Power BI vs vecvec

vecvec provides an R class that holds multiple vectors as one logical vector without copying them together, for cases where concatenating would be wasteful. The 1.0.0 rewrite moved the class off vctrs onto S7 while keeping user-facing code working, and added matrix and array behaviour. Since then the work has been about whether the abstraction actually saves anything: 1.3.0 makes duplicated(), equality proxies, casting and array formatting compute slot-wise instead of materialising, and adds vecvec_mapply() to apply a function across several vecvecs at once.

The arc runs from proving the idea to making it cheap, and 1.3.0 is the widest pass yet at the second half. Early releases established constructors and vctrs dispatch; 1.0.0 rebuilt the internals on S7; the three releases since have worked through the operations that were quietly defeating the point — printing, duplicate detection, casting, equality — and made each compute on storage slots rather than elements. ALTREP detection has moved from parsing .Internal(inspect()) output to a C-level check, the same work on a firmer footing. The other visible thread is a widening apply surface: vec_apply() per vector, now vecvec_mapply() across several.

With the main vctrs operations converted to slot-wise computation, the remaining materialisation points are the natural next target; the entries name no specific one, and the internal structure reserved at 1.0.0 still leaves room for the faster special-case representations flagged then.