Bryce Adelstein Lelbach

I'll rank ABI stability at the bottom. I need the other two, and I believe C++ is marked for EoL if we do not decide on that as a language.

I rank it at the bottom too, but I'm concerned that the reason for that is that we've been able to take it for granted for so long. With other languages, build tools, and even operating systems I've used historically, interface instability has been a *massive* setback.

1 and 3, build everything from sources

But many of us can't. There are industries where you rely on algorithms bought from third parties who don't share their code, you only get headers + precompiled binaries. Recompiling everything is simply not an option. Instead, you get stuck on older compilers, which is very sad.

Totally 1 and 3. Use C APIs for binary interoperability.

In the best case create C++ user side wrappers over the C API. In the worse case, create C API wrappers as binary bridge DLLs to retrofit older ABI APIs (fortunately I’ve never had to do this).

1 and 3. ABI compatibility might be required by compiler vendors, but it’s ultimately not a requirement for most devs. And if required, wouldn’t a C API compatibility be enough?

As a Rust programmer, I'll say that it's a little rough having only the C ABI and not your own lang's ABI.