Render library specification
| Renderer consistency | |
|---|---|
| Project state | Specification |
| Tracker issue | - |
| Discussion | - |
| Assignee | Damyon, Sam |
Project goals
- Make it easier to write UI pages in Moodle
- Make it easier for themers to customise UI pages in Moodle
- Make it easier to switch CSS frameworks
- Make the appearance of Moodle more consistent (through reuse of components)
Note: This page is a work-in-progress. Feedback and suggested improvements are welcome. Please join the discussion on moodle.org or use the page comments.
Understanding renderers in Moodle
Renderers have been around since Moodle 2.0 and are now the requirement when writing any code that produces output. Renderers come in many shapes and forms depending upon the developers take on how output should be produced. They are typically given a range of arguments all of which fall into the space created by three concepts:
- Simple arguments - scalars, booleans the like; either being content or an option.
- Multi-purpose objects - objects that serve both to provide data for output and as a functional, interactional object. If you are familiar with the MVC architecture pattern then this is the model. It may relate directory to what is being output (it may be a forum post object when printing a forum post) but more likely its relation is to that of the component or a subsystem there of (giving a forum object to method producing a output for a forum discussion).
- Single-purpose object - here we have an object that is built to solely supplement the output process. Its properties are usables and its methods relate purely to the state of the object in relation to its display. It is to say just a container of useful information and devoid of logic.
Because there is no mandated organisation, nor has there ever being there is immense variation in the what constitutes content to a renderer. There is no way to predict what is provided, not that there should necessarily be of course.
One of the upsides to our current render system is that you are not limited to having a single renderer for a component, you can make use of language capabilities and the output framework and create layers of renderers to cut back code duplication and to aid in preparing a base for consistent structuring and output.
Into the equation we add targets and subtypes. Targets all us to create both standard renderers and renderers specific to the needs of the output method. Standard being HTML, without output methods including CLI and AJAX. Subtypes allow for compartmentalisation to occur in component architecture allowing for greater granularity and organisation when writing a render and design its use within the component.
People have never being constrained as to how they implement a renderer, we've being happy enough to simply guide them towards using renderers because of the benefits they provide and the benefits are numerous but there are really two primary benefits:
1. A change in thinking: It required developers to think about their code organisation. They were introduced in a time before Object Orientation was used in design, a time when logic and output were tightly intermingled in most code.
2. An extensible approach to output: The render system allowed for the overriding renderers to occur. Renderers were supplied in a chained stack. This benefiting both theme designers and developers. Theme designers got the final say - when they overrode a renderer they determined what would occur. When the renderer was used if they hadn't overridden the method being called it would fall back to the next renderer in the stack. It also benefited developers, when you wrote a renderer for a component you were working on you a) knew that your output could be customised by themers and b) you could rely on the stack to re-use existing render methods published on the core_renderer. Another words if you need to create a notification you could ask your renderer to produce it and as you'd not defined in your renderer it'd fall to the responsibility of the next renderer in the chain. This allowed you to produce "widgets" within your output that were more consistent across components. Of course subjectively you could override those core methods to alter how a "widget" was rendered throughout your entire component. It is highly flexible.
Understanding themes
Themers are without a doubt the most widely customised plugin provided by Moodle. But there are some gotchas to how we've created themes in Moodle, especially as we currently stand. For a start there is a lose relation between the renderers and the output in that the HTML is created by the render in relation to the component whereas the CSS is in many cases added to the theme in order to style the HTML without relation to the component. This leads to the renderer being dependent on the theme to do the right thing. This is true for core by the way. Plugins can have styles.css files and as such can lay groundwork CSS without the need to inject it into a theme. For core we work around this by having a "base" theme(s). The base theme is responsible primarily for laying the ground work for core and for applying some minimal default styling. However the issue is still there, the relation betweem HTML and CSS in Moodle is largely indirect.
