Phase 4: sync

Implements PLAN.md Phase 4: the read model. /sync, /messages and /context,
the filters that narrow them, and the streams a response carries that are
not room events — typing, receipts, presence and to-device messages. With
this a client can be used against Warp, which is milestone M1.

The one architectural choice is that the syncapi has no copy of the rooms:
it reads them through the roomserver. Dendrite gives its syncapi its own
tables and feeds them from the output stream, which is right for a design
where the two can be separate processes, and Dendrite's could be. Warp is
one process by decision, and there the copy buys nothing and costs the
hardest class of bug a homeserver has — a read model that disagrees with
the room, where a client is shown a message that was rejected and the only
way to find out is to compare two tables. So the roomserver grew a read
model (stream.go, roomsync.go, visibility.go) rather than exporting its
store: reading an event means parsing it under its room's version, and a
second copy of the version registry, the parser and the client projection
would be a second place for them to disagree.

syncapi: stream tokens naming a position in each of five independent
streams, so a client that has seen receipt 7 but only event 4 can say so in
one opaque string. A token with fewer positions than this build writes is
accepted and the rest read as zero, which is what lets a token stored
before a stream existed go on working. The PDU position comes from
MAX(stream_ordering) and not from the counter row the other four use: the
counter is advanced inside the writing transaction, so a token taken from
it would name a position whose event no reader can see yet, and the event
would be skipped for good once that transaction committed.

Long polling goes through a notifier that wakes by room and by user rather
than globally — an event wakes the listeners of its room, a membership
event also wakes the user it names, so an invite reaches a client that is
not in the room yet. A wake-up is only ever a hint: the listener re-reads
the streams and decides for itself what changed, which is what makes it
safe for the notifier to hold no state about what each client has seen. The
listener registers before the read, so an event landing in between leaves
the channel signalled and the sync reads again rather than blocking on a
change it just missed. EventPersisted must not block, because it is called
with the room's write lock still held.

"Something happened" is not "a token moved". A stream can move without this
user being concerned by it, and returning on every such move would turn a
long poll back into a busy loop. The same trap caught presence twice over:
a client says it is still online by syncing, so ?set_presence= makes every
sync a presence write, and a stream that moved each time would wake every
peer, which would sync, which would move it again. SetPresence therefore
reports whether anything a peer would notice actually changed, and a write
that only refreshes the last-active timestamp leaves the stream alone.

Timeline windows are named by their two ends and a direction, Low exclusive
and High inclusive in both directions, rather than by "from" and "to" as
the wire format spells them. Each of those would otherwise be the low bound
in one direction and the high bound in the other, which is exactly the
off-by-one every pagination bug is made of; the handlers translate once and
nothing below them thinks about it again. A sync's timeline is read
backwards and reported forwards, because a client wants the newest events
of the window it missed — a forward read with a limit hands it the oldest
and then calls the rest a gap.

visibility: history visibility is now a span rather than a yes or no, which
is what Phase 3 deferred here. A user who has left a room is served the
room as it was when they left — their state reads, their members and their
timeline — instead of being refused, and is bounded above at their own
membership event so a kicked client cannot go on streaming. Their span is
found from their own member events, so a user who joined, left and rejoined
sees their latest span. ?at= on /members is honoured for the same reason.
What is left is that the span is computed from the room's current
history_visibility where the spec judges each event by the one in force
when it was sent; that is a state read per event and belongs with open
question 1.

Typing is held in memory and nowhere else: notifications expire in seconds,
are worthless across a restart, and arrive one per keystroke-burst per
room, so persisting them would be the busiest table in the database holding
the least valuable data in it. Receipts only ever advance, compared on the
position of the event acknowledged rather than on arrival, because two of a
user's clients read at different speeds and the slower one must not drag
the marker backwards. To-device queues are per device and drained by the
next sync rather than by the one that delivered them, so a client that
crashes in between gets the batch again — the right way round, since a lost
key share leaves a message permanently undecryptable and a duplicate is
ignored. A wildcard addressee is expanded when sent, not stored, because
"every device" means the ones that existed then.

Presence is off by default, which answers PLAN.md open question 3. It is
the noisiest thing in Matrix and the one genuinely fan-out notification
here; a private homeserver gains little from it. With it off the endpoints
refuse rather than disappearing, because a client that gets a 404 from
/presence concludes the server is too old to have it and stops asking.

storage: one migration per driver adding the receipt, to-device, presence
and filter tables with a counter row each, and a stream position on the
existing account data table rather than a parallel one — there is exactly
one current value per (user, room, type), and a second table would be a
second thing to keep in step. Account data is reported in an event list but
is not an event, so it gets its own response type: a room event with an
empty sender and no timestamp would make a client decide what that meant.

