MSC3267: Reference relations (#3267)

* initial draft of reference relations msc

* change MSC number

* Apply formatting

* Convert to point at present rather than ideal

* Clarify that multiple relations is a thing we don't have and won't fix here

* Fix wording to match reality, again

* fix a typo

Co-authored-by: Travis Ralston <travisr@matrix.org>
Co-authored-by: Andrew Morgan <andrewm@element.io>

proposals/3267-reference-relations.md

@@ -0,0 +1,92 @@
+# MSC3267: reference relationships
+
+## Proposal
+
+This proposal defines a relation type (using
+[MSC2674 relations](https://github.com/matrix-org/matrix-doc/pull/2674))
+for events to make a reference to another event.
+
+A `rel_type` of `m.reference` is defined as a generic way to associate an
+event with another event. As a bundle, `m.reference` relations appear as
+an object with a single `chunk` field. The `chunk` is an array of objects
+with a single `event_id` field for all the child events which `m.reference`
+the parent.
+
+There are no implied semantics by a reference relation: the feature or event
+type which makes use of the `rel_type` should specify what sort of semantic
+behaviour there is, if any. For example, describing that a poll response event
+*references* the poll start event, or that a location update *references* a
+previous location update.
+
+Reference relations are used by [MSC2241](https://github.com/matrix-org/matrix-doc/pull/2241)
+to tie all events together for the same verification request.
+
+For instance, an `m.room.message` which references an existing event
+would look like:
+
+```json5
+{
+    // Unimportant fields omitted
+    "type": "m.room.message",
+    "content": {
+        "msgtype": "m.text",
+        "body": "i <3 shelties",
+        "m.relates_to": {
+            "rel_type": "m.reference",
+            "event_id": "$another_event_id"
+        }
+    }
+}
+```
+
+## Server aggregation
+
+[MSC2674](https://github.com/matrix-org/matrix-doc/pull/2674) states
+that values for `rel_type` should define how the server should aggregate the
+`rel_type`. For `m.reference`, child events are bundled to appear as follows
+under `m.relations`:
+
+```json5
+{
+  "m.reference": {
+    "chunk": [
+      {"event_id": "$one"},
+      {"event_id": "$two"},
+      {"event_id": "$three"}
+    ]
+  }
+}
+```
+
+Note that currently only the `event_id` is noted in the chunk, however a future
+MSC might add more fields.
+
+## Limitations
+
+Different subtypes of references could be defined through additional fields on
+the `m.relates_to` object, to distinguish between other forms of semantic behaviour
+independent of type (hypothetical threads, replies, etc if we didn't have a system
+for them already). This MSC doesn't attempt to define these subtypes.
+
+This relation cannot be used in conjunction with another relation due to `rel_type`
+being a single value. This is known and unfortunately not resolved by this MSC.
+A future MSC might address the concern.
+
+## Edge Cases
+
+Can you reference an event which doesn't have logical meaning? Eg to a [reaction](https://github.com/matrix-org/matrix-doc/pull/2677)?
+ * Yes, at the protocol level.  But you shouldn't expect clients to do anything
+   useful with it.
+ * The relationship is effectively pointless, so the event would appear as though
+   there was no reference relationship.
+
+Do we need to support retrospective references?
+ * For something like "m.duplicate" to retrospectively declare that one event
+   dupes another, we might need to bundle two-levels deep (subject+ref and then
+   ref+target).  We can cross this bridge when we get there though, as another
+   aggregation type
+
+## Unstable prefix
+
+Unfortunately not applicable - this MSC was used in production and appears in the
+specified version of the [key verification framework](https://spec.matrix.org/v1.2/client-server-api/#key-verification-framework).