BUG: iloc on DataFrame with reference is a silent no-op with unsorted indexes

[rhshadrach]

• closes BUG: Swapping values with iloc causes aligns axes, wrong example in docs #65446 (Replace xxxx with the GitHub issue number)

This appears to me to be a somewhat severe bug - using iloc has no impact when there is another reference to the DataFrame.

df = pd.DataFrame({'A': [1, 2], 'B': [3, 4]})
ref = df[['A', 'B']]  # Create a reference
df.iloc[:, [1, 0]] = np.array([[10, 30], [20, 40]])
print(df)
#    A  B
# 0  1  3
# 1  2  4

The only thing that makes it not so severe is that the indices have to be not strictly increasing. I believe this was introduced in 2.1 with CoW enabled in #51435 but haven't run a git bisect here.

[jbrockmendel]

do we have a lib.is_range_indexer is something like that we can use here?

[jbrockmendel]

claude found an existing bug that this changes to a different bug:

df = pd.DataFrame({"A": [1, 2], "B": [3, 4]})
ref = df[["A", "B"]]      # hold a reference -> CoW path
df.iloc[[1, 0], [1, 0]] = 99

on main the setitem is (incorrectly) a no-op. Under this PR, it sets the diagonal elements when it should set all the elements.

Fine if you want to consider out of scope.

[jorisvandenbossche]

Suggested change

	blk_loc, inverse = np.unique(blk_loc, return_inverse=True)
	blk_loc_unique, inverse = np.unique(blk_loc, return_inverse=True)

So we keep the original blk_loc? (eg I would assume the values = values[blk_locs] below needs to us the original blk_locs?