Add Canonical Name

[adraffy]

• changed IRegistry
  • added getCanonicalName() returns (name)
  • added event CanonicalNameUpdated
• changed IStandardRegistry
  • added setCanonicalName(name)
• changed PermissionedRegistry
  • added ROLE_SET_CANONICAL_NAME
  • implemented setCanonicalName() and getCanonicalName()
  • added tests
• changed LibRegistry
  • added findCanonicalRegistry(root, name, offset) returns (registry)
    • tries to avoid reverting and returns null when reasonable
  • added tests
• changed UniversalResolverV2
  • added isCanonical(name) returns (bool) — unsure if this is the right function to expose

---

• an alternative to Add Canonical Parent #209
• does a namehash() == namehash() check that could be a keccak check

---

⭐️ Personally, I like the #209 approach (as immediate access to the parent seems useful and it avoids someone using an non-canonical name) but this matches the original design that was discussed. It's also cleaner when you move things around, since only 1 canonical reference needs changed, instead of every descendant.

[immunefi-magnus]

🛡️ Immunefi PR Reviews

This pull request is not eligible for a PR Reviews review. Please contact Immunefi support.

Reason: This PR can't be reviewed because no PR Reviews plan is configured for your organisation. Please ask your admin to set up a plan to enable reviews.

[Arachnid]

We discussed this a while ago and you made a good argument that having a (registry, token id) pair was a better option than a name. Have you changed your mind?

[adraffy]

Yeah, there's a slack discussion (where I'm arguing with myself 🤣️) and #209 is the other implementation.

They're basically equivalent and only differ in their properties:

#209 (registry, label) uses slightly more storage for short names and the only way to get the canonical name is to construct it. This means you never have a non-canonical "canonical" name: you either have a canonical name or nothing.

#217 name gives each registry a name, but you must check that its actually canonical. The "fix" is requiring that each ancestor is also canonical and the optimization is checking that each parents canonical name has a common suffix with the child:

/// @dev `name` is canonical if it's ancestors are canonical and share parential suffixes:
/// eg.        Canonical    not-Canonical     not-Canonical
///            =========    =============     =============
///     abc.sub.test.eth          abc.eth      a.b.test.eth
///         sub.test.eth            > xyz <      b.test.eth
///             test.eth            <root>       > nick.eth <
///                  eth                                eth
///                <root>                             <root>

---

Ignoring storage, it's ultimately the choice between:

• The registry has an explicit parent but UR.findCanonicalName(registry) returns (name?) is required to derive the name
• The registry has an explicit name but can only be used if UR.isCanonical(name) returns (bool) is true

[adraffy]

Using #209