Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

MSC3440: Threading via m.thread relation #3440

Merged
merged 73 commits into from Mar 9, 2022
Merged
Changes from all commits
Commits
Show all changes
73 commits
Select commit Hold shift + click to select a range
beda89e
Threading via relation
germain-gg Oct 13, 2021
2fa27ac
Add explainer on how to handle m.in_reply_to
germain-gg Oct 18, 2021
d8a0a94
Clarify wording on threading MSC
germain-gg Oct 18, 2021
7d39887
Mention MSC3051 in the alternative section of MSC3440
germain-gg Oct 18, 2021
6e37911
Clarify updates to MSC2675 for MSC3440
germain-gg Oct 18, 2021
c142b17
Line wrap the MSC
germain-gg Oct 18, 2021
c578f75
More line wrapping for MSC3440
germain-gg Oct 18, 2021
33acdf4
Clarify single-layer event aggregation section
germain-gg Oct 25, 2021
7102165
Update thread-as-rooms advantages
germain-gg Oct 25, 2021
f84f949
Clarify backwards compatibility and incremental support
germain-gg Nov 17, 2021
f02dc8d
Clarify wording and correct typos
germain-gg Nov 17, 2021
3e46728
Splitting Cerulean and MSC2836 in alternatives section
germain-gg Nov 17, 2021
44e967f
Add dependencies for threads MSC
germain-gg Nov 19, 2021
65d0d55
Clarify intro to threads as rooms
germain-gg Nov 19, 2021
2b76a6e
Add currentUserParticipated flag
germain-gg Nov 19, 2021
99c5b2e
snake_case over camelCase
germain-gg Nov 19, 2021
4ee42b1
Adding dependency to MSC3567
Dec 14, 2021
fc81bbd
Add threads capability
Jan 5, 2022
91e6ec7
Merge branch 'main' into gsouquet/threading-via-relations
germain-gg Jan 5, 2022
eaeef00
Fix typo
germain-gg Jan 7, 2022
26fb5f2
Update syntax highlighting to use jsonc
germain-gg Jan 11, 2022
a23c795
Add limitations when fetching thread content by relation type
germain-gg Jan 11, 2022
6b1a368
Add reply chain fallback via m.in_reply_to
germain-gg Jan 13, 2022
f227592
Clarity in wording and fix typo
Jan 18, 2022
b493f21
Cosmetic changes based on pull request feedback
germain-gg Jan 18, 2022
46e1e9b
Add note to allow clients to omit fallback for rich replies
germain-gg Jan 18, 2022
e40efa0
fix typo
germain-gg Jan 18, 2022
23928e7
Clarify wording to not confuse thread answers with quote-replies
germain-gg Jan 20, 2022
0880a86
move relations justification to alternatives section
germain-gg Jan 20, 2022
1bbb021
Clarify handling of m.in_reply_to missing rel_type:m.thread
germain-gg Jan 20, 2022
3c977f7
Fix typo
germain-gg Jan 20, 2022
0140454
Fix typo
germain-gg Jan 20, 2022
700464c
Declare MSC2781 as a dependency
germain-gg Jan 21, 2022
0035202
Use rich reply over quote reply
germain-gg Jan 24, 2022
e3cb699
Depend on MSC3676 rather than MSC2781
ara4n Jan 26, 2022
847f468
Remove full stop typo
germain-gg Feb 10, 2022
c8ffa62
Clarify new filtering parameters.
clokep Feb 14, 2022
a7cbf8d
Fix typo.
clokep Feb 15, 2022
5896d69
Update wording for client side considerations
germain-gg Feb 16, 2022
ee5df80
Add m.in_reply_to mixin to thread fallback
germain-gg Feb 16, 2022
00daf64
Add guidance for clients and servers for thread invalid relations
germain-gg Feb 16, 2022
a5d8aab
update thread root wording
germain-gg Feb 17, 2022
d7ed3c4
Add better definition to reply target event
germain-gg Feb 22, 2022
5c04906
Add note regarding forward compatibility
germain-gg Feb 22, 2022
d667a0b
link to MSC2674
richvdh Feb 22, 2022
b157dfd
Update proposals/3440-threading-via-relations.md
germain-gg Feb 22, 2022
3162bea
Clarification on responsibilities for the reply fallback
germain-gg Feb 22, 2022
fa232f4
Update `/messages` API endpoint version on example
germain-gg Feb 22, 2022
68d9c42
Apply wording suggestions from code review
germain-gg Feb 23, 2022
5bbb015
Add notes on server-side invalid relation filtering
germain-gg Feb 22, 2022
707af2b
Fix typo
germain-gg Feb 22, 2022
b28a365
reword paragraph about forwarding m.thread relation
germain-gg Feb 23, 2022
8f82dfa
Add unstable prefix for capability endpoint
germain-gg Feb 23, 2022
8f8be64
Re-order alternatives to match intro paragraph
germain-gg Feb 23, 2022
b6d8076
rework relation_senders and relation_types definition
germain-gg Feb 23, 2022
cd671ef
Apply wording suggestions from code review
germain-gg Feb 24, 2022
a61c01e
Clarify fallback mechanism
germain-gg Feb 24, 2022
362e661
Rename filter property names
germain-gg Feb 24, 2022
b831fb3
Change m.render_in to m.display_reply_fallback
germain-gg Feb 24, 2022
e2dde8e
Clarify what endpoints support the new filter
germain-gg Feb 25, 2022
e640f6b
Switch from /capabilities to /versions
germain-gg Feb 25, 2022
bda3a1e
remove references to Cerulean
germain-gg Feb 25, 2022
89c4b5e
Update latest_event description
germain-gg Feb 25, 2022
61bb518
Clarity in wording and fix typo
germain-gg Mar 1, 2022
9159a5a
rename m.display_reply_fallback to hide_reply
germain-gg Mar 1, 2022
a97307a
remove redundant paragraph about forward compat
germain-gg Mar 1, 2022
f541dab
Improve bundled relationship example
germain-gg Mar 1, 2022
82b4c62
Explain context on why a thread-unaware client might want to send m.t…
germain-gg Mar 1, 2022
75f4cb2
Clarify `hide_reply`
germain-gg Mar 4, 2022
54ce185
Rename hide_reply to show_reply
germain-gg Mar 8, 2022
6d6baa2
rename show_reply to is_falling_back
germain-gg Mar 8, 2022
893cf1f
Add note about stable support.
clokep Mar 8, 2022
641e326
Update proposals/3440-threading-via-relations.md
germain-gg Mar 9, 2022
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
385 changes: 385 additions & 0 deletions proposals/3440-threading-via-relations.md
@@ -0,0 +1,385 @@
# MSC3440 Threading via `m.thread` relation

