This is needed in some situations when we only want to update a small single component, sometimes even using a different controller to the one implied in the URL. An example here is reloading dynamically the subsite dropdown without reloading the entire page, updating a filter sidebar or suchlike.
26 KiB
CMS architecture
Introduction
A lot can be achieved in SilverStripe by adding properties and form fields
to your own page types (via [api:SiteTree->getCMSFields()]
), as well as creating
your own data management interfaces through [api:ModelAdmin]
. But sometimes
you'll want to go deeper and tailor the underlying interface to your needs as well.
For example, to build a personalized CMS dashboard, or content "slots" where authors
can drag their content into. At its core, SilverStripe is a web application
built on open standards and common libraries, so lots of the techniques should
feel familiar to you. This is just a quick run down to get you started
with some special conventions.
For a more practical-oriented approach to CMS customizations, refer to the Howto: Extend the CMS Interface which builds
Markup and Style Conventions
While SilverStripe is intended to work with JavaScript only, we're following the principles of "Progressive Enhancement" where feasible, relying on a comparatively light layer of JavaScript to enhance forms and markup generated on the server. This allows seamless customization of aspects like form fields. We're explaining this philosophy in more detail on our blog).
All CSS in the CMS UI is written in the SCSS language extensions
and the Compass framework, which helps
us maintain expressive and concise style declarations. The files are located in framework/admin/scss
(and if you have the cms
module installed, in cms/scss
), and are compiled to a css
folder on the
same directory path. Changes to the SCSS files can be automatically converted by installing
the "compass" module for SilverStripe,
although installing the compass framework directly works as well.
Each file describes its purpose at the top of the declarations. Note that you can write
plain CSS without SCSS for your custom CMS interfaces as well, we just mandate SCSS for core usage.
As there's a whole lot of CSS driving the CMS, we have certain best practives around writing it:
- Use dashed lowercase naming for both
id
andclass
attributes (my-class-name
), instead of camel case (myClassName
) - Use the
id
attribute sparingly. Remember that it "closes off" the structure to code reuse, as HTML elements require uniqueid
attributes. Code reuse can happen both in CSS and JavaScript behaviour. - Separate presentation from structure in class names, e.g.
left-menu
is encoding the component position (which might change later on). A more structural name could becms-menu
(orcms-tools-menu
for a more specific version) - Class naming: Use the
cms-
class prefix for major components in the cms interface, and thess-ui-
prefix for extensions to jQuery UI. Don't use theui-
class prefix, its reserved for jQuery UI built-in styles. - Use jQuery UI's built-in styles where possible, e.g.
ui-widget
for a generic container, orui-state-highlight
to highlight a specific component. See the jQuery UI Theming API for a full list.
See our system requirements for a list of supported browsers.
Templates and Controllers
The CMS backend is handled through the [api:LeftAndMain]
controller class,
which contains base functionality like displaying and saving a record.
This is extended through various subclasses, e.g. to add a group hierarchy ([api:SecurityAdmin]
),
a search interface ([api:ModelAdmin]
) or an "Add Page" form ([api:CMSPageAddController]
).
The controller structure is too complex to document here, a good starting point
for following the execution path in code are [api:LeftAndMain->getRecord()]
and [api:LeftAndMain->getEditForm()]
.
If you have the cms
module installed, have a look at [api:CMSMain->getEditForm()]
for a good
example on how to extend the base functionality (e.g. by adding page versioning hints to the form).
CMS templates are inherited based on their controllers, similar to subclasses of
the common Page
object (a new PHP class MyPage
will look for a MyPage.ss
template).
We can use this to create a different base template with LeftAndMain.ss
(which corresponds to the LeftAndMain
PHP controller class).
In case you want to retain the main CMS structure (which is recommended),
just create your own "Content" template (e.g. MyCMSController_Content.ss
),
which is in charge of rendering the main content area apart from the CMS menu.
Depending on the complexity of your layout, you'll also need to overload the
"EditForm" template (e.g. MyCMSController_EditForm.ss
), e.g. to implement
a tabbed form which only scrolls the main tab areas, while keeping the buttons at the bottom of the frame.
This requires manual assignment of the template to your form instance, see [api:CMSMain->getEditForm()]
for details.
Often its useful to have a "tools" panel in between the menu and your content,
usually occupied by a search form or navigational helper.
In this case, you can either overload the full base template as described above.
To avoid duplicating all this template code, you can also use the special [api:LeftAndMain->Tools()]
and
[api:LeftAndMain->EditFormTools()]
methods available in LeftAndMain
.