Deliberately missing, each named in docs/sync.md: notification counts (they
need push rules, Phase 8), device lists (the keyserver, Phase 7), anything
remote (Phase 5), threads (Phase 10). device_lists is sent as two empty
lists because a client reads a missing one as malformed; the notification
counts are left out entirely, because a zero there is a claim and an absent
field is not.

Tested through HTTP rather than against the builder, because what is worth
pinning down is the response a client sees — which section a room lands in,
whether a token handed back can be handed in again, whether paging
terminates and loses nothing. The storage tests run against both SQL
drivers; the notifier has its own tests for the concurrent case. A live
server was driven end to end, which is what caught the two bugs the unit
tests had missed: a backwards /messages page must be in reverse
chronological order, and `start` must name where the page began rather than
position zero when ?from= was omitted.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

server/internal/roomserver/stream.go

@@ -0,0 +1,481 @@
+package roomserver
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"slices"
+
+	"github.com/dariuszjaniak/warp/server/internal/matrix/event"
+	"github.com/dariuszjaniak/warp/server/internal/matrix/roomversion"
+	"github.com/dariuszjaniak/warp/server/internal/storage"
+)
+
+// The read model: the event stream as /sync, /messages and /context read
+// it.
+//
+// It is here rather than in the syncapi because reading an event means
+// parsing it under its room's version, and that is the roomserver's
+// business — the syncapi would otherwise need its own copy of the room
+// version registry, the event parser and the client projection, and a
+// second copy of those would be a second place for them to disagree. What
+// the syncapi gets instead is windows of ClientEvents with their positions,
+// and it assembles the response shape around them.
+
+// MembershipAt is a user's current membership in a room, with the position
+// of the event that set it.
+type MembershipAt struct {
+	RoomID     string
+	Membership string
+	EventID    string
+
+	// Position is the stream position of the membership event.
+	Position int64
+}
+
+// TimelineRequest is one window of a room's event stream.
+//
+// The bounds are the store's: Low exclusive, High inclusive, both
+// directions, and zero meaning unbounded at that end. Direction decides
+// only which end the limit trims and the order the store reads in — the
+// events always come back oldest first.
+type TimelineRequest struct {
+	RoomID string
+
+	// User is the full user ID the window is read on behalf of. The result
+	// is clamped to what their membership and the room's history
+	// visibility allow.
+	User string
+
+	Low  int64
+	High int64
+
+	Direction storage.TimelineDirection
+
+	// Limit caps how many events come back.
+	Limit int
+
+	// Filter narrows which events are read. Applying it here rather than
+	// after the limit is what keeps a filtered page from coming back short
+	// and looking like the end of the room.
+	Filter storage.TimelineFilter
+}
+
+// Timeline is a window of a room's event stream.
+type Timeline struct {
+	// Events are in stream order, oldest first, whichever direction was
+	// asked for: a client renders a timeline forwards even when it paged
+	// backwards to get it.
+	Events []ClientEvent
+
+	// Start and End are the positions bounding the window: Start below the
+	// first event and End at the last. They are the two tokens a client
+	// needs to ask for the next page in either direction, and either can
+	// be handed back as a bound without re-reading the event it names.
+	Start int64
+	End   int64
+
+	// Limited reports that the limit cut the window short, so there is
+	// more of it beyond the end the limit trimmed.
+	Limited bool
+}
+
+// StreamTop returns the top of this server's event stream.
+func (r *API) StreamTop(ctx context.Context) (int64, error) {
+	return r.store.MaxStreamOrdering(ctx)
+}
+
+// UserMemberships returns the rooms a user has one of the given memberships
+// in, with the position of each membership event.
+func (r *API) UserMemberships(ctx context.Context, userID string, memberships []string) ([]MembershipAt, error) {
