stdint.h sorts before stdio.h.

Same problem here, no?

Yes, I guess that path just didn't get executed during buildworld. Will update patch.

And here.

This field is needed since otherwise the compiler complains: https://godbolt.org/z/TqeEcn
flexible array member 'wstr' not allowed in otherwise empty struct

I would say so. There are "only" 57 instances of ->len (and no .len) so it wouldn't be _that_ painful to fix if someone felt sufficiently motivated.

I agree it would be much nicer, maybe it's not too difficult to update with a bit of search+replace

Should fix accesses the non-active union member

What's the reasoning behind this one? You should have size_t * int which will give either int or size_t depending on 32-bit vs 64-bit, I believe, both of which should then work fine as an offset. But if there's something weird going on with that would it not be best to make vlen a size_t instead?

This changes based on char signedness?

Is setting the MSB with | less UB than shifting into it?

Does this want to be (int64_t)-1 or something instead? Using UINT64_MAX is a bit weird.

I can't remember what the error was, I think it was that vlen can be zero and UBSan complains that adding a large unsigned number to dptr overflows it.

(int64_t)-1 still shifts the sign bits. We have to use an unsigned type to silence the warning.

Then the if below should be:

if (vlen > 0 && ((u_short *)dptr)[vlen - 1] == 0)

and this line can go?

Ah, right, the UB is specifically anything that's not left-shifting a non-negative value to give a still-representable result.

I didn't get an error, but it's possible that it would be one. Maybe it just wasn't triggered with the given inputs.