_Node_handle: destroy with value_type allocator

[CaseyCarter]

Drive-by cleanup:

• explicitly compare pointers with nullptr in preference to if (ptr) and if (!ptr)
• _STL_INTERNAL_CHECK checkable preconditions
• Use constinit when available to verify constexpr constructor

Fixes #1309.

[BillyONeal]

construct isn't supposed to require that the allocator be rebound first; this introduces an extra rebind to support a case the standard doesn't require.

This is also inconsistent with what we do everywhere else in the node-based containers, for example

STL/stl/inc/list

Line 317 in 5f3e912

allocator_traits<_Alnode>::destroy(_Al, _STD addressof(_Ptr->_Myval));

[CaseyCarter]

construct isn't supposed to require that the allocator be rebound first; this introduces an extra rebind to support a case the standard doesn't require.

On the contrary; this introduces an extra rebind to support what the Standard explicitly requires in [container.node.cons]/3.1 and [container.node.dtor]/1.

[BillyONeal]

I see, so the standard does say that the type needs to match. We make no attempt to do that right now; all construct and destroy calls should probably be audited for this in the node based containers.

[CaseyCarter]

I see, so the standard does say that the type needs to match. We make no attempt to do that right now; all construct and destroy calls should probably be audited for this in the node based containers.

The quoted paragraphs have nothing to do with nodes stored in node-based containers, they only describe the behavior of node-handle while it holds a node. No, I don't know why the Standard wants these two cases to have different behavior - it seems completely pointless to me. I assume it was the simplest way to specify node-handle.

[BillyONeal]

The quoted paragraphs have nothing to do with nodes stored in node-based containers, they only describe the behavior of node-handle while it holds a node. No, I don't know why the Standard wants these two cases to have different behavior - it seems completely pointless to me. I assume it was the simplest way to specify node-handle.

I view it as a serious problem if construct and destroy do not match the ones the container used.

[CaseyCarter]

I view it as a serious problem if construct and destroy do not match the ones the container used.

IMO you're asking for trouble If construct and destroy don't have the same behavior within a rebind family.

[BillyONeal]

IMO you're asking for trouble If construct and destroy don't have the same behavior within a rebind family.

I mean, I agree, but that argues to not do the rebind.

[StephanTLavavej]

Looks good to me - I found a couple of potential clarity improvements in the test. I'll verify whether they actually compile; if so, I'll push changes.

[CaseyCarter]

🦶 : bug smashed.
🐞