path: root/doc/todo/structured_page_data.mdwn

This is an idea from [[JoshTriplett]].  --[[Joey]]

* See further discussion at [[forum/an_alternative_approach_to_structured_data]].

Some uses of ikiwiki, such as for a bug-tracking system (BTS), move a bit away from the wiki end
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.
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
currently overrides the standard [[!cpan CGI::FormBuilder]] generated form,
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
[[!cpan CGI::FormBuilder]]. The attachment plugin uses the hack of bouilding
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.

Additional tie-ins:

* Pagespecs that can select pages with a field with a given value, etc.
  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.
* 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]]

> 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]]

>> Okie.  I've had a play with this.  A 'form' plugin is included inline below, but it is only a rough first pass to
>> get a feel for the design space.
>>
>> The current design defines a new type of page - a 'form'.  The type of page holds YAML data
>> defining a FormBuilder form.  For example, if we add a file to the wiki source `test.form`:

    ---
    fields:
      age:
        comment: This is a test
        validate: INT
        value: 15

>> The YAML content is a series of nested hashes.  The outer hash is currently checked for two keys:
>> 'template', which specifies a parameter to pass to the FromBuilder as the template for the
>> form, and 'fields', which specifies the data for the fields on the form.
>> each 'field' is itself a hash.  The keys and values are arguments to the formbuilder form method.
>> The most important one is 'value', which specifies the value of that field.
>>
>> Using this, the plugin below can output a form when asked to generate HTML.  The Formbuilder
>> arguments are sanitized (need a thorough security audit here - I'm sure I've missed a bunch of
>> holes).  The form is generated with default values as supplied in the YAML data.  It also has an
>> 'Update Form' button at the bottom.
>>
>>  The 'Update Form' button in the generated HTML submits changed values back to IkiWiki.  The
>> plugin captures these new values, updates the YAML and writes it out again.  The form is
>> validated when edited using this method.  This method can only edit the values in the form.
>> You cannot add new fields this way.
>>
>> It is still possible to edit the YAML directly using the 'edit' button.  This allows adding new fields
>> to the form, or adding other formbuilder data to change how the form is displayed.
>>
>> One final part of the plugin is a new pagespec function.  `form_eq()` is a pagespec function that
>> takes two arguments (separated by a ',').  The first argument is a field name, the second argument
>> a value for that field.  The function matches forms (and not other page types) where the named
>> field exists and holds the value given in the second argument.  For example:
    
    \[[!inline pages="form_eq(age,15)" archive="yes"]]
    
>> will include a link to the page generated above.

>>> Okie, I've just made another plugin to try and do things in a different way.
>>> This approach adds a 'data' directive.  There are two arguments, `key` and `value`.
>>> The directive is replaced by the value.  There is also a match function, which is similar
>>> to the one above.  It also takes two arguments, a key and a value.  It returns true if the
>>> page has that key/value pair in a data directive.  e.g.:

    \[[!data key="age" value="15"]]

>>> then, in another page:

    \[[!inline pages="data_eq(age,15)" archive="yes"]]

>>> I expect that we could have more match functions for each type of structured data,
>>> I just wanted to implement a rough prototype to get a feel for how it behaves.  -- [[Will]]

>> Anyway, here are the plugins.  As noted above these are only preliminary, exploratory, attempts. -- [[Will]]

>>>> I've just updated the second of the two patches below.  The two patches are not mutually
>>>> exclusive, but I'm leaning towards the second as more useful (for the things I'm doing). -- [[Will]]

I think it's awesome that you're writing this code to explore the problem
space, [[Will]] -- and these plugins are good stabs at at least part of it.
Let me respond to a few of your comments.. --[[Joey]]

On use cases, one use case is a user posting a bug report with structured
data in it. A template is one way, but then the user has to deal with the
format used to store the structured data. This is where a edit-time form
becomes essential.

> This was the idea with the 'form' plugin.  With the 'data' plugin I was exploring
> a different approach: try to keep the markup simple enough that the user can edit
> the markup directly, and still have that be ok.  I admit it is a stretch, but I thought
> it worth exploring.

Another use case is, after many such bugs have been filed,
wanting to add a new field to each bug report. To avoid needing to edit
every bug report it would be good if the fields in a bug report were
defined somewhere else, so that just that one place can be edited to add
the new field, and it will show up in each bug report (and in each bug
report's edit page, as a new form field).

