Plans for c++23

We had two recent crashes caused by constructing a std::string from a nullptr.

We've adopted every C++ language version since C++11 about 3 years after it was ratified (despite usually trying to be faster). Even then, that tended to cause issues with other toolchains not being ready yet. So I'd expect this to be another 1-3 years out.

Yes, both crashes were due to converting existing JNI code that returned a jstring (which is a pointer) to returning std::string. These refactorings are pretty mechanical, and the recent one was semi-automated. The fact that "return nullptr" turns into a runtime crash is very unfortunate when converting for these conversions.

We have an automated read-only mirror of upstream libc++ to make it impossible for us to have a local diff, so that's sadly not possible :/

I started kicking the tires on c++23 recently, and it seems mostly baked and not TOO hard to get to. If we want to, it would probably be feasible to take it around August/September, with Clang 21 and the annual AOSP upgrade. However, I also got immediate feedback that c++23 isn't a priority and I shouldn't spend time on it, so if ATLs want to do that, it needs to be an explicit decision. Otherwise I'm not going to bother doing the compat work for it for the foreseeable future, and I don't know of anyone else planning to drive these upgrades.

We talked about this particular diagnostic a bit in the libc++ discord channel ([1]). It didn't immediately lead to something, but maybe we can get a warning for this in the next few months, without having to bump language versions.

Patching locally is a good idea for these migrations. Thanks for that idea :)

Yay!