Packed struct field alignment should be able to be derived from the alignment of the struct itself

[chr-1x]

Zig Version

0.10.0-dev.3559+d2342370f

Steps to Reproduce

const expect = @import("std").testing.expect;

const Foo = packed struct {
    a: i32,
    b: u8,
};

var buffer: [256]u8 = undefined;

test {
    var f1: *align(16) Foo = @alignCast(16, @ptrCast(*Foo, &buffer[0]));
    // Given: alignment of f1 is 16 (in this case, because we said it is, but
    // the aligned pointer could come from anywhere).
    try expect(@typeInfo(@TypeOf(f1)).Pointer.alignment == 16); // passes
    // Packed struct memory layout is guaranteed, and has no padding. So the
    // first field in the struct will have the same address as the struct.
    try expect(@ptrToInt(f1) == @ptrToInt(&f1.a)); // passes
    // Yet, the type of the address of the first field is *align(1) i32, not
    // *align(16) i32!
    try expect(@typeInfo(@TypeOf(&f1.a)).Pointer.alignment == 16); // fails
    // This would merely be an annoyance, but it makes atomic loads and stores
    // to these fields impossible without a cast, which shouldn't be necessary:
    //   @atomicStore(i32, &f1.a, 123, .Unordered);
    // error: expected type '*i32', found '*align(1) i32'
}

Expected Behavior

Taking the pointers of fields of a packed struct through an aligned pointer should give you aligned pointers to the fields.

Actual Behavior

All fields in a packed struct are *align(1), regardless if the pointer to the packed struct has a higher alignment.