Prevent make_span from working with vectors and arrays of pointers

[are-you-tilted-already]

The goal is to fix the problem with a confusing behaviour of flatbuffers::make_span with flatbuffers::Vector<(const)? T*>.
Consider the following flatbuffers schema:

// example.fbs
struct ExampleStruct {
  id: uint64;
}

table ExampleStructs {
  values: [ExampleStruct];
}

and some code working with it:

#include "example.fbs.h"

void ProcessStruct(const ExampleStruct& value);

void ProcessStructs(const flatbuffers::span<const ExampleStruct*> values) {
  for (const ExampleStruct* const value : values) {
    ProcessStruct(*value);
  }
}

const ExampleStructs* const table = ...;

// UB and most likely SIGSEGV, because the correct way to read is the one that flatbuffers::IndirectHelper<const T*> uses
// https://github.com/google/flatbuffers/blob/master/include/flatbuffers/buffer.h#L168
ProcessStructs(flatbuffers::make_span(table->values());

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[are-you-tilted-already]

Some people (including me, to be honest) use spans almost whenever they could, so it's better to make flatbuffers::Vector<const T*> to flatbuffers::Span<const T*> conversion impossible.

[are-you-tilted-already]

@aardappel @dbaileychess what do you think of it?

[aardappel]

I agree such an illegal conversion should not be possible, so disabling it is better than nothing.

Ideally it be converted to flatbuffers::Span<const T> instead, if on little endian? do we want to make that possible somehow?

And looks like CI wants some parentheses to clarify precedence.

[are-you-tilted-already]

Ideally it be converted to flatbuffers::Span<const T> instead, if on little endian? do we want to make that possible somehow?

I've made it possible via make_structs_span on little endian. Not sure if we need something like that for flatbuffers::Array<const T*>, though.

And looks like CI wants some parentheses to clarify precedence.

It is just std::is_pointer_v<T>, which requires at least C++ 17 :)

[aardappel]

It is just std::is_pointer_v, which requires at least C++ 17 :)

No, the actual error is error: '&&' within '||' and indeed there is a && and || at the same level.

Besides that, you'll need to modify this such that the version using std::is_pointer_v<T> is only used on compilers supporting it, or better yet, implement it in a different way that also works on older compilers. We still support all the platforms on CI.

[are-you-tilted-already]

It is just std::is_pointer_v, which requires at least C++ 17 :)

No, the actual error is error: '&&' within '||' and indeed there is a && and || at the same level.

Besides that, you'll need to modify this such that the version using std::is_pointer_v<T> is only used on compilers supporting it, or better yet, implement it in a different way that also works on older compilers. We still support all the platforms on CI.

My bad: fixed. Seems like it's more convenient to use std::is_pointer<T>::value, which requires C++ 11 only and is supported on most(?) compilers.

[aardappel]

Thanks!