## Problem

Threading allows users to branch out a new conversation from the main timeline of a room
to each other. This is particularly useful in high traffic rooms where multiple
conversations can happen in parallel or when a single discussion might stretch
over a very long period of time.

The main goal of implementing threads is to facilitate conversations that are easier
to follow and smoother to read.
Threading is very clearly a core requirement for any modern messaging
solution, and Matrix uptake is suffering due to the lack of progress.

## Proposal

### Event format

A new relation type (see [MSC2674](https://github.com/matrix-org/matrix-doc/pull/2674))
`m.thread` expresses that an event belongs to a thread.

```json
"m.relates_to": {
"rel_type": "m.thread",
"event_id": "$thread_root"
}
```
Where $thread_root is the event ID of the root message in the thread.

When a thread root is aggregated (as in MSC2675), it returns a summary of the thread:
the latest message, a list of participants and the total count of messages.
I.e. in places which include bundled relations (per
[MSC2675](https://github.com/matrix-org/matrix-doc/pull/2675)), the thread root
would include additional information in the `unsigned` field:
germain-gg marked this conversation as resolved.
Show resolved Hide resolved

```jsonc
{
"event_id": "$root_event",
"unsigned": {
"m.relations": {
"m.thread": {
"latest_event": {
"event_id": "$thread_event",
// ...
},
"count": 7,
"current_user_participated": true
}
}
}
}
```

* `latest_event`: The most recent event which relates to this event, with
`rel_type` of `m.thread`.
* `count`: An integer counting the number of `m.thread` events
* `current_user_participated`: A flag set to `true` if the current logged in user
has participated in the thread
clokep marked this conversation as resolved.
Show resolved Hide resolved