> If I was going to do that, I'd use a perl script on a checked out
> workspace.  I think you're describing a rare operation and
> so I'd be happy not having a web interface for it.  Having said that,
> if you just wanted to change the form for *new* pages, then you
> can just edit the template used to create new pages.

Re the form plugin, I'm uncomfortable with tying things into
[[!cpan CGI::FormBuilder]] quite so tightly as you have.

> Yeah :).  But I wanted to explore the space and that was the
> easiest way to start.

CGI::FormBuilder
could easily change in a way that broke whole wikis full of pages. Also,
needing to sanitize FormBuilder fields with security implications is asking
for trouble, since new FormBuilder features could add new fields, or
add new features to existing fields (FormBuilder is very DWIM) that open
new security holes. 

> There is a list of allowed fields.  I only interpret those.

I think that having a type system, that allows defining specific types,
like "email address", by writing code (that in turn can use FormBuilder),
is a better approach, since it should avoid becoming a security problem.

> That would be possible.  I think an extension to the 'data' plugin might
> work here.

One specific security hole, BTW, is that if you allow the `validate` field,
FormBuilder will happily treat it as a regexp, and we don't want to expose
arbitrary perl regexps, since they can at least DOS a system, and can
probably be used to run arbitrary perl code.

> I validate the validate field :).  It only allows validate fields that match
> `/^[\w\s]+$/`.  This means you can really only use the pre-defined
> validation types in FormBuilder.

The data plugin only deals with a fairly small corner of the problem space,
but I think does a nice job at what it does. And could probably be useful
in a large number of other cases.

> I think the data plugin is more likely to be useful than the form plugin.
> I was thinking of extending the data directive by allowing an 'id' parameter.
> When you have an id parameter, then you can display a small form for that
> data element.  The submission handler would look through the page source
> for the data directive with the right id parameter and edit it.  This would
> make the data directive more like the current 'form' plugin.

> That is making things significantly more complex for less significant gain though. --[[Will]]

> Oh, one quick other note.  The data plugin below was designed to handle multiple
> data elements in a single directive.  e.g.

    \[[!data key="Depends on" link="bugs/bugA" link="bugs/bugB" value=6]]

> would match `data_eq(Depends on,6)`, `data_link(Depends on,bugs/bugA)`, `data_link(Depends on,bugs/bugB)`
> or, if you applied the patch in [[todo/tracking_bugs_with_dependencies]] then you can use 'defined pagespecs'
> such as `data_link(Depends on,~openBugs)`. <a id="another_kind_of_links" />The ability to label links like this allows separation of
> dependencies between bugs from arbitrary links.
>> This issue (the need for distinguished kinds of links) has also been brought up in other discussions: [[tracking_bugs_with_dependencies#another_kind_of_links]] (deps vs. links) and [[tag_pagespec_function]] (tags vs. links). --Ivan Z.

----

    #!/usr/bin/perl
    # Interpret YAML data to make a web form
    package IkiWiki::Plugin::form;
    
    use warnings;
    use strict;
    use CGI::FormBuilder;
    use IkiWiki 2.00;
    
    sub import {
    	hook(type => "getsetup", id => "form", call => \&getsetup);
    	hook(type => "htmlize", id => "form", call => \&htmlize);
    	hook(type => "sessioncgi", id => "form", call => \&cgi_submit);
    }
    
    sub getsetup () {
    	return
    		plugin => {
    			safe => 1,
    			rebuild => 1, # format plugin
    		},
    }
    
    sub makeFormFromYAML ($$$) {
    	my $page = shift;
    	my $YAMLString = shift;
    	my $q = shift;
    
    	eval q{use YAML};
    	error($@) if $@;
    	eval q{use CGI::FormBuilder};
    	error($@) if $@;
    	
    	my ($dataHashRef) = YAML::Load($YAMLString);
    	
    	my @fields = keys %{ $dataHashRef->{fields} };
    	
    	unshift(@fields, 'do');
    	unshift(@fields, 'page');
    	unshift(@fields, 'rcsinfo');
    	
    	# print STDERR "Fields: @fields\n";
    	
    	my $submittedPage;
    	
