Impagination

Put the fun back in lazy, asynchronous, paged, datasets.

Impagination is a lazy data layer for your paged records. All you provide Impagination is the logic to fetch a single page, plus how many records you want it to pre-fetch ahead of you.

Impagination is built using an event-driven immutable style, so it is ideal for use with UI frameworks like Ember, Angular, or React. That said, it has zero dependencies apart from JavaScript, so it can be used from node as well.

Installation

npm install impagination

Upgrading

If you are upgrading Impagination to the 1.0 release, consider checking out the Migration Guide

Usage

To get started, create a dataset. There are only two required parameters fetch, and pageSize:

import Dataset from 'impagination';

let dataset = new Dataset({
  pageSize: 5, // num records per page
  loadHorizon: 10, // window of records to keep (default: pageSize)
  fetch: function(pageOffset, pageSize, stats) { // How to `fetch` a page
    stats.totalPages = 4;
    // Returns a `thenable` which resolves with page's `records`
    return $.ajax({ method, url });
  },
  unfetch: function(records, pageOffset) {} // invoked whenever a page is unloaded
  filter: function(element, index, array) {} // filters `records` whenever a page resolves
  observe: function(nextState) { // invoked whenever a new `state` is generated
    dataset.state = nextState;
  }
});

Calling new Dataset() will emit a state immediately. However this state will be empty.

dataset.state.length //=> 0;

let record = dataset.state.getRecord(0); // Empty Record
record.isRequested //=> false
record.isPending //=> false
record.isResolved //=> false
record.content //=> null

To start fetching pages and build the state, we need to start reading from an offset. To do this, we will update the dataset's readOffset.

dataset.setReadOffset(0);

With a pageSize of 5, this will immediately call fetch twice (for records 0-4, and 5-9), and emit a new state indicating that these records are in flight.

dataset.setReadOffset(0);
dataset.state.length //=> 10;

// Records 0-9 are Pending Records
let record = dataset.state.getRecord(0);
record.isRequested //=> true
record.isPending //=> true
record.isResolved //=> false
record.content //=> null

Load Horizon

How did it know which records to fetch? The answer is in the loadHorizon parameter that we passed into the constructor. We set the loadHorizon to 10. This tells the dataset, that it should keep all records within 10 of the current read offset loaded. That's why it fetched the first two pages.

I hope this ASCII Dataset adds some clarity:

dataset = new Dataset(...).setReadOffset(0); // builds the dataset below

           Read
          Offset
             ┃
             ┃
<──────────Load Horizon──────────>
             ┃
             ▼
             ┌ ─ ─ ─ ─ ─ ─ ─ ─ ─ ┐
              * * * * * * * * * *
             ◇ ─ ─ ─ ─ ◇─ ─ ─ ─  ┘
             │         │         │
             p0        p1        length = 10

ASCII  Legend:
`xx` - unrequested record
`*`  - pending record
`n`  - record inex
`◇`  - page boundary

Resolving Asynchrnous Pages

Once the asynchronous fetch for a page resolves, the dataset will emit a new state with the updated resolved records.

Continuing our previous example, we assume the the request on page 0 resolves and the request on page 1 is not yet resolved. That state will still contain the resolved records as well the pending records.

dataset.state.length //=> 20;

// Assumes the page `0` resolves and page `1` is pending
let record = state.getRecord(0);
record.page.offset = 0;
record.isPending //=> false
record.isResolved //=> true
record.content //=> { name: 'Record 3' }

record = state.get(5)
record.page.offset = 1;
record.isPending //=> true
record.isResolved //=> false
record.content //=> null

Another interesting thing that happened here is that the length of the dataset has also changed.

dataset.state.length //=> 20 (stats.totalPages: 4, pageSize: 5)

That's because the stats parameter that is passed into our example fetch function tells our dataset there are 5 total pages in our dataset. This value allows the fetch function to optionally specify the total extent of the dataset if that information is available. This can be useful when rendering native scrollbars or other UI elements that indicate the overall length of a list. If stats are never updated, then the dataset will just expand indefinitely. Now our state looks like this:

            Read
           Offset
              ┃
              ┃
<──────────Load Horizon──────────>
              ┃
              ▼
              ┌ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─  ─ ─ ─ ─ ─ ─ ─ ── ─ ┐
               0 1 2 3 4 * * * * * xx xx xx xx xx xx xx xx xx xx│
              ◇ ─ ─ ─ ─ ◇ ─ ─ ─ ─ ◇ ─ ─ ─ ─ ─ ─  ◇ ─ ─ ─ ─ ─ ─  ┘
              │         │         │              │              │
             p0        p1        p2             p3              length = 20

We have records 0-4, whilst records 5-9 are in flight, and records 10-20 have yet to be requested.

//from the last ◇ page (p3)
record = state.getRecord(17);
record.isRequested //=> false
record.isPending //=> false
record.content //> null

Dataset API

There are a number of public impagination functions which we provide as actions to update the dataset.

