There are still ways for a sort function to not work properly with the comparer, even if the criterion is correct. As mentioned in the article, the comparer might have internal state that the sorter duplicates during intermediate steps, thus breaking assumptions made by the caller. The example the article gives is a comparer that increments a variable each time it's called, to count the number of comparisons. Depending on how this is done and how the sorter is implemented, copying the comparer may break this behavior.

> to not work properly with the comparer, even if the criterion is correct.

Then the algorithm is incorrect or its specification for the compare function.

> The example the article gives is a comparer that increments a variable each time it's called, to count the number of comparisons. Depending on how this is done and how the sorter is implemented, copying the comparer may break this behavior.

Yep, then that means the comparer is not valid and does not satisfy the necessary requirements.

In C++ your relational operators can return std::partial_ordering etc which is a bit more semantically meaningful, for the reader and the compiler.

Agreed. And I think that's a good idea to do. It's still possible to do wrong.

> What should it do?

If I'm reading the article correctly, the user can be informed at runtime with little to no performance impact. That certainly sounds preferable to returning an unsorted list (that may not even match the input).

Almost every Rust sort does this, but no C/C++ sorts do. That appears to support that author's conclusion that this is a cultural thing.

> If I'm reading the article correctly, the user can be informed at runtime with little to no performance impact.

It's not possible to verify a function is a strict weak ordering without enumerating the entire domain.

Right, but restricting the domain to just the list you're already enumerating makes things easier.

Yep, so now your sort function isn't just a sort. It's a sort with a unit test at the beginning. This burden is placed on EVERY use.

I'm not sure enumerating the entire domain is even possible in this setting, as the comparison function could return random values for each call.

The idea is that the implementation doesn't promise it will detect a strict weak ordering violation, but it has code that will detect the effects of such violations with a high probability. It's not a sampled approach, but rather as part of the small-sort a bi-directional merge is performed and if the pointers don't line up, the only explanation is a strict weak ordering violation, the result of the merge is discarded and a panic is raised.