More tweaks to the design.

[roundup.git] / doc / templating.txt

==========================
HTML Templating Mechanisms
==========================

:Version: $Revision: 1.11 $

Current Situation and Issues
============================

Syntax
------

Roundup currently uses an element-based HTML-tag-alike templating syntax::

   <display call="checklist('status')">

The templates were initially parsed using recursive regular expression
parsing, and since no template tag could encapsulate itself, the parser
worked just fine. Then we got the ``<require>`` tag, which could have other
``<require>`` tags inside. This forced us to move towards a more complete
parser, using the standard python sgmllib/htmllib parser. The downside of this
switch is that constructs of the form::

   <tr class="row-<display call="plain('status')">">

don't parse as we'd hope. We can modify the parser to work, but that doesn't
another couple of issues that have arisen:

1. the template syntax is not well-formed, and therefore is a pain to parse
   and doesn't play well with other tools, and
2. user requirements generally have to be anticipated and accounted for in
   templating functions (like ``plain()`` and ``checklist()`` above), and
   we are therefore artificially restrictive.

Arguments for switching templating systems:

*Pros*

  - more flexibility in templating control and content
  - we can be well-formed

*Cons*

  - installed user base (though they'd have to edit their templates with the
    next release anyway)
  - current templating system is pretty trivial, and a more flexible system
    is likely to be more complex


Templates
---------

We should also take this opportunity to open up the flexibility of the
templates through:

1. allowing the instance to define a "page" template, which holds the overall
   page structure, including header and footer



Possible approaches
===================

Zope's PageTemplates
--------------------

Using Zope's PageTemplates seems to be the best approach of the lot.
In my opinion, it's the peak of HTML templating technology at present. With
appropriate infrastructure, the above two examples would read:

  <span tal:replace="item/status/checklist">status checklist</span>

  <tr tal:attributes="class string:row-${item/status/name}">

... which doesn't look that much more complicated... honest...

Other fun can be had when you start playing with stuff like:

  <table>
   <tr tal:repeat="message item/msg/list">
    <td tal:define="from message/from">
     <a href="" tal:attributes="href string:mailto:${from/address}"
                tal:content="from/name">mailto link</a>
    </td>
    <td tal:content="message/title">subject</td>
    <td tal:content="message/created">received date</td>
   </tr>
  </table>

Note: even if we don't switch templating as a whole, this document may be
applied to the ZRoundup frontend.

PageTemplates in a Nutshell
~~~~~~~~~~~~~~~~~~~~~~~~~~~

PageTemplates consist of three technologies:

TAL - Template Attribute Language
  This is the syntax which is woven into the HTML using the ``tal:`` tag
  attributes. A TAL parser pulls out the TAL commands from the attributes
  runs them using some expression engine.

TALES - TAL Expression Syntax
  The expression engine used in this case is TALES, which runs the expressions
  that form the tag attribute values. TALES expressions come in three
  flavours:

  Path Expressions - eg. ``item/status/checklist``
   These are object attribute / item accesses. Roughly speaking, the path
   ``item/status/checklist`` is broken into parts ``item``, ``status``
   and ``checklist``. The ``item`` part is the root of the expression.
   We then look for a ``status`` attribute on ``item``, or failing that, a
   ``status`` item (as in ``item['status']``). If that
   fails, the path expression fails. When we get to the end, the object we're
   left with is evaluated to get a string - methods are called, objects are
   stringified. Path expressions may have an optional ``path:`` prefix, though
   they are the default expression type, so it's not necessary.

  String Expressions - eg. ``string:hello ${user/name}``
   These expressions are simple string interpolations (though they can be just
   plain strings with no interpolation if you want. The expression in the
   ``${ ... }`` is just a path expression as above.

  Python Expressions - eg. ``python: 1+1``
   These expressions give the full power of Python. All the "root level"
   variables are available, so ``python:item.status.checklist()`` would be
   equivalent to ``item/status/checklist``, assuming that ``checklist`` is
   a method.

PageTemplates
  The PageTemplates module glues together TAL and TALES.


Implementation
~~~~~~~~~~~~~~

I'm envisaging an infrastructure layer where each template has the following
"root level" (that is, directly accessible in the TALES namespace) variables
defined:

*user*
  The current user node as an HTMLItem instance

*class*
  The current class of node being displayed as an HTMLClass instance

*item*
  The current node from the database, if we're viewing a specific node, as an
  HTMLItem instance. If it doesn't exist, then we're on a new item page.

(*classname*)
  this is one of two things:

  1. the *item* is also available under its classname, so a *user* node
     would also be available under the name *user*. This is also an HTMLItem
     instance.
  2. if there's no *item* then the current class is available through this
     name, thus "user/name" and "user/name/menu" will still work - the latter
     will pull information from the form if it can.

*form*
  The current CGI form information as a mapping of form argument name to value

*request*
  Includes information about the current request, including:
   - the url
   - the current index information (``filterspec``, ``filter`` args,
     ``properties``, etc) parsed out of the form. 
   - methods for easy filterspec link generation

*instance*
  The current instance

*db*
  The current open database

*config*
  The current instance config

*modules*
  python modules made available (XXX: not sure what's actually in there tho)

Accesses through the *user*::

    class HTMLUser:
        def hasPermission(self, ...)