Updating the Dataset

| Actions | Parameters | Description | | ------------- |:--------------:|:--------------| | refilter | [filterCallback] | Reapplies the filter for all resolved pages. If filterCallback is provided, applies and sets the new filter. | reset | [offset] | Unfetches all pages and clears the state. If offset is provided, fetches records starting at offset. | setReadOffset | [offset] | Sets the readOffset and fetches records resuming at offset

Updating the State

| Actions| Parameters | Defaults |Description | | ------ |:-----------:|:--------------|:--------------| | post | data, index | index = 0 | Inserts data into state at index. | put | data, index | index = state.readOffset | Merges data into record at index. | delete | index | index= state.readOffset | Deletes data from state at index.

setReadOffset Example

Let's say the we change our viewport to item 2 in our UI. We want to tell impagination to move the read head to offset 2 with a call to dataset.setReadOffset(2). This will immediately emit a new state that looks like this:

                 Read
                Offset
                   ┃
                   ┃
     <──────────Load Horizon──────────>
                   ┃
                   ▼
              ┌ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─  ─ ─ ─ ─ ─ ─ ── ─ ┐
               0 1 2 3 4 * * * * *  * * * * *   xx xx xx xx xx│
              ◇ ─ ─ ─ ─ ◇ ─ ─ ─ ─ ◇ ─ ─ ─ ─ ─  ◇ ─ ─ ─ ─ ─ ─  ┘
              │         │         │            │              │
             p0        p1        p2            p3             length = 20

You'll notice that the page at offset p2 has now been requested because it contains records that fall within the load horizon. The page at p1 is still pending, but now p2 is as well. What happens if the request for p2 resolves before the request for p1? In that case, the dataset emits this state:

                 Read
                Offset
                   ┃
                   ┃
     <──────────Load Horizon──────────>
                   ┃
                   ▼
              ┌ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─  ─ ─ ─ ─  ── ─ ── ─ ┐
               0 1 2 3 4 * * * * * 10 11 12 13 14 xx xx xx xx xx│
              ◇ ─ ─ ─ ─ ◇ ─ ─ ─ ─ ◇ ─ ─ ─ ─ ─ ─  ◇ ─ ─ ─ ─ ─ ─  ┘
              │         │         │              │              │
             p0        p1        p2             p3              length = 20

In this way, impagination is resilient to the order of network requests because the records are "always available" and in their proper order, albeit in their unrequested, pending, or resolved states.

//records on p2 are now available
record = state.getRecord(10);
record.isResolved //=> true
record.content //=> 10

//records on p1 are still pending
record = state.getRecord(5);
record.isResolved //=> false
record.isPending //=> true
record.content //=> null

Filtering Records

We fetch records using an immutable style, but we often require filtering by mutable values in our dataset. To enable filtering, pass a filter callback to impagination as you would to Array.prototype.filter(). The filters are applied as soon as a page is resolved. To filter a page at other times in your application see refilter.

Here we filter by records whose content contains an even number

let dataset = new Dataset({ ...
  // filter() function which returns only _even_ records
  filter: function(content) { return content % 2 === 0 }
});

                 Read
                Offset
                   ┃
                   ┃
     <──────────Load Horizon──────────>
                   ┃
                   ▼
              ┌ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─  ─ ─ ─ ─  ── ─ ── ─ ┐
               0   2   4 * * * * * ** ** ** ** ** xx xx xx xx xx│
              ◇ ─ ─ ─ ─ ◇ ─ ─ ─ ─ ◇ ─ ─ ─ ─ ─ ─  ◇ ─ ─ ─ ─ ─ ─  ┘
              │         │         │              │              │
             p0        p1        p2             p3              length = 18

// Finding even numbered records
record = state.getRecord(1);
record.isResolved //=> true
record.page.offset //=> 0
record.content //=> 2
state.length //=> 18 (stats.totalPages: 4, pageSize: 5, rejected records by filter: 2)

//records on p1 are still pending
record = state.getRecord(3); // The record at index 3 now exists on p1
record.isResolved //=> false
record.isPending //=> true
record.page.offset //=> 1
record.content //=> null

Impagination and Immutability

In the mutable style of reactivity, you listen to events that report what changed about a datastructure, and then you're left to realize the implications of that change in your internal data structures (such as changing a record from isPending to isResolved). By contrast, Impagination uses an immutable style.

In Impagination, each event is the fully formed datastructure in its entirety. This eliminates all guesswork and ambiguity from what the implications are so that you, the developer, have to do less work to maintain consistency.

What this means in practice is that each of the states observed by the observe function are unique structures that are considered immutable. Each one stands alone and will continue to function properly even if you discard references to all other states and the dataset object itself. Furthermore, altering them will have no effect on neither prior nor subsequent states.

You may be asking, is it not wasteful to recreate an entire potentially infinite data structure with every state transition? The answer is that each state is lazy and stores as little information as it needs to provide its API. The state contains lazy array interfaces for pages and records.