vector-im-hydrogen-web/src/matrix/storage/idb/schema.js

import {iterateCursor} from "./utils.js";
import {RoomMember, EVENT_TYPE as MEMBER_EVENT_TYPE} from "../../room/RoomMember.js";

// FUNCTIONS SHOULD ONLY BE APPENDED!!
// the index in the array is the database version
export const schema = [
    createInitialStores,
    createMemberStore,
];
// TODO: how to deal with git merge conflicts of this array?


// how do we deal with schema updates vs existing data migration in a way that 
//v1
function createInitialStores(db) {
    db.createObjectStore("session", {keyPath: "key"});
    // any way to make keys unique here? (just use put?)
    db.createObjectStore("roomSummary", {keyPath: "roomId"});

    // need index to find live fragment? prooobably ok without for now
    //key = room_id | fragment_id
    db.createObjectStore("timelineFragments", {keyPath: "key"});
    //key = room_id | fragment_id | event_index
    const timelineEvents = db.createObjectStore("timelineEvents", {keyPath: "key"});
    //eventIdKey = room_id | event_id
    timelineEvents.createIndex("byEventId", "eventIdKey", {unique: true});
    //key = room_id | event.type | event.state_key,
    db.createObjectStore("roomState", {keyPath: "key"});
    db.createObjectStore("pendingEvents", {keyPath: "key"});
}
//v2
async function createMemberStore(db, txn) {
    const roomMembers = db.createObjectStore("roomMembers", {keyPath: [
        "roomId",
        "userId"
    ]});
    // migrate existing member state events over
    const roomState = txn.objectStore("roomState");
    await iterateCursor(roomState.openCursor(), entry => {
        if (entry.event.type === MEMBER_EVENT_TYPE) {
            roomState.delete(entry.key);
            const member = RoomMember.fromMemberEvent(entry.roomId, entry.event);
            if (member) {
                roomMembers.add(member.serialize());
            }
        }
    });
}

function migrateKeyPathToArray(db, isNew) {
    if (isNew) {
        // create the new stores with the final name
    } else {
        // create the new stores with a tmp name
        // migrate the data over
        // change the name
    }

    // maybe it is ok to just run all the migration steps?
    // it might be a bit slower to create a store twice ...
    // but at least the path of migration or creating a new store
    // will go through the same code
    // 
    // might not even be slower, as this is all happening within one transaction
}