fix(cpp1): support empty and multi-argument indexing

[JohelEGP]

Resolves #482.
Partially addresses #387 (operator[]).

The added test requires C++23.

Testing summary.

100% tests passed, 0 tests failed out of 684

Total Test time (real) =  98.23 sec

Acknowledgements.

• Tested thanks to https://github.com/modern-cmake/cppfront.
• The commit message uses Conventional Commits 1.0.0.

[JohelEGP]

As you can see, this needs to run in C++23 only.

I'll have to enhance https://github.com/modern-cmake/cppfront for this.
Someone will have to do the same for the scripts bundled in this repository.

[JohelEGP]

See the generated test passing at https://cpp2.godbolt.org/z/PfoWWqznG in C++23.

[MaxSagebaum]

One question would be how the range checker can be extended to multiple dimensions. I do not know the standard proposal for this therefore I ask the question: How does the c++ standard defines the size operator for multi dimensional arrays? Can we use this mechanism to check the range of other dimensions?

[MaxSagebaum]

Maybe add a TODO, that this needs to be extended for multiple dimensions.

[JohelEGP]

I think that perhaps the correct thing to do
is to only avoid cpp2::assert_in_bounds when there are no index arguments,
and add a multi-dimensional overload for cpp2::assert_in_bounds that doesn't assert.
Then someone versed in std::mdspan can just add an overload for it that asserts.

[hsutter]

Thanks! I think it's fine to do only single-dimensional bounds checks for the time being -- it will be quite a while before multidimensional is actually usable/used in the field.

[JohelEGP]

How does the c++ standard defines the size operator for multi dimensional arrays?

There's none.
The resolved #482 links to the only occurrence in the C++ standard library:
https://en.cppreference.com/w/cpp/container/mdspan/operator_at.

[MaxSagebaum]

Ok thanks for applying my remarks.

It doesn't seem trivial to support multi-argument indexing in cpp2::assert_in_bounds.
See https://en.cppreference.com/w/cpp/container/mdspan/operator_at.

I had a look at cpp2::assert_in_bounds. Out of my had, one option would be:

template<typename ... Args>
auto assert_in_bounds_multi(auto&& x CPP2_SOURCE_LOCATION_PARAM_WITH_DEFAULT, Args&& ... args ) -> decltype(auto)
    requires ((std::is_integral_v<CPP2_TYPEOF(args)> && ...) &&
             requires { std::ssize(x, 1); x[arg]; std::begin(x) + 2; })
{
   
    int dim = 0;
    (Bounds.expects(0 <= args && args < std::ssize(x, dim++), "out of bounds access attempt detected" CPP2_SOURCE_LOCATION_ARG)), ...;
    return CPP2_FORWARD(x) [ CPP2_FORWARD(args)... ];
}

I have not tested this. The increment of dim in the expansion is kind of hacky. Here we would assume that there is a std::ssize(container, dim) function. Would this make sense?

Do you want to pursue such an implementation or should we leave it. I you want I can try and implement it.

[JohelEGP]

That doesn't work for std::mdspan.
See how its precondition has to look like: https://eel.is/c++draft/containers#mdspan.mdspan.members-note-1.

It could work for a container of containers
that added support for multi-dimensional indexing.

Without actual use-cases, it's hard to argue in favor of an overload that asserts.

[JohelEGP]

[Note 1: This implies that map_(I) < map_.required_span_size() is true.
— end note]

It seems like std::mdspan provides the interface to implement that precondition,
but I seems to be a pack.
Also, it can't be tested today.
Here's the implementation status, with libc++ noted "std::extents only".

[hsutter]

We should still have an error here though, should we? Like the one for ( at line 4181 below.