>     Would this work:
>
>     - The launcher looks for the S-expression with the declarations.
>
>     - If it's found, use the "new style" semantics.
>
>     - If it's not found, use the semantics in the R6RS appendix.
>
> How can this work with the strict guarantee of R6RS to report every error?

I don't understand what you mean. The launcher isn't an R6RS
implementation; it hands off the script to an R6RS implementation.

> "#!r6rs" is not mandatory for R6RS and is not a marker for R6RS
> semantics. It is just a marker for R6RS lexical (!) syntax without
> extensions.

Will continue to think about this.

> If you rather want to choose pain than a painless approach, I'm out.

Indeed, explicitly choosing pain up front has been the guiding principle
behind all of my Scheme work. The fastest way to improve Scheme would be
having more people who do that.

It's always painful up front to organize something that resists being
organized. But it's more painful in the long term not to organize it.
Here as in so many things, we don't have the luxury of a painless approach.

Is it not painful to you to have a language "ecosystem" where things
don't work in the obvious way, and when you look it up, it's due to a
random technicality from 15 years ago that people didn't clean up when
they could? The story of `make` requiring tabs instead of spaces, and
not being fixed because it already had 10 users, comes to mind.