PERF: use copy=False in constructor in unstack (CoW)

[jorisvandenbossche]

xref #57431

I think we are sure that the values from self.get_new_values(..), which we pass here to the DataFrame constructor, are always new values owned by us (not shared by any other object), and so we can safely set copy=False in that case.

Mimicking the time_unstack benchmark:

m = 100
n = 1000

levels = np.arange(m)
index = pd.MultiIndex.from_product([levels] * 2)
columns = np.arange(n)
values = np.arange(m * m * n).reshape(m * m, n)
df = pd.DataFrame(values, index, columns)

In [2]: %timeit df.unstack()
316 ms ± 44.9 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)   # main
228 ms ± 25.1 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)   # PR

[phofl]

yeah this should be fine

should be good to backport as well

[rhshadrach]

I think we are sure that the values from self.get_new_values(..)... are always new values owned by us

I'm not sure that's the case. This path:

pandas/pandas/core/reshape/reshape.py

Lines 251 to 272 in c5bf3d3

	sorted_values = self._make_sorted_values(values)
	# place the values
	length, width = self.full_shape
	stride = values.shape[1]
	result_width = width * stride
	result_shape = (length, result_width)
	mask = self.mask
	mask_all = self.mask_all
	# we can simply reshape if we don't have a mask
	if mask_all and len(values):
	# TODO: Under what circumstances can we rely on sorted_values
	# matching values? When that holds, we can slice instead
	# of take (in particular for EAs)
	new_values = (
	sorted_values.reshape(length, width, stride)
	.swapaxes(1, 2)
	.reshape(result_shape)
	)
	new_mask = np.ones(result_shape, dtype=bool)
	return new_values, new_mask

with self.sort=False only does reshape and swapaxes. Both of those can be no copy, right?

[jorisvandenbossche]

I'm not sure that's the case. This path:

Ah, yes, I missed the early return there.

Continued in -> #57487