These placeholders are populated by auto-detected templates,
with the naming convention of "_Tools.ss" and "_EditFormTools.ss".
So to add or "subclass" a tools panel, simply create this file and it's automatically picked up.
Layout and Panels
The various panels and UI components within them are loosely coupled to the layout engine through the data-layout-type
attribute. The layout is triggered on the top element and cascades into children, with a redraw
method defined on
each panel and UI component that needs to update itself as a result of layouting.
Refer to Layout reference for further information.
Forms
SilverStripe constructs forms and its fields within PHP,
mainly through the [getCMSFields()](api:DataObject->getCMSFields())
method.
This in turn means that the CMS loads these forms as HTML via Ajax calls,
e.g. after saving a record (which requires a form refresh), or switching the section in the CMS.
Depending on where in the DOM hierarchy you want to use a form,
custom templates and additional CSS classes might be required for correct operation.
For example, the "EditForm" has specific view and logic JavaScript behaviour
which can be enabled via adding the "cms-edit-form" class.
In order to set the correct layout classes, we also need a custom template.
To obey the inheritance chain, we use $this->getTemplatesWithSuffix('_EditForm')
for
selecting the most specific template (so MyAdmin_EditForm.ss
, if it exists).
The form should be of type CMSForm
rather than Form
, since it allows the use
of a PjaxResponseNegotiator
to handle its display.
Basic example form in a CMS controller subclass:
:::php
class MyAdmin extends LeftAndMain {
function getEditForm() {
return CMSForm::create(
$this,
'EditForm',
new FieldSet(
TabSet::create(
'Root',
Tab::create('Main',
TextField::create('MyText')
)
)->setTemplate('CMSTabset')
),
new FieldSet(
FormAction::create('doSubmit')
)
)
// JS and CSS use this identifier
->setHTMLID('Form_EditForm')
// Render correct responses on validation errors
->setResponseNegotiator($this->getResponseNegotiator());
// Required for correct CMS layout
->addExtraClass('cms-edit-form')
->setTemplate($this->getTemplatesWithSuffix('_EditForm'));
}
}
Note: Usually you don't need to worry about these settings,
and will simply call parent::getEditForm()
to modify an existing,
correctly configured form.
JavaScript through jQuery.entwine
jQuery.entwine is a thirdparty library
which allows us to attach behaviour to DOM elements in a flexible and structured mannger.
It replaces the behaviour.js
library used in previous versions of the CMS interface.
See Topics: JavaScript for more information on how to use it.
In the CMS interface, all entwine rules should be placed in the "ss" entwine namespace.
If you want to call methods defined within these rules outside of entwine logic,
you have to use this namespace, e.g. $('.cms-menu').entwine('ss').collapse()
.
Note that only functionality that is custom to the CMS application needs to be built in jQuery.entwine, we're trying to reuse library code wherever possible. The most prominent example of this is the usage of jQuery UI for dialogs and buttons.
The CMS includes the jQuery.entwine inspector. Press Ctrl+` ("backtick") to bring down the inspector. You can then click on any element in the CMS to see which entwine methods are bound to any particular element.
JavaScript and CSS dependencies via Requirements and Ajax
The JavaScript logic powering the CMS is divided into many files,
which typically are included via the [api:Requirements]
class, by adding
them to [api:LeftAndMain->init()]
and its subclassed methods.
This class also takes care of minification and combination of the files,
which is crucial for the CMS performance (see [api:Requirements::combine_files()]
).
Due to the procedural and selector-driven style of UI programming in jQuery.entwine,
it can be difficult to find the piece of code responsible for a certain behaviour.
Therefore it is important to adhere to file naming conventions.
E.g. a feature only applicable to ModelAdmin
should be placed in
framework/admin/javascript/ModelAdmin.js
, while something modifying all forms (including ModelAdmin forms)
would be better suited in framework/admin/javascript/LeftAndMain.EditForm.js
.
Selectors used in these files should mirrow the "scope" set by its filename,
so don't place a rule applying to all form buttons inside ModelAdmin.js
.
The CMS relies heavily on Ajax-loading of interfaces, so each interface and the JavaScript driving it have to assume its underlying DOM structure is appended via an Ajax callback rather than being available when the browser window first loads. jQuery.entwine is effectively an advanced version of jQuery.live and jQuery.delegate, so takes care of dynamic event binding.
