel free to submit a pull request if you think the codebase can be improved.
This packages includes parts of the python documentation (http://www.python.org) Copyright (C) 2001-2016 Python Software Foundation. Licensed under the Python Software Foundation License.
| [1] | (1, 2) As of Python 3.6, the builtin range class is not compliant
with the Python 3.5 specification of collections.abc.Sequence.
See Issue 28197 |
| [2] | The long long type should fill any current CPU architecture (read 64bit).
Since Python cannot parallelize instructions,
this should provide maximum range with optimal performance for small values. |