Why does MemoryBlock::create_derived require parent memory to not be mapped?

[Silverlan]

I have a parent buffer with a large chunk of memory. Since I need to make frequent changes to it, I keep the entire buffer permanently mapped. However, I also want to split this buffer into many smaller derived sub-buffers. Since the parent buffer is already mapped at this point, this will trigger an assertion failure in _MemoryBlock::createderived:

Anvil::MemoryBlockUniquePtr Anvil::MemoryBlock::create_derived(MemoryBlock* in_parent_memory_block_ptr,
                                                               VkDeviceSize in_start_offset,
                                                               VkDeviceSize in_size)
{
    MemoryBlockUniquePtr result_ptr(nullptr,
                                    std::default_delete<MemoryBlock>() );

    if (in_parent_memory_block_ptr == nullptr)
    {
        anvil_assert(in_parent_memory_block_ptr != nullptr); // Assertion failure!

        goto end;
    }

    result_ptr.reset(
        new Anvil::MemoryBlock(in_parent_memory_block_ptr,
                               in_start_offset,
                               in_size)
    );

end:
    return result_ptr;
}

The question is, is that assert really necessary? Writing to (or mapping) a sub-buffer redirects to the parent buffer anyway, so I don't see any harm in the parent buffer being already mapped at the time the sub-buffer is created.

[DominikWitczakAMD]

That assertion check acts more of a safe-guard against use cases we currently do not have test coverage for internally. (yes, we actually do have a fairly thorough test coverage for Anvil but it can't be shared for various reasons). I believe it dates back to the old times where buffer chaining used to be a mess and use cases like the one you mention here would basically explode.

If you can confirm everything works fine for you, I'm happy to ditch that assertion check.

[DominikWitczakAMD]

Addressed internally.