plone.app.discussion/plone/app/discussion/PRINCIPLES.txt

78 lines
3.1 KiB
Plaintext

=============================================
plone.app.discussion architectural principles
=============================================
This document outlines architectural principles used in the design of
plone.app.discussion.
Discussion items have a portal_type
This makes it easier to search for them and manage them using existing
CMF and Plone UI constructs.
Discussion items are cataloged
It should be possible to search for discussion items like any other
type of content.
Discussion items are subject to workflow and permission
Moderation, anonymous commenting, and auto approve/reject should be
handled using workflow states, automatic and manual transitions, and
permissions.
Discussion items are light weight objects
All discussion item objects should be as light weight as possible.
Ideally, a discussion item should be as lightweight as a catalog brain.
This may mean that we forego convenience base classes and re-implement
certain interfaces. Comments should not provide the full set of dublin
core metadata, though custom indexers can be used to provide values for
standard catalog indexes.
Optimise for retrival speed
HTML filtering and other processing should happen on save, not on render,
to make rendering quick.
Settings are stored using plone.registry
Any global setting should be stored in plone.registry records
Forms are constructed using extensible z3c.form forms
This allows plugins (such as spam protection algorithms) to provide
additional validation
Discussion items are stored in a BTree container
This allows faster lookup and manipulation
Discussion items are accessed using a dict-like interface
This makes iteration and manipulation more natural. Even if comments are
not stored threaded, the dict interface should act as if they are, i.e.
calling items() on a comment should return the replies to that comment
(in order).
Discussion items are retrieved in reverse creation date order
Discussion items do not need to support explicit ordering. They should
always be retrieved in reverse creation date order (most recent for).
They can be stored with keys so that this is always true.
Discussion items do not need readable ids
Ids can be based on the creation date.
Discussion items send events
The usual zope.lifecycleevent and zope.app.container events should be
fired when discussion items are added, removed, or modified.
Outstanding questions
---------------------
* Should comments use the 'talkback' URL structure currently used in CMF,
or a ++comments++ namespace?
* Should discussion items be stored in one container, maintaining parent
pointers as references or attributes, or in a nested structure?
- it may be possible to use the navtree algorithm to turn a search of
comments into a nested structure
* How can we ensure that discussion items are not removed from the catalog
when the user initiates a "clear and rebuild"?
* Can we find a way to avoid having to explicitly migrate all existing
comments?