Add additional heterogeneous overloads.

[BenKaufmann]

• Add heterogeneous overloads for try_emplace, insert_or_assign, and operator[] to {b}hopscotch_map and hopscotch_hash.

• Similar overloads were added to the standard associative containers with C++26 (see P2363R5).

[Tessil]

Thanks, it looks good.

Few comments:

• Would it be possible to also add the changes for the (b)hopscotch_set that are in the P2363R5?
• For the comment with P2363R5, could we use the This overload only participates in the overload resolution if... instead for all these new methods?
• At the time heterogeneous overloads were added to the project (2017), only std::map/set had these in the standard. The std::unordered_map/set ones were added in C++20.
  I used the same convention as std::map, requiring has_is_transparent<KeyEqual> but I didn't check for has_is_transparent<Compare>.
  Ideally this should be changed but I have kept it like that for now to keep backward compatibility.
  Could the new methods also only check has_is_transparent<KeyEqual> for consistency?

Thanks

[BenKaufmann]

Few comments:

* Would it be possible to also add the changes for the `(b)hopscotch_set`  that are in the P2363R5?

Done.

* For the comment with P2363R5, could we use the `This overload only participates in the overload resolution if...` instead for all these new methods?

Done.

* At the time heterogeneous overloads were added to the project (2017), only `std::map/set` had these in the standard. The `std::unordered_map/set` ones were added in C++20.
  I used the same convention as `std::map`, requiring `has_is_transparent<KeyEqual>` but I didn't check for `has_is_transparent<Compare>`.
  Ideally this should be changed but I have kept it like that for now to keep backward compatibility.
  Could the new methods also only check `has_is_transparent<KeyEqual>` for consistency?

I don't think that we can remove the is_convertible check from the try_emplace overloads as otherwise,
try_emplace(K&& k, Args&&... args) and try_emplace(const_iterator hint, K&& k, Args&&... args) are ambiguous. However, I adjusted the code such that it no longer uses std::is_convertible_v.

[Tessil]

Thank you for the change and sorry for the delay.

I don't think that we can remove the is_convertible check from the try_emplace overloads as otherwise,
try_emplace(K&& k, Args&&... args) and try_emplace(const_iterator hint, K&& k, Args&&... args) are ambiguous. However, I adjusted the code such that it no longer uses std::is_convertible_v.

Sorry, I meant to only have has_is_transparent<KeyEqual> as is transparent check and not has_is_transparent<Compare> but keep all other necessary checks (is_convertible). Could you check to put them back? Thanks!

I just want to avoid incoherences in the API were some heterogeneous methods require only has_is_transparent<KeyEqual> but other require has_is_transparent<KeyEqual> & has_is_transparent<Compare>.

[BenKaufmann]

Sorry, I meant to only have has_is_transparent<KeyEqual> as is transparent check and not has_is_transparent<Compare> but keep all other necessary checks (is_convertible). Could you check to put them back? Thanks!

I just want to avoid incoherences in the API were some heterogeneous methods require only has_is_transparent<KeyEqual> but other require has_is_transparent<KeyEqual> & has_is_transparent<Compare>.

Sorry, I'm not sure I can follow:

• I did not remove the is_convertible checks. I only changed my previous use of std::is_convertible_v (added with C++17) to a C++11 compatible std::is_convertible<...>::value under the assumption that this library should work with a C++11 standard-compliant compiler.
• I added has_is_transparent<Compare> in some places precisely to avoid further incoherences. Both bhopscotch_map.h and bhopscotch_set.h already require has_is_transparent<Compare> in the existing heterogeneous functions, like .e.g. erase(). So, for consistency, I also added this requirement to the new functions.

Could you please clarify your concerns so that I can update this PR again if needed?

[Tessil]

Sorry if I was unclear, it seems some std::is_convertible were removed (example) even though they are part of P2363R5. Is there any reason why they were removed?

[BenKaufmann]

it seems some std::is_convertible were removed [...] Is there any reason why they were removed?

Given that iterator is convertible to const_iterator, I assumed that checking for the latter is enough. Of course, this assumption is wrong in the general case. I now restored the previously removed checks and also added a test for the constraints.

[Tessil]

Sorry for the review delay and thanks for your patience.

It looks good to me in general. Just a bit worried about the static int inst in heterogeneous_test if we ever want to run the tests in parallel, but it isn't supported by Boost.Test currently anyway. It can always be changed later.

I merged the change.

Thanks again for your contribution.

[BenKaufmann]

Just a bit worried about the static int inst in heterogeneous_test if we ever want to run the tests in parallel, but it isn't supported by Boost.Test currently anyway. It can always be changed later.

I understand your concern and drafted an alternative approach: master...BenKaufmann:hopscotch-map:simplify-unit-tests.