Indexing / slicing

The problem, from first principles

Three ways of indexing into a tensor/array look similar (all three select a subset of elements), but they have a genuinely important difference underneath: some return a view into the original memory, others return an independent copy. Code that treats every indexed subset the same way will work fine for one kind and silently misbehave for another the moment it mutates the result.

From theory to code

Implement basic_slice, boolean_mask, fancy_index, and is_a_view_of — a way to actually verify, rather than assume, whether a given result shares memory with its source.

Constraints

• basic_slice(a, start, stop) returns a[start:stop], sharing memory with a.
• boolean_mask(a, mask) and fancy_index(a, indices) both return independent copies, not sharing memory with a.
• is_a_view_of(child, parent) returns a plain bool: True iff child and parent share underlying memory.
• None of the four mutate their inputs directly (though a caller mutating a returned view will, by definition, affect the original — that's the point being tested).

Hints

Open one at a time. Each gives away a little more than the last.