When a form element doesn't specify a #weight, it is assumed internally as #weight 0. However, to ensure that our form elements display visually *as they were defined in the array* we, in form_builder, count the number of elements, divide by 1000, and set that as the weight:

# Assign a decimal placeholder weight to preserve original array order
if (!isset($form[$key]['#weight'])) {
  $form[$key]['#weight'] = $count/1000;
}

The above code will set the #weights of elements that have not defined a weight to something like 0 (first element in array definition), 0.001, 0.002, and so on. However, anytime a form element *explicitly* defines a #weight of 0, that #weight is kept at exactly 0, which would cause that form element to appear BEFORE the elements that didn't have a #weight defined (and thus received a #weight such as 0.002).

Consider the following pseudo example:

$form['game_title'] = array(
    '#type' => 'textfield',
    ...
    );
$form['game_description'] = array(
    '#type' => 'textarea',
    ...
    );
$form['game_format'] = filter_form(variable_get('game_format', NULL));
return $form;

Here, we're not definiing weights on our two textfields. We then add an filter_form. The second parameter of the filter_form is $weight, which defaults to 0. After this $form hits form_builder, we have weights 0 (game_title), 0.001 (game_description), and 0 (filter_form) respectively. This is then sorted by weight, which causes filter_form (the third element in the array) to appear BEFORE game_description (0 is lighter than 0.001).

The short lesson is: explicitly defining #weight 0 for a form element is probably a bad idea. This patch changes the default #weight of filter_form to NULL, instead of 0, and also removes any other explicit setting of #weight to 0 in core.

- Patch #39778 by chx: obliterate nodeapi op form in favor of the forms API's way of doing things.  Tested with help from webchick.

  TODO:
  + The contact.module was broken; a new patch for contact.module is needed.
  + Documentation is needed.
  + The most important modules need to be updated ASAP.

All node revisions were stored in a serialized field in the node table and retrieved for _each_ page view although they are rarely needed. We created a separate revisions table which would be in principle identical to the node table, only that it could have several old copies of the same node.  This also allows us to revision-related information, and to provide log entries to non-book pages when a new revision is being created.

TODO:

1. Provide upgrade instructions for node module maintainers!
2. Upgrade modules that implement node types.
3. Provide an upgarde path for revisions.  Dependency on the upgrade system.

  part of the node system!  If you have a module that implements node
  types, you'll have to udpate its CVS HEAD version.

  We replaced _node_name() and _node_types() by _node().  The new _node()
  hook let's you define one or more node types, including their names.
  The implementation of the _node() hook needs to:

   return array($type1 => array('name' => $name1, 'base' => $base1),
                $type2 => array('name' => $name2, 'base' => $base2));

  where $type is the node type, $name is the human readable name of the type
  and $base is used instead of <hook> for <hook>_load, <hook>_view, etc.

  For example, the story module's node hook looks like this:

    function story_node() {
      return array('story' => array('name' => t('story'), 'base' => 'story'));
    }

  The page module's node hook module like:

    function page_node() {
      return array('page' => array('name' => t('page'), 'base' => 'page'));
    }

  However, more complex node modules like the project module and the
  flexinode module can use the 'base' parameter to specify a different base.

  The project module implements two node types, proejcts and issues, so it
  can do:

    function project_node() {
      return array(
       array('project_project' => array('name' => t('project'), 'base' => 'project'),
       array('project_issue' => array('name' => t('issue'), 'base' => 'project_issue'));
    }

  In the flexinode module's case there can only one base ...

  This hook will simplify the CCK, and will make it easy (or easier) to merge
  the story and page module.

  In addition, node_list() became node_get_types().  In addition, we created
  the following functions: node_get_name($type) and node_get_base($type).

- Patch #16962 by Neil: "The page module's long help text is a bunch of lies and then it briefly explains it's permissions. IMO it should just be taken out. I can't think of what help should be there."

- Refactored the queue module: removed the queue module's field from the node table.  With help from Gerhard.

- Slight addition to INSTALL.txt with regard to PHP versions.

- Updated/reworded some node type descriptions as per Boris' suggestions.

- Adding missing {} around a table name in update.php.

  TODO:
   + Update Doxygen hook list (if required).
   + Document this in the developer's upgrade section.

- Removed the link feature.  It was death code, bound to confuse people.

  TODO after Drupal 4.5.0: clean up the page module and remove additional cruft.

Modified version of #10230: Put placement of filter format selector in a module's hands, and move it below the relevant textarea.

- Patch #8179 by JonBob: reintroduced menu caching.

- Patch by JonBob: for consistency and readability, add brief descriptions of each source file inside the @file comment block at the head of the file. This helps with Doxygen indexing, and also allows neophytes to see what a file does immediately on opening the source, regardless of the organization of the hooks.

The Input formats - filter patch has landed. I still need to make update instructions for modules and update the hook docs.

