add array.zeroes

[alex]

BPO	12615
Nosy	@rhettinger, @ncoghlan, @alex

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = 'https://github.com/rhettinger'
closed_at = <Date 2011-07-26.10:23:03.941>
created_at = <Date 2011-07-22.18:35:11.527>
labels = ['type-feature', 'library']
title = 'add array.zeroes'
updated_at = <Date 2011-07-26.10:23:03.939>
user = 'https://github.com/alex'

bugs.python.org fields:

activity = <Date 2011-07-26.10:23:03.939>
actor = 'rhettinger'
assignee = 'rhettinger'
closed = True
closed_date = <Date 2011-07-26.10:23:03.941>
closer = 'rhettinger'
components = ['Library (Lib)']
creation = <Date 2011-07-22.18:35:11.527>
creator = 'alex'
dependencies = []
files = []
hgrepos = []
issue_num = 12615
keywords = []
message_count = 4.0
messages = ['140901', '140905', '141133', '141139']
nosy_count = 4.0
nosy_names = ['rhettinger', 'ncoghlan', 'alex', 'santoso.wijaya']
pr_nums = []
priority = 'normal'
resolution = 'rejected'
stage = None
status = 'closed'
superseder = None
type = 'enhancement'
url = 'https://bugs.python.org/issue12615'
versions = ['Python 3.3']

[alex]

From python-ideas:

introduce array.zeroes, a new classmethod that provides an alternative constructor, its signature is zeroes(typecode, length), which allows for preallocating an array, with a lower overhead than methods such as array(typecode, [0]) * length, and is more readable (plus consistant with the numpy method of the same name).

[rhettinger]

-1

Array's API much more closely matches that of list than it does numpy's arrays which are profoundly different. So we should stick the one-way-to-do-it idiom: array.array('c', '\0') * 100 or array.array('d', [0.0]) * 100.

Also, the current idiom is much more flexible in that it works with any initializer, not just a zero (compare with numpy's ones() constructor for example).

[ncoghlan]

Compare however the behaviour of bytes() and bytearray() which support direct zero initialisation of a region of memory by passing an integer to their constructor.

[rhettinger]

Array has been around for a very long time and this functionality has never been requested. That and the fact that we already have a way to do it (the same way as for lists) is a strong indication that this isn't needed at all.

[sethtroisi]

I was surprised to learn about array.array('t', [0]) * 100 today and to find it the fastest way to initialize an empty array.

Adding a little timing info for people who might find this at some later date

import array
import itertools
from time import time

import numpy


def time_long(n, typecode='I'):
    s = time()
    counts = array.array(typecode, [0] * n)
    e = time()
    print(f"array.array({typecode!s}, [0] * {n}) = {e - s:.2f}")

    s = time()
    size = counts.itemsize
    counts = array.array(typecode)
    counts.frombytes(bytearray(size * n))
    e = time()
    print(f"array.frombytes(bytearray({size} * {n})) = {e - s:.2f}")

    s = time()
    counts = array.array(typecode, (0 for i in range(n)))
    e = time()
    print(f"array.array({typecode!s}, (0 for _ in range({n})) = {e - s:.2f}")

    s = time()
    counts = array.array(typecode, itertools.repeat(0, n))
    e = time()
    print(f"array.array({typecode!s}, itertools.repeat(0, {n})) = {e - s:.2f}")

    s = time()
    counts = array.array(typecode, [0]) * n
    e = time()
    print(f"array.array({typecode!s}, [0]) * {n} = {e - s:.2f}")

    s = time()
    counts = numpy.zeros(n, dtype=typecode)
    e = time()
    print(f"numpy.zeros({n}, dytpe={typecode!r}) = {e - s:.2f}")

time_long(10 ** 8)

array.array(I, [0] * 100000000) = 1.61
array.frombytes(bytearray(4 * 100000000)) = 0.59
array.array(I, (0 for _ in range(100000000)) = 4.17
array.array(I, itertools.repeat(0, 100000000)) = 1.51
array.array(I, [0]) * 100000000 = 0.32
numpy.zeros(100000000, dytpe='I') = 0.05