htmlpurifier/docs/proposal-config.txt

Configuration

Configuration is documented on a per-use case: if a class uses a certain
value from the configuration object, it has to define its name and what the
value is used for.  This means decentralized configuration declarations that
are nevertheless error checking and a centralized configuration object.

Directives are divided into namespaces, indicating the major portion of
functionality they cover (although there may be overlaps.  Please consult
the documentation in ConfigDef for more information on these namespaces.

Since configuration is dependant on context, internal classes require a
configuration object to be passed as a parameter.  (They also require a
Context object).

In relation to HTMLDefinition and CSSDefinition, there could be a special class
of directives that influence the *construction* of the Definition object.
A theoretical call pattern would look like:

1. Client calls Config->getHTMLDefinition()
2. Config calls HTMLDefinition->createNew(this)
3. HTMLDefinition constructs itself with base configuration
4. HTMLDefinition calls Config->get('HTML')
5. Config returns array of directives
6. HTMLDefinition performs operations and changes specified by directives
7. HTMLPurifier returns constructed definition
8. Config caches definition so it doesn't have to be generated again
9. Config returns definition

You could also override Config's copy of the definition with your own
custom copy, which OVERRIDES all directives.  Only the base, vanilla copy
is the Singleton, the object actually interfaced with is a operated-upon
clone of that object.  Also, if an update to the directives would update
the definition, you'd have to force reconstruction.

In practice, the pulling directives from the config object are
solely need-based, and the flex points are littered throughout the
setup() function.  Some sort of refactoring is likely in order. See
ref-xhtml-1.1.txt for more info.