This PEP proposes to strip the xrange() object from some rarely
used behavior like x[i:j] and x*n.
The xrange() function has one idiomatic use:
for i in xrange(...): ...
However, the xrange() object has a bunch of rarely used behaviors
that attempt to make it more sequence-like. These are so rarely
used that historically they have has serious bugs (e.g. off-by-one
errors) that went undetected for several releases.
I claim that it's better to drop these unused features. This will
simplify the implementation, testing, and documentation, and
reduce maintenance and code size.
I propose to strip the xrange() object to the bare minimum. The
only retained sequence behaviors are x[i], len(x), and repr(x).
In particular, these behaviors will be dropped:
x*n, n*x (sequence-repeat)
cmp(x1, x2) (comparisons)
i in x (containment test)
x.start, x.stop, x.step attributes
I also propose to change the signature of the PyRange_New() C API
to remove the 4th argument (the repetition count).
By implementing a custom iterator type, we could speed up the
common use, but this is optional (the default sequence iterator
does just fine).
This PEP affects the xrange() built-in function and the
PyRange_New() C API.
Somebody's code could be relying on the extended code, and this
code would break. However, given that historically bugs in the
extended code have gone undetected for so long, it's unlikely that
much code is affected.
For backwards compatibility, the existing functionality will still
be present in Python 2.2, but will trigger a warning. A year
after Python 2.2 final is released (probably in 2.4) the
functionality will be ripped out.
This document has been placed in the public domain.