[span] Make span standard_layout

[miscco]

This moves span from inheritance to composition, which is problematic given that we define members both in the base class (_Mysize) and the derived class (_Mydata)

However, this is blocked on [[no_unique_address]] working on MSVC

[AdamBucior]

The design @ned14 proposed in #874 (comment) would be much better.

[miscco]

@AdamBucior couldyou elaborate on that as this is essentially the same design.

The one difference with inheritance would be that One would have to sprinkle this-> all over the place

[AdamBucior]

The one difference with inheritance would be that One would have to sprinkle this-> all over the place

You can use using _Mybase::_Some_member

[miscco]

Again I am unsure what benefit inheritance has over composition?

[AdamBucior]

Again I am unsure what benefit inheritance has over composition?

I feel like inheritance is more trustworthy and harder to get wrong. Inheritance is already widely used in the STL and span already uses it, so will be easier to merge. Also won't have to be blocked on no_unique_address.

[miscco]

Ok so given that MSVC does not support [[no_unique_address]] yet I went again with inheritance as this does not create any ABI problems.

[AlexGuteniev]

this is blocked on [[no_unique_address]] working on MSVC

@aymarino explained:

Regarding the feature itself, we do plan on implementing it, but since it is an ABI-breaking change we will not be able to enable it by default until we release an ABI-breaking toolset.

See here: https://developercommunity.visualstudio.com/solutions/973748/view.html

So I'm afraid that if you can't do it without [[no_unique_address]], you cannot do it until vNext.

[miscco]

Note that the _Other.size() calls in the constructors cannot be removed as _Other might be a different type where we do not have access to the private members

[miscco]

Ok. I am going out on a limb an declare that the futures::wait_until test was not broken by these changes.

[cbezault]

/azp run

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[cbezault]

We'll re-run and see if it's spurious. If it is I'll open an issue to track it/investigate it later.

[StephanTLavavej]

I think this is a good improvement (despite requesting more information in #874 about why pointer-length layout is so important). Noticed minor stylistic issues.

[StephanTLavavej]

Looks good, but I'd like to push a change to reverse the internal constructor parameters.

[StephanTLavavej]

Thanks for simplifying span's representation and improving its debug codegen before the ABI freeze! 😺