+	stored, err := r.store.MembershipsForUser(ctx, userID, memberships)
+	if err != nil {
+		return nil, err
+	}
+	out := make([]MembershipAt, 0, len(stored))
+	for _, m := range stored {
+		out = append(out, MembershipAt{
+			RoomID:     m.RoomID,
+			Membership: m.Membership,
+			EventID:    m.EventID,
+			Position:   m.StreamOrdering,
+		})
+	}
+	return out, nil
+}
+
+// RoomsWithEvents returns the subset of rooms that have an event after from
+// and at or before to.
+func (r *API) RoomsWithEvents(ctx context.Context, roomIDs []string, from, to int64) ([]string, error) {
+	return r.store.RoomsWithEvents(ctx, roomIDs, from, to)
+}
+
+// Timeline reads one window of a room's event stream, as far as the user is
+// allowed to see it.
+func (r *API) Timeline(ctx context.Context, req TimelineRequest) (Timeline, error) {
+	room, rules, err := r.roomAndRules(ctx, req.RoomID)
+	if err != nil {
+		return Timeline{}, err
+	}
+	visible, err := r.requireReadable(ctx, room.ID, req.User)
+	if err != nil {
+		return Timeline{}, err
+	}
+	return r.timeline(ctx, rules, visible, req)
+}
+
+// timeline is Timeline with the room's rules and the user's visible range
+// already in hand, for the sync loop that reads many rooms at once.
+func (r *API) timeline(ctx context.Context, rules roomversion.Rules, visible VisibleRange, req TimelineRequest) (Timeline, error) {
+	// An empty window is reported at the bound the caller came in on, so
+	// that the tokens it hands back are still usable.
+	empty := Timeline{Start: req.Low, End: req.Low}
+	if req.Direction == storage.DirectionBackward && req.High > 0 {
+		empty = Timeline{Start: req.High, End: req.High}
+	}
+	if req.Limit <= 0 {
+		return empty, nil
+	}
+
+	low, high, any := visible.Clamp(req.Low, req.High)
+	if !any {
+		return empty, nil
+	}
+
+	// One more than asked for, so that "there is more" can be told from
+	// "that was all" without a second query.
+	stored, err := r.store.Timeline(ctx, storage.TimelineRequest{
+		RoomID:    req.RoomID,
+		Low:       low,
+		High:      high,
+		Direction: req.Direction,
+		Limit:     req.Limit + 1,
+		Filter:    req.Filter,
+	})
+	if err != nil {
+		return Timeline{}, err
+	}
+
+	out := empty
+	if len(stored) > req.Limit {
+		out.Limited = true
+		stored = stored[:req.Limit]
+	}
+	if len(stored) == 0 {
+		return out, nil
+	}
+
+	// The store returned them in the direction asked for; a client always
+	// renders a timeline oldest first.
+	if req.Direction == storage.DirectionBackward {
+		slices.Reverse(stored)
+	}
+
+	out.Events = make([]ClientEvent, 0, len(stored))
+	for _, s := range stored {
+		e, err := parseEvent(s, rules)
+		if err != nil {
+			return Timeline{}, err
+		}
+		out.Events = append(out.Events, r.clientEvent(s, e, req.RoomID))
+	}
+	// Start sits below the oldest event, so that handing it back as a High
+	// bound returns the page before this one; End is at the newest, so that
+	// handing it back as a Low bound returns the page after.
+	out.Start = stored[0].StreamOrdering - 1
+	out.End = stored[len(stored)-1].StreamOrdering
+	return out, nil
+}
+
+// EventPosition returns an event's position in the stream, for a user who
+// may see it.
+//
+// /context needs it: the timeline reads are bounded by position rather than
+// by event ID, so the event a context is centred on has to be located
+// before either side of it can be read.
+func (r *API) EventPosition(ctx context.Context, roomID, userID, eventID string) (int64, error) {
+	visible, err := r.requireReadable(ctx, roomID, userID)
+	if err != nil {
+		return 0, err
+	}
+	stored, err := r.store.Event(ctx, roomID, eventID)
+	if err != nil {
+		return 0, err
+	}
+	if !visible.Contains(stored.StreamOrdering) {
+		return 0, fmt.Errorf("event %q: %w", eventID, ErrNoSuchEvent)
+	}
+	return stored.StreamOrdering, nil
+}
+
+// StateAt returns a room's state as of a stream position, in the client
+// shape.
+//
+// It is what a sync with a gap in it and a leaver's state read both need:
+// the state current state cannot give, because the room has moved on since.
+func (r *API) StateAt(ctx context.Context, roomID, userID string, pos int64) ([]ClientEvent, error) {
+	room, rules, err := r.roomAndRules(ctx, roomID)
+	if err != nil {
