Deprecate finalize_all()

[dalg24]

Oversight on our part, not used and not tested

[ldh4]

Does this mean that this functionality is not needed or is this already being done in the regular finalize?

[ldh4]

If this is to be deprecated, should Impl::finalize_internal also be refactored to not have a default argument? Currently, it seems to be only called from finalize and finalize_all.

[masterleinad]

I pretty much have the same question. What functionality do we need to offer to initialize and finalize individual backends? What are we promising to, say, Legion?

[dalg24]

I dropped the boolean argument from Implemented::finalize().
Cleaning further raises a few questions

kokkos/core/src/impl/Kokkos_Core.cpp

Lines 123 to 127 in 7a413b5

	void ExecSpaceManager::finalize_spaces(const bool all_spaces) {
	for (auto& to_finalize : exec_space_factory_list) {
	to_finalize.second->finalize(all_spaces);
	}
	}

kokkos/core/src/impl/Kokkos_ExecSpaceInitializer.hpp

Line 56 in 7a413b5

virtual void finalize(const bool all_spaces) = 0;

Presumably they are space we don't know about that inherits from ExecSpaceInitializerBase so we need to be careful there.

[masterleinad]

IMHO, it's fair to also drop it from

void ExecSpaceManager::finalize_spaces(const bool all_spaces)

since we control that as well but I agree to be careful about going deeper.

[dalg24]

Done

[dalg24]

Failures (one of the machine out of disk space) are unrelated. This is ready for merge.