Whether an enum struct variant is multi-line depends on doc comments of previous variants

[RalfJung]

rustfmt considers this to be correctly formatted:

/// Some item that needs to monomorphize successfully for a MIR body to be considered well-formed.
#[derive(Copy, Clone, PartialEq, Eq, Debug, Hash, HashStable, TyEncodable, TyDecodable)]
#[derive(TypeFoldable, TypeVisitable)]
pub enum MentionedItem<'tcx> {
    /// A function that gets called. We don't necessarily know its precise type yet, since it can be
    /// hidden behind a generic.
    Fn(Ty<'tcx>),
    /// A type that has its drop shim called.
    Drop(Ty<'tcx>),
    /// Unsizing casts might require vtables, so we have to record them.
    UnsizeCast { source_ty: Ty<'tcx>, target_ty: Ty<'tcx> },
    /// A closure that is coerced to a function pointer.
    Closure(Ty<'tcx>),
}

And this:

/// Some item that needs to monomorphize successfully for a MIR body to be considered well-formed.
#[derive(Copy, Clone, PartialEq, Eq, Debug, Hash, HashStable, TyEncodable, TyDecodable)]
#[derive(TypeFoldable, TypeVisitable)]
pub enum MentionedItem<'tcx> {
    Fn(Ty<'tcx>),
    Drop(Ty<'tcx>),
    /// Unsizing casts might require vtables, so we have to record them.
    UnsizeCast {
        source_ty: Ty<'tcx>,
        target_ty: Ty<'tcx>,
    },
    /// A closure that is coerced to a function pointer.
    Closure(Ty<'tcx>),
}

For some reason, adding doc comments to Fn and Drop variants changes whether the UnsizeCast variant should be multi-line. That doesn't make any sense to me...

rustfmt.toml:

version = "Two"
use_small_heuristics = "Max"
merge_derives = false

[ytmimi]

Similar to #6106 but I'm pretty sure this is a duplicate of #5662