Continue from #33824 (comment):

I was thinking about permitting nullptr when size=0, or even just permitting nullptr for any size, when the data pointer is followed by a size?

Afaict our de-serialization can handle this case (nullptr when size=0) correctly, but fails when passed a nullptr with non-zero size. Maybe a null check in btck_block_create(...) would be better? What do you think?

Yes, either should be fine. To clarify I had two alternative suggestions:

• Remove all non-null attributes for spans (data+size) and replace them with null checks when the size is one or larger.
• Remove all non-null attributes for spans without replacement.

I was wondering if the std::span nullptr check was part of the C++26 library hardening, but it seems not (https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2025/p3471r4.html#span-constructors), so I guess if we want it, we'd still have to write it ourselves.

Thanks for picking this up! Looks like a simple and small change in your pull request.