#### Rich replies in a thread

Rich replies are still handled via the `m.in_reply_to` field of `m.relates_to`.
However clients should specify that this is not a thread fallback by setting
the `is_falling_back` property to `false`.

```json
"m.relates_to": {
"rel_type": "m.thread",
"event_id": "$thread_root",
"is_falling_back": false,
"m.in_reply_to": {
"event_id": "$event_target"
}
}
```

It is possible that an `m.in_reply_to` event targets an event that is outside the
related thread. Clients should always do their utmost to display the rich reply
and when clicked, the event should be displayed and highlighted in its original context.

A rich reply without `rel_type: m.thread` targeting a thread relation must be
rendered in the main timeline. This will allow users to advertise threaded messages
germain-gg marked this conversation as resolved.
Show resolved Hide resolved
in the room.

### Backwards compatibility

A thread will be displayed as a chain of replies on clients unaware of threads.

Thread-ready clients should always include an `m.in_reply_to` property when sending
a threaded event. Unless the user is explicitly replying to another event (see "Rich replies in a thread", above),
the `m.in_reply_to` property should reference the latest message-like event in the
thread, and clients should also specify that `m.in_reply_to`
is a fallback mechanism (rather than a genuine reply) by setting the `is_falling_back` property to `true`.

(If omitted, `is_falling_back` defaults to `false`, and receiving clients will treat the
`m.in_reply_to` part of the event as a genuine reply.)

```jsonc
"m.relates_to": {
"rel_type": "m.thread",
"event_id": "ev1",
"is_falling_back": true,
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

So one thing I noticed when looking at the push rules again: I expected this key to be in the "m.in_reply_to" relation, not on the thread relation. Is that intentional or not? Could this possibly be moved to the relation, that is actually falling back, so in this case the reply? Otherwise this seems to be a bit ambiguous.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Moving it at this point would mean having to handle it in both cases indefinitely to not break the semantics of existing history.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Well, the only consequence of that change would be that some old events would render with a reply. Considering the short timeframe that the use of m.thread has been legal and that this is not in any spec release yet and won't be for 3 months, I think fixing that now would be the much, much smaller pain point. Clients could accept both for a few months in the transition period and after that old threads might have some extra replies in them, but since they are still marked as beta, that will be justifyable. If we don't fix it, we will have to introduce a new field to mark other relations as fallbacks in the future, which will be then something we will actually have to have compatibility with for years.

So no, I don't agree we would need to handle that indefinitely. Implementing a feature before it is in a spec release has some risk and the problems with that will be minimal. I can write an MSC for that though, if it is something we need to argue about.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think @deepbluev7 has a point, but as I just wrote on #3664: please can you open a new issue for it.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I wrote an MSC instead: #3825

"m.in_reply_to": {
"event_id": "last_event_id_in_thread",
}
}
```

Historically replies have been limited to text messages due to the legacy fallback
prepended to `formatted_body`. This MSC is dependant on
[MSC3676](https://github.com/matrix-org/matrix-doc/pull/3676) which strips that
requirement to unlock use of any event type in this context.

### Fetch all relations to a thread root

To fetch an entire thread, the `/relations` API can be used as defined in
[MSC2675](https://github.com/matrix-org/matrix-doc/pull/2675)

```
GET /_matrix/client/unstable/rooms/!room_id:domain/relations/$thread_root/m.thread
```
novocaine marked this conversation as resolved.
Show resolved Hide resolved

Where `$thread_root` is the event ID of the root message in the thread.

> Any API which receives events should bundle relations (apart from non-gappy
incremental syncs), for instance: initial sync, gappy incremental sync,
/messages and /context.

### Fetch all threads in a room

[Event filters](https://spec.matrix.org/v1.2/client-server-api/#filtering) (as
used by endpoints including `/messages`, `/sync` and `/context`) are extended
with new options to allow filtering events by their relating events:

* `related_by_rel_types`: A list of relation types to include. An event `A` is included
in the filter only if there exists another event `B` which relates to `A` with a
`rel_type` which is defined in the list
* `related_by_senders`: A list of senders to include. An event `A` is included in
the filter only if there exists another event `B` which relates to `A`, and
which has a `sender` which is in the list.