Most interfaces will require their own JavaScript and CSS files, so the Ajax loading has
to ensure they're loaded unless already present. A custom-built library called
jQuery.ondemand
(located in framework/thirdparty
) takes care of this transparently -
so as a developer just declare your dependencies through the [api:Requirements]
API.
Ajax Loading and Browser History
SilverStripe uses the HTML5 browser history to modify the URL without a complete window refresh, and load its UI via Ajax by hooking into browser navigation events (through the history.js wrapper library). This technique has an impact on how any Ajax load needs to happen: In order to support browser history (and change the URL state), a CMS developer needs to fire a navigation event rather than invoking the Ajax call directly.
The main point of contact here is $('.cms-container').loadPanel(<url>, <title>, <data>)
in LeftAndMain.js
. The data
object can contain additional state which is required
in case the same navigation event is fired again (e.g. when the user pressed the back button).
No callbacks are allowed in this style of Ajax loading, as all state needs
to be "repeatable". Any logic required to be exected after the Ajax call
should be placed in jQuery.entinwe onmatch()
rules which apply to the newly created DOM structures.
See $('.cms-container').handleStateChange()
in LeftAndMain.js
for details.
Alternatively, form-related Ajax calls can be invoked through their own wrappers,
which don't cause history events and hence allow callbacks: $('.cms-container').submitForm()
.
PJAX: Partial template replacement through Ajax
Many user interactions can change more than one area in the CMS. For example, editing a page title in the CMS form changes it in the page tree as well as the breadcrumbs. In order to avoid unnecessary processing, we often want to update these sections independently from their neighbouring content.
In order for this to work, the CMS templates declare certain sections as "PJAX fragments"
through a data-pjax-fragment
attribute. These names correlate to specific
rendering logic in the PHP controllers, through the [api:PjaxResponseNegotiator]
class.
Through a custom X-Pjax
HTTP header, the client can declare which view he's expecting,
through identifiers like CurrentForm
or Content
(see [api:LeftAndMain->getResponseNegotiator()]
).
These identifiers are passed to loadPanel()
via the pjax
data option.
The HTTP response is a JSON object literal, with template replacements keyed by their Pjax fragment.
Through PHP callbacks, we ensure that only the required template parts are actually executed and rendered.
When the same URL is loaded without Ajax (and hence without X-Pjax
headers),
it should behave like a normal full page template, but using the same controller logic.
Example: Create a bare-bones CMS subclass which shows breadcrumbs (a built-in method), as well as info on the current record. A single link updates both sections independently in a single Ajax request.
:::php
// mysite/code/MyAdmin.php
class MyAdmin extends LeftAndMain {
private static $url_segment = 'myadmin';
public function getResponseNegotiator() {
$negotiator = parent::getResponseNegotiator();
$controller = $this;
// Register a new callback
$negotiator->setCallback('MyRecordInfo', function() use(&$controller) {
return $controller->MyRecordInfo();
});
return $negotiator;
}
public function MyRecordInfo() {
return $this->renderWith('MyRecordInfo');
}
}
:::js
// MyAdmin.ss
<% include CMSBreadcrumbs %>
<div>Static content (not affected by update)</div>
<% include MyRecordInfo %>
<a href="admin/myadmin" class="cms-panel-link" data-pjax-target="MyRecordInfo,Breadcrumbs">
Update record info
</a>
:::ss
// MyRecordInfo.ss
<div data-pjax-fragment="MyRecordInfo">
Current Record: $currentPage.Title
</div>
A click on the link will cause the following (abbreviated) ajax HTTP request:
GET /admin/myadmin HTTP/1.1
X-Pjax:MyRecordInfo,Breadcrumbs
X-Requested-With:XMLHttpRequest
... and result in the following response:
{"MyRecordInfo": "<div...", "CMSBreadcrumbs": "<div..."}
Keep in mind that the returned view isn't always decided upon when the Ajax request
is fired, so the server might decide to change it based on its own logic,
sending back different X-Pjax
headers and content.
On the client, you can set your preference through the data-pjax-target
attributes
on links or through the X-Pjax
header. For firing off an Ajax request that is
tracked in the browser history, use the pjax
attribute on the state data.
$('.cms-container').loadPanel('admin/pages', null, {pjax: 'Content'});
Loading lightweight PJAX fragments
Normal navigation between URLs in the admin section of the Framework occurs through loadPanel
and submitForm
.
These calls make sure the HTML5 history is updated correctly and back and forward buttons work. They also take
care of some automation, for example restoring currently selected tabs.
