Fix aliasing issues with new host function system

[arhag]

Change Description

This PR change the implementation of is_aliasing in preconditions.hpp to fix bugs exhibited during certain corner cases of aliasing. The previous implementation would consider those cases (having to do with empty spans) to be aliasing whereas the new one is more permissive. The new behavior allows us to define the aliasing behavior simply: aliasing occurs if the intersection of two ranges is not empty.

The PR also makes additional bug fixes in preconditions.hpp (see definition of core_precondition and the change in the definition of to_span) to allow alias checking involving vm::argument_proxy<T*> to work properly; before it would never consider such pointers to be aliasing even in cases where it should have been.

The kv_tests/alias unit tests were added to check for various cases of aliasing or non-aliasing with the two KV database intrinsics kv_set and kv_get.

In addition, the changes to to_span are now also careful to avoid constructing a span<T> using a possibly unaligned void*. To be safe, we need to ensure that the void* passed by the WASM code is properly aligned before using it to construct a span<T>. This requires changes in EOS VM. We currently don't have an issue because we only use span<char> and span<const char>. More work is involved if we want to support span<T> for general T with stricter alignment requirements. Therefore, the currently unnecessary alignment check for spans in core_precondition has been removed and the comment under is_whitelisted_type has been updated.

Consensus Changes

• Consensus Changes

API Changes

• API Changes

Documentation Additions

• Documentation Additions