Perhaps, if its not already a NodeArray, we should instead clone it via .slice() before modifying it? Then we wouldn't need the Nodes<T> type, though at first glance I'm not certain what the memory/performance impact would be.

I tested out adding slice inside of createNodeArray (at the bottom of the if (elements) block), and actually saw a slight decrease in emit time for some reason -- might have to do with the fact that createNodeArray would otherwise have to change the shape of an array object acquired from elsewhere, but if we create a fresh slice it may be allocated with the correct shape already.

The most efficient option may be to avoid ever calling createNodeArray on an existing array, and instead always start by calling createNodeArray to get an empty mutable array, then pushing to it.