Farid Zaripov wrote:
If the _RWSTD_NO_LIST_NODE_BUFFER macro is not defined, then
list::splice() method do not conform the stadnard requirements (linear
complexity instead of constant time; invalidated iterators after
operation).
That's correct :( (If we don't have one yet, we should open
a bug report in Jira with a test case demonstrating the
problems.)
I see that the _RWSTD_NO_LIST_NODE_BUFFER macro is not defined
by default.
Also correct :(
Perhaps that macro should be defined by default?
It's definitely something to think about. As Liviu mentioned,
we discussed it a while back and decided to leave the current
non-conforming implementation enabled by default. Apparently
some users preferred fast insertion over fast (and conforming)
splicing.
I've thought about it several times since then and my feeling
is that we should probably change our implementation to provide
conforming semantics by default. Users who care more about the
speed of insertion should pick a different container (such as
deque).
If we do make the change, I think we might as well remove the
current default implementation altogether.
Thoughts?
Martin