+		return nil, err
+	}
+	visible, err := r.requireReadable(ctx, room.ID, userID)
+	if err != nil {
+		return nil, err
+	}
+	if visible.To != 0 && (pos == 0 || pos > visible.To) {
+		pos = visible.To
+	}
+	return r.stateAt(ctx, roomID, rules, pos)
+}
+
+// stateAt is StateAt with the checks already done.
+//
+// A position of zero is the room's current state, which is both cheaper to
+// read and what every caller that is up to date wants.
+func (r *API) stateAt(ctx context.Context, roomID string, rules roomversion.Rules, pos int64) ([]ClientEvent, error) {
+	var (
+		entries []storage.StateEntry
+		err     error
+	)
+	if pos == 0 {
+		stored, err := r.store.CurrentStateEvents(ctx, roomID)
+		if err != nil {
+			return nil, err
+		}
+		return r.clientEvents(stored, rules, roomID)
+	}
+	if entries, err = r.store.StateAtStream(ctx, roomID, pos); err != nil {
+		return nil, err
+	}
+	return r.clientEventsByID(ctx, roomID, rules, stateEntryIDs(entries))
+}
+
+// StateBeforeEvent returns the state of a room as of an event's
+// prev_events: the state /context reports alongside it.
+func (r *API) StateBeforeEvent(ctx context.Context, roomID, userID, eventID string) ([]ClientEvent, error) {
+	room, rules, err := r.roomAndRules(ctx, roomID)
+	if err != nil {
+		return nil, err
+	}
+	visible, err := r.requireReadable(ctx, room.ID, userID)
+	if err != nil {
+		return nil, err
+	}
+
+	stored, err := r.store.Event(ctx, roomID, eventID)
+	if err != nil {
+		return nil, err
+	}
+	if !visible.Contains(stored.StreamOrdering) {
+		return nil, fmt.Errorf("%w: %s may not see %s", ErrForbidden, userID, eventID)
+	}
+
+	entries, err := r.store.StateBefore(ctx, roomID, eventID)
+	if err != nil {
+		return nil, err
+	}
+	state, err := r.clientEventsByID(ctx, roomID, rules, stateEntryIDs(entries))
+	if err != nil {
+		return nil, err
+	}
+	// The event itself, when it is a state event, is part of the state
+	// after it but not before it. /context wants the state the event was
+	// sent against, so it is left out — except that a client rendering the
+	// event needs the sender's member event, which is in the snapshot
+	// already.
+	return state, nil
+}
+
+// MemberEvents returns the m.room.member events of specific users in a
+// room, in the client shape.
+//
+// Lazy-loading is what this is for: a client that asked for it is sent the
+// members who appear in the timeline it was given, rather than all of them.
+func (r *API) MemberEvents(ctx context.Context, roomID string, userIDs []string) ([]ClientEvent, error) {
+	if len(userIDs) == 0 {
+		return nil, nil
+	}
+	_, rules, err := r.roomAndRules(ctx, roomID)
+	if err != nil {
+		return nil, err
+	}
+	return r.memberEvents(ctx, roomID, rules, userIDs)
+}
+
+// memberEvents is MemberEvents with the rules in hand.
+func (r *API) memberEvents(ctx context.Context, roomID string, rules roomversion.Rules, userIDs []string) ([]ClientEvent, error) {
+	out := make([]ClientEvent, 0, len(userIDs))
+	for _, userID := range userIDs {
+		stored, err := r.store.CurrentStateEvent(ctx, roomID, event.TypeMember, userID)
+		switch {
+		case isNotFound(err):
+			// A member who appears in the timeline but is not in the
+			// room's current state: their membership was reset by state
+			// resolution, or the event is old enough to predate it. Not an
+			// error, and nothing to send.
+			continue
+		case err != nil:
+			return nil, err
+		}
+		e, err := parseEvent(stored, rules)
+		if err != nil {
+			return nil, err
+		}
+		out = append(out, r.clientEvent(stored, e, roomID))
+	}
+	return out, nil
+}
+
+// StrippedEvent is a piece of room state as it travels with an invite:
+// enough for a client to show the user what they are being asked to join,
+// and nothing that would let them read the room.
+//
+// It has no event ID and no timestamp on purpose — the spec's "stripped
+// state" is explicitly not an event, and a client must not put it in a
+// timeline.
+type StrippedEvent struct {
+	Type     string          `json:"type"`
+	StateKey string          `json:"state_key"`
+	Sender   string          `json:"sender"`
+	Content  json.RawMessage `json:"content"`
+}
+
+// strippedStateTypes is what the spec says an invite carries: the state a
+// client needs to render the room in an invite list.