Next up lets consider frontend frameworks. Moodle has a chosen frontend framework at present, however rather than being located in core it is located within a theme. This has implications on its use throughout Moodle in that it cannot directly be used within renderers as other themes may not be running the framework, or may be running a different framework. While there is definitely some cross over there is much of the chosen framework that is simply out of reach short of overriding renderers and going to town to create a "secondary" display of Moodle.
So whats the problem
There are several. Lets be blunt about this. This specification is not a single task. It is about improving the state of output in Moodle and there are going to be several steps in achieving this. On a positive note these steps can in many cases be broken down into granular tasks that can be worked on asynchronously. First lets understand some of the major problems:
- We've not established best practices, nor even documentation on what constitutes a good renderer, or considerations that should be paid mind in planning and designing of code.
- We've not got an established style guide/pattern library/element library, nor anything that could be referenced when someone starts tossing around those buzz words.
- Backend congestion. Through flexibility and a lack of the above documentation and gudies we've got substantial fragmentation of our design. Spread across all layers of code and deep into the crevasses within our code. Design and output is duplicated in numerous places, and consequently ends up further fragmenting as further work forgets to update all places (or simply cannot update all places).
- Developers are encouraged to push functionality boundaries in order to take it places observed in other projects or to challenge oneself. While consistency is often talked about it is rarely observed when designing output. We end up with numerous "similiar" looking widgets all functioning slightly differently and consequently there is a feeling that this is the way to achieve a new fresh look. One widget at a time.
Identified issues with renderers in 2.7
- Poorly implemented renderers are full of logic/db queries/access checks that would need to be duplicated by a themer wanting to change the HTML (and forces the themer to be good at PHP and know all the Moodle apis)
- Lack of consistency
- Lack of documentation
- No abstraction from CSS framework
- Requirement to style each page individually causes the CSS to balloon
- No common technique to template UIs generated by javascript
Tasks
- Rewrite the docs for renderers
- Read and rate each of the existing docs pages (see related links below)
- Write new complete docs for everyone (at Output API)
- Cross link old docs to the new ones
- Add an admin tool that can show a list of renderable objects for each plugin/and core - filled with test data. The goal of this tool is to
- validate that the renderer/renderable follows best practice (no DB queries, complex types, complex logic)
- provide a page for themers to check that their new theme correctly displays all the renderables
- provide a page for developers to see all the reusable renderables (in a well defined structure) they can use when building their own pages
- Define a comprehensive list of low level widgets we can add to core as renderables. This is a library of components that should be used by other renderers (by compositing them). E.g. list, table, warning. These renderables should be structured into layers, starting from renderables that cannot be broken into smaller parts, up to widgets made of smaller renderables, up to layouts (and possibly up to entire pages). Sources for this list should come from:
- Existing CSS Frameworks (bootstrap, pureio) (but must be framework agnostic)
- Existing renderables in Moodle that we see as “perfect” already (maybe there are some)
- Existing patterns in Moodle that we do over and over but don’t have a renderable for yet
- Build this list of widgets as core renderables with a generator so they are listed in the admin tool
- Use compositions of these new core renderables for all new code (policy decision)
- Slowly convert the old renderers to compositions of the new ones with minimal logic etc (no eta)
Notes on renderers / renderables
Proposed improvements to the docs on renderers / renderables. First - the docs need an overhaul to more clearly explain how renderers / renderables work for core devs, plugin devs and themers. There needs to be clearer guidelines for the things that can be done in a renderable, and in a renderer.
A "good renderer" is:
- Receives some “data” through a renderable
- Produces some “output”
- Where “output” is some HTML and any javascript init calls
- Can use basic PHP functions/loops, like for(), foreach(), count()
A "good renderer" is not:
- nasty php logic (or even non-nasty php logic)
- access checks
- non-trivial function calls
- using any Moodle functions not related to output
- calling the database
- unshaven
- a bad renderer
A “renderable” is:
- All the data required to generate the output.
- Properties should be simple data types only, or renderables, or array of those