    	$submittedPage = $q->param('page') if defined $q;
    	
    	if (defined $q && defined $submittedPage && ! ($submittedPage eq $page)) {
    		error("Submitted page doensn't match current page: $page, $submittedPage");
    	}
    	
    	error("Page not backed by file") unless defined $pagesources{$page};
    	my $file = $pagesources{$page};
    	
    	my $template;
    	
    	if (defined $dataHashRef->{template}) {
    		$template = $dataHashRef->{template};
    	} else {
    		$template = "form.tmpl";
    	}
    	
    	my $form = CGI::FormBuilder->new(
    		fields => \@fields,
    		charset => "utf-8",
    		method => 'POST',
    		required => [qw{page}],
    		params => $q,
    		action => $config{cgiurl},
    		template => scalar IkiWiki::template_params($template),
    		wikiname => $config{wikiname},
    		header => 0,
    		javascript => 0,
    		keepextras => 0,
    		title => $page,
    	);
    	
    	$form->field(name => 'do', value => 'Update Form', required => 1, force => 1, type => 'hidden');
    	$form->field(name => 'page', value => $page, required => 1, force => 1, type => 'hidden');
    	$form->field(name => 'rcsinfo', value => IkiWiki::rcs_prepedit($file), required => 1, force => 0, type => 'hidden');
    	
    	my %validkey;
    	foreach my $x (qw{label type multiple value fieldset growable message other required validate cleanopts columns comment disabled linebreaks class}) {
    		$validkey{$x} = 1;
    	}
    
    	while ( my ($name, $data) = each(%{ $dataHashRef->{fields} }) ) {
    		next if $name eq 'page';
    		next if $name eq 'rcsinfo';
    		
    		while ( my ($key, $value) = each(%{ $data }) ) {
    			next unless $validkey{$key};
    			next if $key eq 'validate' && !($value =~ /^[\w\s]+$/);
    		
    			# print STDERR "Adding to field $name: $key => $value\n";
    			$form->field(name => $name, $key => $value);
    		}
    	}
    	
    	# IkiWiki::decode_form_utf8($form);
    	
    	return $form;
    }
    
    sub htmlize (@) {
    	my %params=@_;
    	my $content = $params{content};
    	my $page = $params{page};
    
    	my $form = makeFormFromYAML($page, $content, undef);
    
    	return $form->render(submit => 'Update Form');
    }
    
    sub cgi_submit ($$) {
    	my $q=shift;
    	my $session=shift;
    	
    	my $do=$q->param('do');
    	return unless $do eq 'Update Form';
    	IkiWiki::decode_cgi_utf8($q);
    
    	eval q{use YAML};
    	error($@) if $@;
    	eval q{use CGI::FormBuilder};
    	error($@) if $@;
    	
    	my $page = $q->param('page');
    	
    	return unless exists $pagesources{$page};
    	
    	return unless $pagesources{$page} =~ m/\.form$/ ;
    	
    	return unless IkiWiki::check_canedit($page, $q, $session);
    
    	my $file = $pagesources{$page};
    	my $YAMLString = readfile(IkiWiki::srcfile($file));
    	my $form = makeFormFromYAML($page, $YAMLString, $q);
    
    	my ($dataHashRef) = YAML::Load($YAMLString);
    
    	if ($form->submitted eq 'Update Form' && $form->validate) {
    		
    		#first update our data structure
    		
    		while ( my ($name, $data) = each(%{ $dataHashRef->{fields} }) ) {
    			next if $name eq 'page';
    			next if $name eq 'rcs-data';
    			
    			if (defined $q->param($name)) {
    				$data->{value} = $q->param($name);
    			}
    		}
    		
    		# now write / commit the data
    		
    		writefile($file, $config{srcdir}, YAML::Dump($dataHashRef));
    
    		my $message = "Web form submission";
    
    		IkiWiki::disable_commit_hook();
    		my $conflict=IkiWiki::rcs_commit($file, $message,
    			$form->field("rcsinfo"),
    			$session->param("name"), $ENV{REMOTE_ADDR});
    		IkiWiki::enable_commit_hook();
    		IkiWiki::rcs_update();
    
    		require IkiWiki::Render;
    		IkiWiki::refresh();
    
    		IkiWiki::redirect($q, "$config{url}/".htmlpage($page)."?updated");
    