+var strippedStateTypes = []string{
+	event.TypeCreate,
+	event.TypeJoinRules,
+	TypeCanonicalAlias,
+	TypeName,
+	TypeAvatar,
+	TypeTopic,
+	TypeEncryption,
+	event.TypeHistoryVisibility,
+}
+
+// StrippedState returns the stripped state of a room, plus the member
+// events of the users named.
+//
+// The member events are what makes an invite legible: a client showing
+// "Alice invited you" needs Alice's display name, and the invitee's own
+// member event carries the invite's reason.
+func (r *API) StrippedState(ctx context.Context, roomID string, members []string) ([]StrippedEvent, error) {
+	_, rules, err := r.roomAndRules(ctx, roomID)
+	if err != nil {
+		return nil, err
+	}
+	return r.strippedState(ctx, roomID, rules, members)
+}
+
+// strippedState is StrippedState with the rules in hand.
+func (r *API) strippedState(ctx context.Context, roomID string, rules roomversion.Rules, members []string) ([]StrippedEvent, error) {
+	var out []StrippedEvent
+
+	appendState := func(eventType, stateKey string) error {
+		stored, err := r.store.CurrentStateEvent(ctx, roomID, eventType, stateKey)
+		switch {
+		case isNotFound(err):
+			return nil
+		case err != nil:
+			return err
+		}
+		e, err := parseEvent(stored, rules)
+		if err != nil {
+			return err
+		}
+		key, _ := e.StateKey()
+		out = append(out, StrippedEvent{
+			Type:     e.Type(),
+			StateKey: key,
+			Sender:   e.Sender(),
+			Content:  e.Content(),
+		})
+		return nil
+	}
+
+	for _, eventType := range strippedStateTypes {
+		if err := appendState(eventType, ""); err != nil {
+			return nil, err
+		}
+	}
+	for _, member := range members {
+		if err := appendState(event.TypeMember, member); err != nil {
+			return nil, err
+		}
+	}
+	return out, nil
+}
+
+// RoomSyncSummary is the `summary` block of a room in a sync response.
+//
+// It exists for rooms with no name: a client showing a direct message has
+// to call it something, and the heroes are the members it names it after.
+type RoomSyncSummary struct {
+	// Heroes are up to five members of the room other than the user
+	// syncing, which is what a client falls back to for a room with no
+	// name and no canonical alias.
+	Heroes []string `json:"m.heroes,omitempty"`
+
+	JoinedMemberCount  int `json:"m.joined_member_count"`
+	InvitedMemberCount int `json:"m.invited_member_count"`
+}
+
+// maxHeroes is how many members a summary names. The spec says five.
+const maxHeroes = 5
+
+// RoomSummaryFor builds the summary block of a room for one user.
+func (r *API) RoomSummaryFor(ctx context.Context, roomID, userID string) (RoomSyncSummary, error) {
+	entries, err := r.store.MembershipEntries(ctx, roomID,
+		[]string{event.MembershipJoin, event.MembershipInvite})
+	if err != nil {
+		return RoomSyncSummary{}, err
+	}
+
+	var summary RoomSyncSummary
+	for _, entry := range entries {
+		switch entry.Membership {
+		case event.MembershipJoin:
+			summary.JoinedMemberCount++
+		case event.MembershipInvite:
+			summary.InvitedMemberCount++
+		}
+		if entry.UserID != userID && len(summary.Heroes) < maxHeroes {
+			summary.Heroes = append(summary.Heroes, entry.UserID)
+		}
+	}
+	return summary, nil
+}
+
+// clientEvents projects stored events into the client shape.
+func (r *API) clientEvents(stored []storage.StoredEvent, rules roomversion.Rules, roomID string) ([]ClientEvent, error) {
+	out := make([]ClientEvent, 0, len(stored))
+	for _, s := range stored {
+		e, err := parseEvent(s, rules)
+		if err != nil {
+			return nil, err
+		}
+		out = append(out, r.clientEvent(s, e, roomID))
+	}
+	return out, nil
+}
+
+// clientEventsByID reads events by ID and projects them.
+func (r *API) clientEventsByID(ctx context.Context, roomID string, rules roomversion.Rules, eventIDs []string) ([]ClientEvent, error) {
+	if len(eventIDs) == 0 {
+		return nil, nil
+	}
+	stored, err := r.store.EventsByID(ctx, roomID, eventIDs)
+	if err != nil {
+		return nil, err
+	}
+	return r.clientEvents(stored, rules, roomID)
+}
+
+// stateEntryIDs projects state entries onto the event IDs behind them.
+func stateEntryIDs(entries []storage.StateEntry) []string {
+	out := make([]string, 0, len(entries))
+	for _, entry := range entries {
+		out = append(out, entry.EventID)
+	}
+	return out
+}