A Minimalistic Wrapper for IndexedDB
Fixes #1955 for real this time (the fix in 4.0.2 wasn't correct enough for npm)
After 3 years is alpha and beta, and a week in RC, Dexie@4 is finally out!
It's still recommended to increment version number when schema has been updated before publishing new versions of an application since it can make opening the db fractions of ms faster. But it's not required.
(If the native version has diverged from the declared version - which happen when extending schema without incrementing version, dexie need to increment the native version to extend schema - and after a page refresh dexie will detect diverged version by catching VersionError and will redo open with the exact installed version under the hood. When measuring the extra time to open an older version in indexedDB and catch VersionError 10,000 times, it only took 1000 ms - 0.1 ms per try (on a macbook pro), so it should be negligible (https://jsfiddle.net/dfahlander/9tg13fwk/19/))
Non-transactional live queries utilize memory cache to assist in query resolution. It is possible to opt-out from using the cache using the option {cache: 'disabled'}. The cache can greatly improve simple pure range-based live queries and reduce the queries towards IndexedDB by reusing common queries from different components. The cache will be continously developed and optimized further to improve the new paging support in Dexie 5.0.
Dexie 4 catches various issues in the IndexedDB support for Chrome and Safari. Without dexie@4 transactions and write operations can suddenly fail in modern versions of Chrome and Safari but dexie makes sure to protect the end user from experience any issues. It does this by reopening the database or redoing the transaction when these things happen - all without the developer having to worry about it. See issues #543 and #613. (A close upcoming release on the 4.0-track will also address #1660 and #1829).
With Dexie 4.0, you can optionally use dexie-cloud-addon and connect your local database with a cloud based database. Dexie Cloud is a complete solution for authorization and synchronization of personal data with support for sharing data between users.
Dexie 4 has some improved typing. Specifically, Collection.modify() and Table.update() use template literals to suggest code completion and type checking of keypaths to update.
The type passed to db.table.add() can be different from the type returned from db.table.get() and tb.table.toArray(). This mirrors reality better since some properties may be optional when adding (such as an auto-incremented primary key or dexie-cloud properties realmId and owner).
The generic type Table<T, TKey> is extended with a 3rd optional parameter TInsertType making it possible to declare a table as such:
class MyDB extends Dexie {
friends!: Table<
{ id: number; name: string; age: number }, // T (id is always there)
number, // TKey (type of primary key)
{ id?: number; name: string; age: number } // TInsertType (id is optional)
>;
}
A new helper generic EntityTable<T> also exists as a "don't repeat yourself" sugar on top of this, and will also omit any method on the type. In case the type is a mapped class with helper methods, you are still able to pass simple POJO objects
to add() and put() with only data properties.
interface Friend {
id: number;
name: string;
age: number;
}
class MyDB extends Dexie {
friends!: EntityTable<Friend, 'id'>; // Automatically makes `id` optional when adding and picks its TKey type.
}
If using Dexie Cloud, there's a DexieCloudTable<T> generic that will declare the implicit Dexie Cloud properties on the entity correctly for insert- and output types:
interface Friend {
id: string;
name: string;
age: number;
realmId: string;
owner: string;
}
class MyDexieCloudDB extends Dexie {
friends!: DexieCloudTable<Friend, 'id'>; // Makes id, realmId and owner optional on insert operations.
}
In dexie 3, a mapped class that needs to access the database from its methods, will needs to access the a db instance. But it is a bit awkward to tie a method body to the same instance exported from the db module. For example, having two instances of db with different constructor parameters would not work as both would use one of the instances from their method bodies.
In Dexie 4.0, this is solved using a lightweight dependency injection. There is a generic class Entity that your classes may extend from in order to get your database instance injected as a protected property db on every entity instance. Your class methods can then perform queries onto the db without being dependant on the db module of your app. This use case will require a subclassed Dexie declaration though. There will still be cyclic import dependencies but only on the type level.
export class Friend extends Entity<AppDB> {
id!: number;
name!: string;
age!: number;
async birthday() {
return await this.db.friends.update(this.id, (friend) => ++friend.age);
}
}
No support for Internet Explorer 11, legacy Edge (non-chromium) nor legacy Safari below version 14.
This is the first release candidate of dexie@4. Please upgrade your dexie@3 to this to help us verify that it is backward compatible and works well for you application. Unless any critical issue shows up within a 5 days, [email protected] will be released as the latest stable version of dexie. At that point, release notes will contain all the news in dexie@4 compared to dexie@3. These changes are briefly covered in https://dexie.org/roadmap/dexie4.0
Please file issue as soon as possible if you find any breaking change or critical issue with this release that would prevent it from being released as the latest stable version of dexie.
