Openhab-paper-ui-ng NPM

openHAB Paper UI (New generation)

This repository contains a design theme study for a new generation of openHAB paper UI. See https://davidgraeff.github.io/paperui-ng/.

Application screenshot

Please participate by editing files in this repository and make pull requests. Especially the help texts and introduction text, graphics and potential videos require help.

Local development

Install npm (node package manager)

Check this repository out
Call npm install
Call npm run dev to start a local webserver
Enter the IP:Port of your openHAB installation in paperui-ng

Check this repository out into your openHAB static html directory, usually under /etc/openhab2/html.
Call npm install
Call npm run build
Browse to http://127.0.0.1:8080/static/paperui-ng/dist (replace the IP and Port with your openHAB installation).

Paper UI

The original Paper UI is a Single Page Application (SPA) developed using a js framework called Angular.

The downside of chosing a js framework is that those come and go. Angular has a step learning curve, is not versioned semantically, does breaking changes regulary and people familiar with the framework are rare because of the many existing and partly easier to use competitors like Reactive, VueJS, Ember, Backbone, Aurelia, Meteor.js.

SPAs require most libraries and interfaces to be loaded and initialized during startup, although only a fraction is used on every rendered page.

Paper UI therefore suffered from feeling unresponsive in certain situations and is hard to extend.

New approach: Architecture and used technology

This new approch uses Web components (v1), static html pages, progressive enhancement and progressive webapps (PWA) technologies like advanced caching.

Web components allow to define own html tags like "oh-binding-doc". Such an "oh-binding-doc" custom component for example would load and display the documentation for a given binding:

<html>
  <head>
    <!-- include the component -->
    <script type="module" src="js/oh-binding-doc.js"></script>
  </head>
  <body>
    <oh-binding-doc binding="mqtt" collapsable collapsed></oh-binding-doc>
  </body>
</html>

Web components (v1) are a standard web technology and work in all evergreen-browsers (Chrome, Firefox, Samsung Internet, Opera, Safari, Edge). No framework is required and at the same time every framework is compatible.
Static Html pages (/index.html, /items.html, /things.html) allow to only load the exact subset of necessary libraries and style sheets per page. Memory consumption is reduced, because of a small DOM. This is especially useful, because the VS-Code editor for example weights 7 MB, and is only loaded on pages that require them.

Page layout and page changing

There is a responsive css grid layout in place, therefore the html layout need to be similar on all pages. Have a look at src/newpage_template.html for a commented html layout that need to be conformed to.

The advantage of SPAs are that the shell of the application stays, so there is no flickering while changing to another subpage. This is realised in this project as a progressive enhancement, by adding the custom component <nav-ajax-page-load></nav-ajax-page-load> to the page. Clicks are intercepted and only parts of the current shown page are replaced.

Static pages

Each html file in src/ is an independant html file.

A build system (Gulp) is in place to copy, but also process html files. For one to implement i18n (translations) and the other reason is to save us from repetitive html like navigation areas. Html partials are used like this:

Store your html fragment into partials/*.html.
Use it in your src/*.html page via <partial src="your-partial-filename.html"></partial>.

Nested partials are supported but should be avoided. Partials do support basic variables. Just use attributes on the markup (like key="value") and use @@key within your partial.

Why no Markdown / other markup language

Most of the time in this application you are not presenting just text (the tuturial section is an exception), but you are presenting interactive content, forms, or need custom html tags (Web components).

The context help texts are written in Markdown though. You find them in assets/contexthelp. They are dynamically fetched when required (and cached as html in the localstorage).

Caching

A service worker cache is in place. While you develop, you should open the DevTools of your browser and tick the checkbox "Disable cache" (in the tab Network on Chrome) or disable the service worker (Tab "Application" -> "Service Worker" -> "Bypass for network" in Chrome.)

All dynamically fetched contents like forum posts, help texts and github data is cached in the users localstorage as prerendered html. The cache has an expire duration of 1 day.

Web component

Webcomponents for

fetching and displaying a context help,
the openHAB REST interface,
the openHAB community forum,
github openHAB addons repository for documentation fetching
navigation components (breadcrumb, prev/next-buttons)

are available in js/bundles/ohcomponents/*.js and js/bundles/uicomponents/*.js and can be used in other projects as well.

The html dom API alone is a bit clumsy for more complex reactive components though. I have used lit-html in some more complex components as renderer. It has a very similar syntax to the vue renderer, but is not "reactive" like vue and therefore adds only 1.7KB (tree shaking not even considered) to the ui-components bundle.

We could think about using lit-elements in the future, which uses lit-html for rendering but also offers one-way and two-way bindings. Or we directly use Vue 3, which thanks to tree-shaking and building up on modern standards only, will also come with a low footprint.

How does interaction with openHAB works

A Model-View-Adapter (MVA) concept is in place and illustrated in the following diagram:

Model-View-Adapter

Image source

There are several components used as View:

The VueJS based ohcomponents/oh-vue-list for reactive list
The uicomponents/ui-dropdown for a dropdown (e.g. selection of Items or Profiles)
The VueJS based ohcomponents/oh-vue for configuration pages

The ohcomponents/oh-vue-list-bind and ohcomponents/oh-vue-bind classes serve as Controllers. They receive all remove and change requests of the Views and also observe the Model and Adapter for any changes.

The listadapter/* classes provide Mixins for the View, but also provide Model Adapters that communicate with the Model (aka Store).

The Model

The app/* bundle finally provides the frontend database, the Model, for this architecture. All requested REST endpoints are cached in a Index DB and kept in sync via SSE (Server Send Events). The storage follows a State-While-Revalidate strategy.

The Index DB access and REST updates happen in a web-worker. Heavy operations like table joining and sorting is outsourced into the web-worker.

Multiple browser tabs stay in sync and only a single SSE connection must be established. (Not true at the moment. Shared web workers do not work in WebKit/Safari.)

This architecture should provide us with low-latency rendering performance.