PartialOrd for opaque?

[pepyakin]

In #882 (PR #1002) we used PartialEq analysis to analyse whether we can derive PartialOrd. Generally this makes sense because if PartialEq can be derived then PartialOrd can be derived too.

But there is a problem. I can imagine what means for opaque types to be PartialEq: we just have to compare byte-by-byte both blobs. But I have a problem imaging what we should do in case of PartialOrd: compare each byte in both blobs?

Maybe we should split result of the analysis and explicitly state whether we can derive PartialOrd or not?

[pepyakin]

Writing this down I realized that it makes no sense to derive PartialEq for blobs either!
I'm not a compiler wizard, but isn't it is at least incorrect (if not UB) to compare byte-by-byte two structs because of stuff like padding bytes which, if I understand correctly, have undefined values.

[fitzgen]

Good point about padding. I guess we should only be "manually" implementing these things so that we don't look into padding.

Or I guess we could define a special type for padding that implements various traits with some kind of default/no-op that doesn't actually read the padding bits, and then we could still derive and the padding would effectively be skipped over.

This doesn't apply for opaque blobs of course.

---

@emilio if we stopped deriving traits for opaque things (except maybe Debug?), would that have bad implications for Stylo?