array insertion fails const iterator conversion

[ned14]

toml::array to, from;
for(size_t n = 0; n < from.size(); n++)
{
  to.insert(to.end(), std::move(from[n]));  // fails with cannot convert argument 1 from 'toml::array::iterator' to 'toml::array::const_iterator'
}

[marzer]

I've implemented a fix for the iterator conversion though in doing so I've realized that what you're trying to do in this snippet won't work anyway; array::operator[] returns a toml::node&, and since toml::node is an abstract base, what you're trying to do wouldn't work without access to the underlying std::unique_ptr<node>. I could make it work as a copy, but it couldn't be a move.

I don't really have a good alternative without restructuring the node<=>table/array/value relationships to rely on something other than inheritance (as mentioned in #31). I'd really like to avoid introducing a 'moved from' or invalid state to the node objects (currently if a node exists it is a valid something), but I can implement something like std::list's splice() for merging arrays as a short-term solution.

I could also provide some means for working with the underlying std::vector<std::unique_ptr<node>> directly. That might actually turn out to be the best option; the simple public interface of the arrays and tables can stay more-or-less how it is (optimized for people largely just interested in querying the data), and people needing to do more advanced stuff can work with the underlying node data directly.

[marzer]

The iterator conversion fix is live in 289c95c, though it doesn't solve the underlying issue I discuss above. I'll leave this issue open and await your ideas on a path forward, since my fix of the issue doesn't really get at the heart of it just yet.

[ned14]

In my code I was actually visiting the node before moving it. The code above was for exposition only. Incidentally, you currently have visit() as a member function, I believe it ought to be an ADL discovered free function overload. See https://en.cppreference.com/w/cpp/utility/variant/visit for the API.

I'm going to consider this closed. Thanks for the rapid fix!

[marzer]

RE visit: toml++'s visit has slightly different semantics to std::visit in that it allows for the visitor to be non-exhaustive; it doesn't have to be callable for every possible node specialization, only the ones you need, so it's not really intended to be compatible with/related to std::visit in any way (I can't think of a situation where code would be so generic that std::visit compatibility would even make sense).