This can also be combined with the `sender` field to search for threads which a
user has participated in (or not participated in).

```
GET /_matrix/client/v3/rooms/!room_id:domain/messages?filter=...
```

The filter string includes the new fields, above. In this example, the URL
encoded JSON is presented unencoded and formatted for legibility:

```jsonc
{
"types": ["m.room.message"],
"related_by_senders": [
// ...
],
"related_by_rel_types": ["m.thread"]
}
```

Note that the newly added filtering parameters return events based on information
in related events. Consider the following events in a room:

* `A`: a `m.room.message` event sent by `alice`
* `B`: a `m.room.message` event sent by `bob` which relates to `A` with type `m.thread`

Using a filter of `"related_by_rel_types": ["m.thread"]` would return event `A` as it
has another event which relates to it via `m.thread`.

Similarly, using a filter of `"related_by_senders": ["bob"]` would return event `A`
as it has another event which relates to it sent by `bob`.

### Server capabilities

Threads might have sporadic support across servers, to simplify feature
detections for clients, a homeserver must advertise unstable support for threads
as part of the `/versions` API:

```jsonc
{
"unstable_features": {
"org.matrix.msc3440": true,
// ...
}
}
```

### Limitations
germain-gg marked this conversation as resolved.
Show resolved Hide resolved

#### Read receipts

Read receipts and read markers assume a single chronological timeline. Threading
changes that assumption making the current API not very practical.

Clients can synthesize read receipts but it is possible that some notifications get
lost on a fresh start where the clients have to start off the `m.read`
information received from the homeserver.

Synchronising the synthesized notification count across devices is out of scope and deferred to a later MSC.

#### Single-layer event aggregation

This MSC does not include support for nested threads.

Nested threading is out of scope for this proposal and would be the subject of
a different MSC.
A `m.thread` event can only reference events that do not have a `rel_type`

```jsonc
[
{
"event_id": "ev1",
// ...
},
{
"event_id": "ev2",
// ...
"m.relates_to": {
"rel_type": "m.thread",
"event_id": "ev1",
"is_falling_back": true,
"m.in_reply_to": {
"event_id": "ev1"
}
}
},
{
"event_id": "ev3",
// ...
"m.relates_to": {
"rel_type": "m.annotation",
"event_id": "ev1",
"key": "✅"
}
}
]
```

Given the above list of events, only `ev1` would be a valid target for an `m.thread`
relation event.
germain-gg marked this conversation as resolved.
Show resolved Hide resolved

Servers should reject attempts to send events with invalid thread relations via the
Client-Server API with an HTTP `400` status code and a
`M_UNKNOWN` error code.
Events received over federation should always be accepted without checking
the validity of the relations as it would break the extensibility of this proposal
in a future MSC.

This means that events with invalid thread relations can make their way into the
network, either due by malicious activity or buggy implementation. If a client
receives such events, it should hide them as soon as it can determine for certain
that the associated event is not a valid target.

Servers are expected to not filter out invalid `m.thread` relations from the results when
serving endpoints that deal with message relations. Clients that call those
endpoints should be aware that they may return events with invalid relations,
and deal with them appropriately.

### Client considerations

#### Sending `m.thread` before fully implementing threads
richvdh marked this conversation as resolved.
Show resolved Hide resolved

There will be clients that will not or can't support threads. Whether this is a
deliberate choice or because the system bridges to a platform that does not support
threads, there are a number of steps developer of those systems can take to ensure
continuity of conversation in the ecosystem.

Clients that do not offer a threading UI should behave as follows when replying, for
best interaction with those that do.
They should set the `m.in_reply_to` part as usual, and then add on
`"rel_type": "m.thread"` and `"event_id": "$thread_root"`, copying `$thread_root`
from the replied-to event.

If the `m.thread` relation type is not present in an incoming event, it should
be treated as not being part of the thread. For example, if a client has a
separate area for displaying threads, clients can render the event in the main
room timeline as a rich reply that will open and highlight the event in the
thread context when clicked.

