Adapting transform to C++20 (non-compilable PR)

[gonidelis]

This is an early adaptation for transform. It includes:

• adaptation for the end/begin iterators to be different in non-segmented transform
• adaptation for the end/begin iterators to be different in segmented transform (didn't touch segmented_transform() functions though)
• transform_projected specialization addition with util::projection_utility

To do

• Add tests for the adaptations.

There is a compilation error when:

make tests.unit.modules.segmented_algorithms.partitioned_vector_transform_binary2 and
make tests.unit.modules.segmented_algorithms.partitioned_vector_transform_binary3

which I believe stems from the projection_identity addition but I can't figure out why.

• Fix test compilation error for partitioned_vector_transform_binary2/3

Important

This is an early adaptation which aims to facilitate the final adaptation that will utilize CPOs.

[gonidelis]

I have adapted the result types for all three overloads of transform. There are lots of compilation errors that come from the way tests are implemented. The errors stem from here:

hpx/libs/algorithms/tests/unit/algorithms/transform_tests.hpp

Lines 66 to 67 in 256b42e

	HPX_TEST(hpx::util::get<0>(result) == iterator(std::end(c)));
	HPX_TEST(hpx::util::get<1>(result) == std::end(d));

and here:

hpx/libs/algorithms/tests/unit/algorithms/transform_tests.hpp

Line 94 in 256b42e

hpx::util::tuple<iterator, base_iterator> result = f.get();

hpx/libs/algorithms/tests/unit/algorithms/transform_tests.hpp

Line 162 in 256b42e

f.get();

as there is no get() function implemented for in_out_result or in_in_out_result types. I would like to discuss with @hkaiser on how to fix that.

• I need to add a projection_identity specialization for transform_binary_projected.

After fixing the compilation issues on the tests, I think I should finally proceed on creating the CPO's.

[hkaiser]

Calling the standard algorithms will have the correct semantics, but exposes inconsistent exception behavior. For the other algorithms we have settled on relying our own implementations even in those cases where semantically the standard version would be sufficient.

[aurianer]

Very nice ;)

[aurianer]

@hkaiser should we use the HPX_DEPRECATED_V macro here?

[hkaiser]

Yes, that's what we use everywhere else now.

[aurianer]

You may also want to adapt the begin/end iter template arguments for transform_loop too

[gonidelis]

Got it! Thanks!

[gonidelis]

This PR is closed and archieved. The transform C++20 adaptation will take place in a new PR #5051.