However there are situations when you would like to only update a small area in the CMS, and when this operation should not trigger a browser's history pushState. A good example here is reloading a dropdown that relies on backend session information that could have been updated as part of action elsewhere, updates to sidebar status, or other areas unrelated to the main flow.
In this case you can use the loadFragment
call supplied by LeftAndMain.js
. You can trigger as many of these in
parallel as you want. This will not disturb the main navigation.
$('.cms-container').loadFragment('admin/foobar/', 'Fragment1');
$('.cms-container').loadFragment('admin/foobar/', 'Fragment2');
$('.cms-container').loadFragment('admin/foobar/', 'Fragment3');
The ongoing requests are tracked by the PJAX fragment name (Fragment1, 2, and 3 above) - resubmission will result in the prior request for this fragment to be aborted. Other parallel requests will continue undisturbed.
You can also load multiple fragments in one request, as long as they are to the same controller (i.e. URL):
$('.cms-container').loadFragment('admin/foobar/', 'Fragment2,Fragment3');
This counts as a separate request type from the perspective of the request tracking, so will not abort the singular
Fragment2
nor Fragment3
.
Upon the receipt of the response, the fragment will be injected into DOM where a matching data-pjax-fragment
attribute
has been found on an element (this element will get completely replaced). Afterwards a afterloadfragment
event
will be triggered. In case of a request error a loadfragmenterror
will be raised and DOM will not be touched.
You can hook up a response handler that obtains all the details of the XHR request like this:
'from .cms-container': {
onafterloadfragment: function(e, data) {
// Say 'success'!
alert(data.status);
}
}
Ajax Redirects
Sometimes, a server response represents a new URL state, e.g. when submitting an "add record" form,
the resulting view will be the edit form of the new record. On non-ajax submissions, that's easily
handled through a HTTP redirection. On ajax submissions, browsers handle these redirects
transparently, so the CMS JavaScript doesn't know about them (or the new URL).
To work around this, we're using a custom X-ControllerURL
HTTP response header
which can declare a new URL. If this header is set, the CMS JavaScript will
push the URL to its history stack, causing the logic to fetch it in a subsequent ajax request.
Note: To avoid double processing, the first response body is usually empty.
State through HTTP response metadata
By loading mostly HTML responses, we don't have an easy way to communicate information which can't be directly contained in the produced HTML. For example, the currently used controller class might've changed due to a "redirect", which affects the currently active menu entry. We're using HTTP response headers to contain this data without affecting the response body.
:::php
class MyController extends LeftAndMain {
class myaction() {
// ...
$this->response->addHeader('X-Controller', 'MyOtherController');
return $html;
}
}
Built-in headers are:
X-Title
: Set window title (requires URL encoding)X-Controller
: PHP class name matching a menu entry, which is marked activeX-ControllerURL
: Alternative URL to record in the HTML5 browser historyX-Status
: Extended status information, used for an information popover.X-Reload
: Force a full page reload based onX-ControllerURL
Special Links
Some links should do more than load a new page in the browser window. To avoid repetition, we've written some helpers for various use cases:
- Load into a PJAX panel:
<a href="..." class="cms-panel-link" data-pjax-target="Content">
- Load URL as an iframe into a popup/dialog:
<a href="..." class="ss-ui-dialog-link">
Buttons
SilverStripe automatically applies a jQuery UI button style
to all elements with the class .ss-ui-button
. We've extended the jQuery UI widget a bit
to support defining icons via HTML5 data attributes (see ssui.core.js
).
These icon identifiers relate to icon files in framework/admin/images/btn-icons
,
and are sprited into a single file through SCSS and the Compass framework
(see tutorial).
Compass also creates the correct CSS classes to show those sprites via background images
(see framework/admin/scss/_sprites.scss
).
Input: <a href="..." class="ss-ui-button" data-icon="add" />Button text</a>
Output: <a href="..." data-icon="add" class="ss-ui-button ss-ui-action-constructive ui-button ui-widget ui-state-default ui-corner-all ui-button-text-icon-primary" role="button"><span class="ui-button-icon-primary ui-icon btn-icon-add"></span><span class="ui-button-text">Button text</span></a>
Note that you can create buttons from pretty much any element, although when using an input of type button, submit or reset, support is limited to plain text labels with no icons.
Menu
The navigation menu in the CMS is created through the [api:CMSMenu]
API,
which auto-detects all subclasses of LeftAndMain
. This means that your custom
ModelAdmin
subclasses will already appear in there without any explicit definition.
