CppCon2025

The Evolution of std::optional: From Boost to C++26

Steve Downey

These are notes from Steve Downey's (Bloomberg) CppCon 2025 talk. He is the primary author of P2988, the proposal that carried std::optional<T&> into C++26. The heart of the talk is one question: a feature that looks like "just a reference that can also be empty" — why was it first proposed in 2005, then held up all the way until the vote at the Sofia meeting in June 2025? The answer runs through the three identities a reference carries in C++, the twenty-year fight between assign-through and rebind, and the final conclusion: "it's just a constrained pointer."

The notes are split into six parts, ordered as "the value-version foundation first, then the reference-version core, then a step back to look at standardization." All the code involving optional<T&> was actually run on GCC 16.1.1 (-std=c++26) — this is not armchair theory.

Contents

STARTFINISHWhy the optional reference took twenty years→STARTFINISHThe Value-Semantics Foundation of std::optional→STARTFINISHWhat an optional reference is, and why assignment is always a rebind→STARTFINISHShallow traps of optional references: const, value_or, and dangling→STARTFINISHThe Move-Semantics Traps Hiding Inside Optional References→STARTFINISHThe Standardization Truth: The Beman Project and a Reference Implementation That Actually Runs→