dont the (planned?) semantics of packed structs mean any undefined bits cause the entire thing to be undefined? probably shouldnt recommend doing that in an upgrade guide

Good call, thank you.

/home/ci/actions-runner1/_work/zig/zig/test/behavior/field_parent_ptr.zig:1755:1: error: packed store

should I disable the test for stage2_aarch64 for now?

Weren't explicit packed union backing ints part of #19754? That's still a compile error after this patch

Sounds like #24715 to me

I thought that's just about making the tag mandatory, I was referring to this comment
This snippet:

const U = packed union(u16) {
    x: packed struct(u16) {
        data: u8,
        padding: u8 = 0,
    },
    y: u16,
};

comptime {
    _ = U;
}

still doesn't compile with:

check.zig:1:24: error: packed union does not support enum tag type
const U = packed union(u16) {
                       ^~~

Sorry, you're right. There is not a proposal open which is limited to supporting backing integer type for packed unions, although several proposals imply its existence. I'll add that right now.

#25350