When replying to the following event, a client that does not support threads should
copy in `rel_type` and `event_id` properties in their reply mixin.
germain-gg marked this conversation as resolved.
Show resolved Hide resolved

```jsonc
{
// ...
"m.relates_to": {
"rel_type": "m.thread",
"event_id": "ev1",
"is_falling_back": false,
"m.in_reply_to": {
"event_id": "$event_target"
}
}
}
```

germain-gg marked this conversation as resolved.
Show resolved Hide resolved
## Alternatives
germain-gg marked this conversation as resolved.
Show resolved Hide resolved

"Threading as rooms", building on `m.in_reply_to`, and [MSC2836](https://github.com/matrix-org/matrix-doc/pull/2836) are the main alternatives here.

It is also worth noting that relations in this MSC could be expressed using the
scalable relation format described in [MSC3051](https://github.com/matrix-org/matrix-doc/pull/3051).

### Threads as rooms

Threads as rooms could provide full server-side APIs for navigating trees of events,
and could be considered an extension of this MSC for scenarios which require that
capability

"Threads as rooms" is the idea that each thread could just get its own Matrix room.

Advantages to "Threads as rooms" include:
* May be simpler for client implementations
* Restricting events visibility as the room creator
* Ability to create read-only threads

Disadvantages include:
* Access control, membership, history visibility, room versions etc needs to be
synced between the thread-room and the parent room
* Harder to control lifetime of threads in the context of the parent room if
they're completely split off
* Clients which aren't aware of them are going to fill up with a lot of rooms.
* Bridging to non-threaded chat systems is trickier as you may have to splice
together rooms

### Threads via m.in_reply_to

The rationale for using a new relation type instead of building on `m.in_reply_to`
is to re-use the event relationship APIs provided by
[MSC2675](https://github.com/matrix-org/matrix-doc/pull/2675). The MSC3267 definition
of `m.reference` relationships could be updated to mention threads (perhaps by
using the key field from [MSC2677](https://github.com/matrix-org/matrix-doc/pull/2677)
as the thread ID), but it is clearer to define a new relation type. It is unclear
what impact this would have on [MSC3267](https://github.com/matrix-org/matrix-doc/pull/3267),
but that is unimplemented by clients.

A big advantage of relations over rich replies is that they can be server-side
aggregated. It means that a client is not bound to download the entire history of
a room to have a comprehensive list of events being part of a thread.

### Threads via serverside traversal of relationships MSC2836

Advantages include:
* Fits other use cases than instant messaging
* Simple possible API shape to implement threading in a useful way

Disadvantages include:
* Relationships are queried using `/event_relationships` which is outside the
bounds of the `/sync` API so lacks the nice things /sync gives you (live updates).
That being said, the event will come down `/sync`, you just may not have the
context required to see parents/siblings/children.
* Threads can be of arbitrary width (unlimited direct replies to a single message)
and depth (unlimited chain of replies) which complicates UI design when you just
want "simple" threading.
* Does not consider use cases like editing or reactions

## Security considerations
germain-gg marked this conversation as resolved.
Show resolved Hide resolved

None

## Unstable prefix

Clients and servers should use list of unstable prefixes listed below while this
MSC has not been included in a spec release.

* `io.element.thread` should be used in place of `m.thread` as relation type
* `io.element.thread` should be used in place of `m.thread` as a capability entry
* `io.element.relation_senders` should be used in place of `related_by_senders`
in the `RoomEventFilter`
* `io.element.relation_types` should be used in place of `related_by_rel_types`
in the `RoomEventFilter`
* `io.element.show_reply` should be used in place of `is_falling_back`

While this MSC is considered stable, but not in v1.3 of the specification, clients
should look for `org.matrix.msc3440.stable` as an unstable feature flag as part of
the `/versions` API response to determine server support for the stable identifiers.
Comment on lines +375 to +377
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

just to ack this here too: we (the SCT) are actively committing to v1.3 having threads, so are okay with this.


## Dependencies

This MSC builds on [MSC2674](https://github.com/matrix-org/matrix-doc/pull/2674),
[MSC2675](https://github.com/matrix-org/matrix-doc/pull/2675),
[MSC3567](https://github.com/matrix-org/matrix-doc/pull/3567) and,
[MSC3676](https://github.com/matrix-org/matrix-doc/pull/3676) (which at the
time of writing have not yet been accepted into the spec).