To modify existing menu entries or create new ones, see [api:CMSMenu::add_menu_item()]
and [api:CMSMenu::remove_menu_item()]
.
New content panels are typically loaded via Ajax, which might change the current menu context. For example, a link to edit a file might be clicked within a page edit form, which should change the currently active menu entry from "Page" to "Files & Images". To communicate this state change, a controller response has the option to pass along a special HTTP response header, which is picked up by the menu:
:::php
public function mycontrollermethod() {
// .. logic here
$this->getResponse()->addHeader('X-Controller', 'AssetAdmin');
return 'my response';
}
This is usually handled by the existing [api:LeftAndMain]
logic,
so you don't need to worry about it. The same concept applies for
'X-Title' (change the window title) and 'X-ControllerURL' (change the URL recorded in browser history).
Note: You can see any additional HTTP headers through the web developer tools in your browser of choice.
Tree
The CMS tree for viewing hierarchical structures (mostly pages) is powered
by the jstree library. It is configured through
framework/admin/javascript/LeftAndMain.Tree.js
, as well as some
HTML5 metadata generated on its container (see the data-hints
attribute).
For more information, see the Howto: Customize the CMS tree.
Note that a similar tree logic is also used for the
form fields to select one or more entries from those hierarchies
([api:TreeDropdownField]
and [api:TreeMultiselectField]
).
Tabs
We're using jQuery UI tabs, but in a customized fashion.
HTML with tabs can be created either directly through HTML templates in the CMS,
or indirectly through a [api:TabSet]
form field. Since tabsets are useable
outside of the CMS as well, the baseline application of tabs happens via
a small wrapper around jQuery.tabs()
stored in TabSet.js
.
In the CMS however, tabs need to do more: They memorize their active tab in the user's browser, and lazy load content via ajax once they're activated.
They also need to work across different "layout containers" (see above),
meaning a tab navigation might be in a layout header, while the tab
content is occupied by the main content area. jQuery assumes a common
parent in the DOM for both the tab navigation and its target DOM elements.
In order to achieve this level of flexibility, most tabsets in the CMS
use a custom template which leaves rendering the tab navigation to
a separate template: CMSMain.ss
. See the "Forms" section above
for an example form.
Here's how you would apply this template to your own tabsets used in the CMS. Note that you usually only need to apply it to the outermost tabset, since all others should render with their tab navigation inline.
Form template with custom tab navigation (trimmed down):
:::ss
<form $FormAttributes data-layout-type="border">
<div class="cms-content-header north">
<% if Fields.hasTabset %>
<% with Fields.fieldByName('Root') %>
<div class="cms-content-header-tabs">
<ul>
<% loop Tabs %>
<li><a href="#$id">$Title</a></li>
<% end_loop %>
</ul>
</div>
<% end_with %>
<% end_if %>
</div>
<div class="cms-content-fields center">
<fieldset>
<% loop Fields %>$FieldHolder<% end_loop %>
</fieldset>
</div>
</form>
Tabset template without tab navigation (e.g. CMSTabset.ss
)
:::ss
<div $AttributesHTML>
<% loop Tabs %>
<% if Tabs %>
$FieldHolder
<% else %>
<div $AttributesHTML>
<% loop Fields %>
$FieldHolder
<% end_loop %>
</div>
<% end_if %>
<% end_loop %>
</div>
Lazy loading works based on the href
attribute of the tab navigation.
The base behaviour is applied through adding a class .cms-tabset
to a container.
Assuming that each tab has its own URL which is tracked in the HTML5 history,
the current tab display also has to work when loaded directly without Ajax.
This is achieved by template conditionals (see "MyActiveCondition").
The .cms-panel-link
class will automatically trigger the ajax loading,
and load the HTML content into the main view. Example:
:::ss
<div id="my-tab-id" class="cms-tabset" data-ignore-tab-state="true">
<ul>
<li class="<% if MyActiveCondition %> ui-tabs-active<% end_if %>">
<a href="admin/mytabs/tab1" class="cms-panel-link">
Tab1
</a>
</li>
<li class="<% if MyActiveCondition %> ui-tabs-active<% end_if %>">
<a href="admin/mytabs/tab2" class="cms-panel-link">
Tab2
</a>
</li>
</ul>
</div>
The URL endpoints admin/mytabs/tab1
and admin/mytabs/tab2
should return HTML fragments suitable for inserting into the content area,
through the PjaxResponseNegotiator
class (see above).