(note that the other permission check implemented by the security module may
 be implemented easily in a tal:condition, so isn't needed here)

Accesses through a class (either through *class* or *db.<classname>*):

    class HTMLClass:
        def __getattr__(self, attr):
            ''' return an HTMLItem instance '''
        def classhelp(self, ...)
        def list(self, ...)
        def filter(self):
            ''' Return a list of items from this class, filtered and sorted
                by the current requested filterspec/filter/sort/group args
            '''

Accesses through an *item*::

    class HTMLItem:
        def __getattr__(self, attr):
            ''' return an HTMLItem instance '''
        def history(self, ...)
        def remove(self, ...)

Note: the above could cause problems if someone wants to have properties
called "history" or "remove"...

String, Number, Date, Interval HTMLProperty
 a wrapper object which may be stringified for the current plain() behaviour
 and has methods emulating all the current display functions, so
 ``item/name/plain`` would emulate the current ``call="plain()``". Also, 
 ``python:item.name.plain(name=value)`` would work just fine::

    class HTMLProperty:
        def __init__(self, instance, db, ...)
        def __str__(self):
            return self.plain()

    class StringHTMLProperty(HTLProperty):
        def plain(self, ...)
        def field(self, ...)
        def stext(self, ...)
        def multiline(self, ...)
        def email(self, ...)

    class NumberHTMLProperty(HTMLProperty):
        def plain(self, ...)
        def field(self, ...)

    class BooleanHTMLProperty(HTMLProperty):
        def plain(self, ...)
        def field(self, ...)

    class DateHTMLProperty(HTMLProperty):
        def plain(self, ...)
        def field(self, ...)
        def reldate(self, ...)

    class IntervalHTMLProperty(HTMLProperty):
        def plain(self, ...)
        def field(self, ...)
        def pretty(self, ...)

Link HTMLProperty
 the wrapper object would include the above as well as being able to access
 the class information. Stringifying the object itself would result in the
 value from the item being displayed. Accessing attributes of this object
 would result in the appropriate entry from the class being queried for the
 property accessed (so item/assignedto/name would look up the user entry
 identified by the assignedto property on item, and then the name property of
 that user)::
    
    class LinkHTMLProperty(HTMLProperty):
        ''' Be a HTMLItem too '''
        def __getattr__(self, attr):
            ''' return a new HTMLProperty '''
        def download(self, ...)
        def checklist(self, ...)

Multilink HTMLProperty
 the wrapper would also be iterable, returning a wrapper object like the Link
 case for each entry in the multilink::

    class MultilinkHTMLProperty(HTMLProperty):
        def __len__(self):
            ''' length of the multilink '''
        def __getitem(self, num):
            ''' return a new HTMLItem '''
        def checklist(self, ...)
        def list(self, ...)
 
*request*
 the request object will handle::

    class Request:
        def __init__(self, ...)
        def filterspec(self, ...)

Template files
~~~~~~~~~~~~~~

Each instance will have the opportunity to supply the following templates:

page
 This is the overall page look template, and includes at some point a TAL
 command that includes the variable "content". This variable causes the actual
 page content to be generated.

[classname].[template type]
 Templates that have this form are applied to item data. There are three forms
 of special template types:

 [classname].index
  This template is used when the URL specifies only the class, and not a node
  designator.  It displays a list of [classname] items from the database, and
  a "filter refinement" form.
  Would perform a TAL ``repeat`` command using the list supplied by
  ``class/filter``. This deviates from the current situation in that currently
  the index template specifies a single row, and the filter part is
  automatically generated.

 [classname].item
  This template is used when the URL specifies an item designator. It's the
  default template used (when no template is explicitly given). It displays
  a single item from the database using the *classname* variable (that
  is, the variable of the same name as the class being displayed. If 

 These two special template types may be overridden by the :template CGI
 variable.

Note that the "newitem" template doesn't exist any more because the item
templates may determine whether the page has an existing item to render. The
new item page would be accessed by "/tracker/url/issue?:template=item".
The old "filter" template has been subsumed by the index template.


Integrating Code
~~~~~~~~~~~~~~~~

We will install PageTemplates, TAL and ZTUtils in site-packages. If there is a
local Zope installation, it will use its own PageTemplates code (Zope modifies
the module search path to give precedence to its own module library).

We will then install the trivial MultiMapping and ComputedAttribute modules in
the Roundup package, and have some import trickery that determines whether
they are required, and if so they will be imported as if they were at the
"top level" of the module namespace.

New CGI client structure
~~~~~~~~~~~~~~~~~~~~~~~~

Handling of a request in the CGI client will take three phases:

1. Determine user, pre-set "content" to authorisation page if necessary
2. Render main page, with callback to "content"
3. Render content - if not pre-set, then determine which content to render


Use Cases
~~~~~~~~~

Meta/parent bug
  Can be done with addition to the schema and then the actual parent heirarchy
  may be displayed with a new template page ":dependencies" or something.

Submission wizard
  Can be done using new templates ":page1", ":page2", etc and some additional
  actions on the CGI Client class in the instance.