ikiwiki/doc/todo/structured_page_data.mdwn

106 lines
5.1 KiB
Plaintext
Raw Normal View History

2008-09-04 20:35:35 +02:00
This is an idea from [[JoshTriplett]]. --[[Joey]]
2008-09-05 00:32:56 +02:00
Some uses of ikiwiki, such as for a bug-tracking system (BTS), move a bit away from the wiki end
2008-09-04 20:35:35 +02:00
of the spectrum, and toward storing structured data about a page or instead
of a page.
For example, in a bug report you might want to choose a severity from a
list, enter a version number, and have a bug submitter or owner recorded,
etc. When editing online, it would be nice if these were separate fields on
the form, rather than the data being edited in the big edit form.
There's a tension here between remaining a wiki with human-editable source
files, containing freeform markup, and more structured data storage. I
think that it would be best to include the structured data in the page,
using a directive. Something like:
part of page content
\[[data yaml="<arbitrary yaml here>"]]
rest of page content
As long as the position of the directive is not significant, it could be
stripped out when web editing, the yaml used to generate/populate form fields,
and then on save, the directive regenerated and inserted at top/bottom of
the page.
Josh thinks that yaml is probably a good choice, but the source could be a
`.yaml` file that contains no directives, and just yaml. An addition
complication in this scenario is, if the yaml included wiki page formatted content,
ikiwiki would have to guess or be told what markup language it used.
Either way, the yaml on the page would encode fields and their current content.
Information about data types would be encoded elsewhere, probably on a
parent page (using a separate directive). That way, all child pages could
be forced to have the same fields.
There would be some simple types like select, boolean, multiselect, string, wiki markup.
2008-09-04 20:35:35 +02:00
Probably lists of these (ie, list of strings). Possibly more complex data
structures.
It should also be possible for plugins to define new types, and the type
definitions should include validation of entered data, and how to prompt
the user for the data.
This seems conceptually straightforward, if possibly quite internally
complex to handle the more complicated types and validation.
One implementation wrinkle is how to build the html form. The editpage.tmpl
2008-09-12 01:40:41 +02:00
currently overrides the standard [[!cpan CGI::FormBuilder]] generated form,
2008-09-04 20:35:35 +02:00
which was done to make the edit page be laid out in a nice way. This,
however, means that new fields cannot be easily added to it using
2008-09-12 01:40:41 +02:00
[[!cpan CGI::FormBuilder]]. The attachment plugin uses the hack of bouilding
2008-09-04 20:35:35 +02:00
up html by hand and dumping it into the form via a template variable.
It would be nice if the type implementation code could just use
FormBuilder, since its automatic form generation, and nice field validation
model is a perfect match for structured data. But this problem with
editpage.tmpl would have to be sorted out to allow that.
2008-09-04 21:48:56 +02:00
Additional tie-ins:
* Pagespecs that can select pages with a field with a given value, etc.
2008-09-08 22:47:40 +02:00
This should use a pagespec function like field(fieldname, value). The
semantics of this will depend on the type of the field; text fields will
match value against the text, and link fields will check for a link
matching the pagespec value.
2008-09-04 21:48:56 +02:00
* The search plugin could allow searching for specific fields with specific
content. (xapian term search is a good fit).
See also:
[[tracking_bugs_with_dependencies]]
2008-09-11 14:27:47 +02:00
> I was also thinking about this for bug tracking. I'm not sure what
> sort of structured data is wanted in a page, so I decided to brainstorm
> use cases:
>
> * You just want the page to be pretty.
> * You want to access the data from another page. This would be almost like
> like a database lookup, or the OpenOffice Calc [VLookup](http://wiki.services.openoffice.org/wiki/Documentation/How_Tos/Calc:_VLOOKUP_function) function.
> * You want to make a pagespec depend upon the data. This could be used
> for dependancy tracking - you could match against pages listed as dependencies,
> rather than all pages linked from a given page.
>
>The first use case is handled by having a template in the page creation. You could
>have some type of form to edit the data, but that's just sugar on top of the template.
>If you were going to have a web form to edit the data, I can imagine a few ways to do it:
>
> * Have a special page type which gets compiled into the form. The page type would
> need to define the form as well as hold the stored data.
> * Have special directives that allow you to insert form elements into a normal page.
>
>I'm happy with template based page creation as a first pass...
>
>The second use case could be handled by a regular expression directive. eg:
>
> \[[regex spec="myBug" regex="Depends: ([^\s]+)"]]
>
> The directive would be replaced with the match from the regex on the 'myBug' page... or something.
>
>The third use case requires a pagespec function. One that matched a regex in the page might work.
>Otherwise, another option would be to annotate links with a type, and then check the type of links in
>a pagespec. e.g. you could have `depends` links and normal links.
>
>Anyway, I just wanted to list the thoughts. In none of these use cases is straight yaml or json the
>obvious answer. -- [[Will]]