Here's an overview of the changes:
1) Multiple Input formats: they are complete filter configurations (what filters to use, in what order and with which settings). Input formats are admin-definable, and usage of them is role-dependant. For example, you can set it up so that regular users can only use limited HTML, while admins can free HTML without any tag limitations.
The input format can be chosen per content item (nodes, comments, blocks, ...) when you add/edit them. If only a single format is available, there is no choice, and nothing changes with before.

The default install (and the upgrade) contains a basic set of formats which should satisfy the average user's needs.

2) Filters have toggles
Because now you might want to enable a filter only on some input formats, an explicit toggle is provided by the filter system. Modules do not need to worry about it and filters that still have their own on/off switch should get rid of it.

3) Multiple filters per module
This was necessary to accomodate the next change, and it's also a logical extension of the filter system.

4) Embedded PHP is now a filter
Thanks to the multiple input formats, I was able to move the 'embedded PHP' feature from block.module, page.module and book.module into a simple filter which executes PHP code. This filter is part of filter.module, and by default there is an input format 'PHP', restricted to the administrator only, which contains this filter.
This change means that block.module now passes custom block contents through the filter system.
As well as from reducing code duplication and avoiding two type selectors for page/book nodes, you can now combine PHP code with other filters.

5) User-supplied PHP code now requires <?php ?> tags.
This is required for teasers to work with PHP code. Because PHP evaluation is now just another step in the filter process, we can't do this. Also, because teasers are generated before filtering, this would result in errors when the teaser generation would cut off a piece of PHP code.

Also, regular PHP syntax explicitly includes the <?php ?> tags for PHP files, so it makes sense to use the same convention for embedded PHP in Drupal.

6) Filter caching was added.
Benchmarking shows that even for a simple setup (basic html filtering + legacy URL rewriting), filtercache can offer speedups. Unlike the old filtercache, this uses the normal cache table.

7) Filtertips were moved from help into a hook_filter_tips(). This was required to accomodate the fact that there are multiple filters per module, and that filter settings are format dependant. Shoehorning filter tips into _help was ugly and silly. The display of the filter tips is done through the input format selector, so filter_tips_short() no longer exists.

8) A more intelligent linebreak convertor was added, which doesn't stop working if you use block-level tags and which adds <p> tags.

- Patch #9543 by JonBob: added node-level access control!

- Patch #5347 by JonBob:

Here's a new patch that unifies the node/52 and book/view/52 paths for nodes. It involves a small change to hook_view(), which is discussed first:

Currently hook_view() expects node modules to return a themed node. However, each module does this the same way; they modify $node as necessary, then call theme('node', $node) and return the result. We can refactor this so that the calling function node_view() calls theme('node') instead. By doing this, it becomes possible for hook_nodeapi('view') to be called after hook_view() where the node contents are filtered, and before theme('node') where the body is enclosed in other HTML. This way the book module can insert its navigation into the body right before the theming.

Advantages of this refactoring:
- I can use it for book.module to remove the extra viewing path.
- The function of hook_nodeapi('view') becomes more like hook_view(), as neither will expect a return value.
- We more closely follow the flow of other nodeapi calls, which usually directly follow their corresponding specific node type hooks (instead of preceding them).
- The attachment.module people could use it to append their attachments in a list after the node.
- Gabor could use it instead of his filter perversion for his "articles in a series" module.
- A little less code in each view hook.
- The content hook is no longer needed, so that means even less code.

Disadvantages:
- Any modules written to use nodeapi('view') could be affected (but these would all be post-4.4 modules).
- Implementations of hook_view() would need to be updated (but return values would be ignored, so most would work without updates anyway).

Now the patch takes advantage of this API shift to inject its navigation at the end of all book nodes, regardless of the viewing path. In fact, since the paths become identical, I've removed the book/view handler entirely. We should probably provide an .htaccess rewrite for this (one is still needed for node/view/nn anyway). At the same time, there is a check in book_block() that shows the block appropriately on these pages.

- Patch #8080 by TDobes: added teaser support to the book and page module.

- Patch #8398 by TDobes: changed permissions for the blog, story, and page modules (and all occurrances elsewhere) to match their 4.4.x equivalents.  In the discussion when these permissions were introduced, it was decided that "edit own ..." was clearer, but "maintain personal ..." slipped into CVS HEAD anyway, while "edit own ..." landed in the 4.4.x branch.

Changes are as follows:
"maintain personal blog" -> "edit own blog" (aggregator.module, blog.module, blogapi.module)
"maintain personal pages" -> "edit own pages" (page.module)
"maintain personal stories" -> "edit own stories (story.module)

- Marked required fields on the node (story, book, page, blog) and comment
  forms using the $required argument of the form_ functions.

- Replaced all Optional's and Required's from the taxonomy forms with proper
  use of the form_ functions.

Please check your contributed modules too!