On Sat, 18 Apr 2026, 11:01 Giuseppe D'Angelo via Std-Proposals, <
std-proposals_at_[hidden]> wrote:

> Hello,
>
> On 18/04/2026 05:04, Muneem via Std-Proposals wrote:
> > I updated my proposal:https://docs.google.com/document/d/1srd5nKI-
> > QZy0vpUdkQoD45JyKIDWhom8/edit?
> > usp=sharing&ouid=114869935649093915681&rtpof=true&sd=true <https://
> > docs.google.com/document/d/1srd5nKI-QZy0vpUdkQoD45JyKIDWhom8/edit?
> > usp=sharing&ouid=114869935649093915681&rtpof=true&sd=true>
> > Thank you a lot to everyone for giving their feedback that led me to
> > refine my proposal
>
> First and foremost: this isn't a proposal. It's just an abstract idea. A
> proposal has to go in detail: discuss tradeoffs, provide a reference
> implementation, tentative wording.
>
>
> In 2.1 there's this claim:
>
> > Existing tuples cannot be optimized for runtime indexing without
> breaking the Application Binary Interface (ABI). Furthermore, switch
> statements are not guaranteed to be the fastest option for tuples with many
> elements.
>
> Could you show some data supporting this statement?
>
> Do you have a prototype implementation where you add such runtime
> `get(index)` to an existing std::tuple implementation, as well as a
> prototype implementation of the proposed runtime_index_tuple, and show
> that the new class brings significant performance gains (... that can't
> otherwise be achieved without breaking std::variant's ABI)?
>
> > This proposal introduces a specialization for std::variant<T&...> with
> the following properties:
> > No Valueless State: The variant cannot be valueless by exception.
> > Immutable Type Selection: Every std::variant<T&...> is constructed with
> a reference to a type T that it holds for its entire lifetime.
> > Assignment Logic: Assignment modifies the underlying value referred to
> by the variant rather than changing the active type.
>
> 1) These are extraordinary claims that aren't backed by extraordinary
> facts.
>
> 2) Adding support for references in std::variant deserves its own paper.
> It's something that was there originally and then removed by P0510.
>


Until this idea has been implemented and tested, this whole thread is a
waste of time.

Try to implement it, then come back with a real proposal with numbers to
prove the claims.