Bugfix: Removed invalid dependency. Dexie has no dependencies #1924
NOTE: Prepare for [email protected] release candidate soon! It will have better stability by working around various browser issues and nicer version handling.
parentPath pattern:One pattern for managing tree structures in a database is to have an indexed property representing the path to the parent node, such as parentPath. This makes it efficient to delete or list all descendants in one query without any need of recursion:
// Add new node
function addNode(childProps, parentId = null) {
return db.transaction('rw', db.treeNodes, async ()=> {
const parent = parentId && await db.treeNodes.get(parentId);
await db.treeNodes.add({
...childProps,
parentPath: parent
? `${parent.parentPath}${parent.id}/`
: '' // If no parent, parentPath will be empty string (added at the root)
});
}
}
// List direct children
function listChildren(node) {
return db.treeNodes.where({parentPath: `${node.parentPath}${node.id}/`}).toArray();
}
// List all descendants without recursion:
function listAllDescendants(node) {
return db.treeNodes.where('parentPath').startsWith(`${node.parentPath}${node.id}/`).toArray();
}
// Load parent
function loadParent(node) {
return db.treeNodes.get(node.parentPath.split('/').at(-2));
}
// Load all ancestors
function async loadAllAncestors(node) {
return (node.parentPath
? await db.treeNodes.bulkGet(node.parentPath.substring(0, node.parentPath.length - 1).split('/'))
: []
);
}
// Delete the node and all its descendants:
function deleteNode(node) {
return db.transaction('rw', db.treeNodes, () => {
db.treeNodes.where('parentPath')
.startsWith(`${node.parentPath}${node.id}/`)
.delete();
db.treeNodes.where({
parentPath: node.parentPath, // for consistence
id: node.id
}).delete();
});
}
// NOTE:
// Where schema is defined, use a compound index on '[parentPath+id]':
// db.version(x).stores({
// treeNodes: 'id, [parentPath+id]' // enables where({parentPath: X, id: Y})
// });
All the mutating operations above are also sync consistent: If one offline client adds a child under "/a/b/c" and another offline client modifies all descendants under "/a/b/c" to have a new property {color: "blue"}, the merge of these operation will set {color: "blue"} on the added child also no matter in which order the clients became online. This works due to Dexie Cloud's built-in CRDT capabilities without the user having to use certain types in the objects they are storing.
But modify-operations that use a JS callback does not benefit from sync consistency, only local consistency. Moving an entire tree from one node to another has been one of those operation that need a JS function because the new parentPath's value depends on its existing value. So a tree move has only been possible with local consistency but not sync consistency before:
function moveNode(node, newParentPath) {
return db.transaction('rw', db.treeNodes, () => {
// Move node. Having the parentPath criteria here is for consistency:
// Only perform the 2 moves if parentPath is still the same.
db.treeNodes.where({
parentPath: node.parentPath,
id: node.id
}).modify({
parentPath: newParentPath
});
// Move all its descendants in relation to their sub path:
db.treeNodes.where('parentPath').startsWith(`${node.parentPath}${node.id}/`).modify(node => {
// This is a JS callback that cannot be expressed to the server
node.parentPath = newParentPath + node.parentPath.substring(node.parentPath.length);
});
});
}
JS code cannot securely be sent to the server due to several reasons: closures information missing + the risk of sending code that injects arbitrary code on the server. The operation above won't be sync consistent. If an offline client did add a new node under /a/b/c and that node is moved to /x/y/z, the merging of these operations would not be consistent - a node would still be placed under /a/b/c even though the c node has moved and doesn't exist anymore. So there is a need for declarative ways of doing certain operations, and moving trees is one of them.
A new export replacePrefix now available for this purpose. We will add more of these in coming versions (such as increment, push, deleteArrayItem, etc). replacePrefix is the first "complex" operation that already has support in Dexie Cloud (if upgrading dexie-cloud-addon to 4.0.1-beta.58) and can be executed to perform sync consistent tree moves:
import { replacePrefix } from 'dexie';
// Move subtree with sync consistency:
function moveNode(node, newParentPath) {
return db.transaction('rw', db.treeNodes, () => {
// Move node
db.treeNodes.where({
parentPath: node.parentPath, // consistency-check
id: node.id
}).modify({
parentPath: newParentPath
});
// Move all its descendants in relation to their sub path:
db.treeNodes
.where('parentPath')
.startsWith(`${node.parentPath}${node.id}/`)
.modify({
// Here we're in a declarative object - not a JS callback - ==> Consistent operation.
parentPath: replacePrefix(node.parentPath, newParentPath);
});
});
}
This transaction does it all - it moves the node and all its descendants consistently in one atomic all-or-nothing transaction but also preserves the where-condition and the replacePrefix operation in the information to the server so that if an offline client added a node under /a/b/c, and then this operation happened, moving all descendants to the new location, and then the offline client comes online again and syncs, the new node would be placed on the correct new location and be hanging below a non-existing /a/b/c.
If two operations competes in a move operation involving the same nodes, consistency is maintained by the extra criteria on parentPath in both the move of the parent node and the operation to move its descendants. A move operation will not be performed if another move operation came first. The end result will always be a consistent tree. The last syncer might then experience that the move operation they made got rolled back after a sync and they would instead see the peer's hierarchy.
Minor fix: Since 4.0.1-beta.10, we're allowing argument { disableAutoOpen: false} to db.delete() (not yet in docs), but it still failed if doing it.
Bugfix