[lmi] When should std::distance (not) be used?

[Greg Chicares]

AFAICT, std::distance() should be used only when truly needed.
lmi uses std::distance() only twice. Here:

template<typename InputIterator, typename T>
bool each_equal(InputIterator first, InputIterator last, T const& t)
{
    return std::distance(first, last) == std::count(first, last, t);
}

it's unavoidable because we have no container, only iterators. But
in multiple_cell_document::parse(), where we push_back() each
element of an xmlwrapp "container" into a std::vector:

    xml::const_nodes_view const subelements(i.elements());
    v.reserve(std::distance(subelements.begin(), subelements.end()));
    for(auto const& j : subelements)

shouldn't xml::const_nodes_view::size() be used instead of distance()
-    v.reserve(std::distance(subelements.begin(), subelements.end()));
+    v.reserve(subelements.size());
because it's less verbose than distance(begin, end) and no slower?
Or is there some subtlety I'm missing here?

The related function std::advance() is used only once in lmi:

    std::vector<double>::const_iterator corr = CompleteGptCorridor().begin();
    LMI_ASSERT
        (   static_cast<unsigned int>(IssueAge)
        <=  CompleteGptCorridor().size()
        );
    std::advance(corr, IssueAge);
    GptCorridor.assign
        (corr
        ,CompleteGptCorridor().end()
        );

and there I think it's better to do

-    std::advance(corr, IssueAge);
     GptCorridor.assign
-        (corr
+        (corr + IssueAge
         ,CompleteGptCorridor().end()
         );

Making that less generic is IMO advantageous: changing these vectors to
lists would be such a bad idea, that it's good to produce an error if
anyone ever tries to do that.