This makes sense to fix, though the annoying part is figuring out whether anywhere else in the code is relying on the current behavior. Or whether it is likely to break any downstream users and thus might warrant additional caution

It looks like a minimal fix is something like so.

This will make all the tests run successfully.

Regarding side effects I would suppose anyone interfering with the TAG data directly might have colliding types. What use case brought you to this, @lgpasquale ?

I suppose taking care of SSHORT is needed as well? This would require introducing an additional Value:

pub enum Value {
    Byte(u8),
    Short(u16),
    Signed(i32),
    SignedBig(i64),
    Unsigned(u32),
    UnsignedBig(u64),
    Float(f32),
    Double(f64),
    List(Vec<Value>),
    Rational(u32, u32),
    RationalBig(u64, u64),
    SRational(i32, i32),
    SRationalBig(i64, i64),
    Ascii(String),
    Ifd(u32),
    IfdBig(u64),
}

Adding support for SSHORT (i16) and SBYTE (i8) at #234, I can try to add the patches for the casting of SHORT as u16 instead of u32 in a follow up PR if there's interest?