    	} else {
    		error("Invalid data!");
    	}
    
    	exit;
    }
    
    package IkiWiki::PageSpec;
    
    sub match_form_eq ($$;@) {
    	my $page=shift;
    	my $argSet=shift;
    	my @args=split(/,/, $argSet);
    	my $field=shift @args;
    	my $value=shift @args;
    
    	my $file = $IkiWiki::pagesources{$page};
    	
    	if ($file !~ m/\.form$/) {
    		return IkiWiki::FailReason->new("page is not a form");
    	}
    	
    	my $YAMLString = IkiWiki::readfile(IkiWiki::srcfile($file));
    
    	eval q{use YAML};
    	error($@) if $@;
    
    	my ($dataHashRef) = YAML::Load($YAMLString);
    
    	if (! defined $dataHashRef->{fields}->{$field}) {
    		return IkiWiki::FailReason->new("field '$field' not defined in page");
    	}
    
    	my $formVal = $dataHashRef->{fields}->{$field}->{value};
    
    	if ($formVal eq $value) {
    		return IkiWiki::SuccessReason->new("field value matches");
    	} else {
    		return IkiWiki::FailReason->new("field value does not match");
    	}
    }
    
    1

----

    #!/usr/bin/perl
    # Allow data embedded in a page to be checked for
    package IkiWiki::Plugin::data;
    
    use warnings;
    use strict;
    use IkiWiki 2.00;
    
    my $inTable = 0;
    
    sub import {
    	hook(type => "getsetup", id => "data", call => \&getsetup);
    	hook(type => "needsbuild", id => "data", call => \&needsbuild);
    	hook(type => "preprocess", id => "data", call => \&preprocess, scan => 1);
    	hook(type => "preprocess", id => "datatable", call => \&preprocess_table, scan => 1);	# does this need scan?
    }
    
    sub getsetup () {
    	return
    		plugin => {
    			safe => 1,
    			rebuild => 1, # format plugin
    		},
    }
    
    sub needsbuild (@) {
    	my $needsbuild=shift;
    	foreach my $page (keys %pagestate) {
    		if (exists $pagestate{$page}{data}) {
    			if (exists $pagesources{$page} &&
    			    grep { $_ eq $pagesources{$page} } @$needsbuild) {
    				# remove state, it will be re-added
    				# if the preprocessor directive is still
    				# there during the rebuild
    				delete $pagestate{$page}{data};
    			}
    		}
    	}
    }
    
    sub preprocess (@) {
    	my @argslist = @_;
    	my %params=@argslist;
    	
    	my $html = '';
    	my $class = defined $params{class}
    			? 'class="'.$params{class}.'"'
    			: '';
    	
    	if ($inTable) {
    		$html = "<th $class >$params{key}:</th><td $class >";
    	} else {
    		$html = "<span $class >$params{key}:";
    	}
    	
    	while (scalar(@argslist) > 1) {
    		my $type = shift @argslist;
    		my $data = shift @argslist;
    		if ($type eq 'link') {
    			# store links raw
    			$pagestate{$params{page}}{data}{$params{key}}{link}{$data} = 1;
    			my $link=IkiWiki::linkpage($data);
    			add_depends($params{page}, $link);
    			$html .= ' ' . htmllink($params{page}, $params{destpage}, $link);
    		} elsif ($type eq 'data') {
    			$data = IkiWiki::preprocess($params{page}, $params{destpage}, 
    				IkiWiki::filter($params{page}, $params{destpage}, $data));
    			$html .= ' ' . $data;
    			# store data after processing - allows pagecounts to be stored, etc.
    			$pagestate{$params{page}}{data}{$params{key}}{data}{$data} = 1;
    		}
    	}
    		
    	if ($inTable) {
    		$html .= "</td>";
    	} else {
    		$html .= "</span>";
    	}
    	
    	return $html;
    }
    
    sub preprocess_table (@) {
    	my %params=@_;
    
    	my @lines;
    	push @lines, defined $params{class}
    			? "<table class=\"".$params{class}.'">'
    			: '<table>';
    
    	$inTable = 1;
    
    	foreach my $line (split(/\n/, $params{datalist})) {
    		push @lines, "<tr>" . IkiWiki::preprocess($params{page}, $params{destpage}, 
    			IkiWiki::filter($params{page}, $params{destpage}, $line)) . "</tr>";
    	}
    
    	$inTable = 0;
    
    	push @lines, '</table>';
    
    	return join("\n", @lines);
    }
    
    package IkiWiki::PageSpec;
    
    sub match_data_eq ($$;@) {
    	my $page=shift;
    	my $argSet=shift;
    	my @args=split(/,/, $argSet);
    	my $key=shift @args;
    	my $value=shift @args;
    
    	if (! exists $IkiWiki::pagestate{$page}{data}) {
    		return IkiWiki::FailReason->new("page does not contain any data directives");
    	}
    	
    	if (! exists $IkiWiki::pagestate{$page}{data}{$key}) {
    		return IkiWiki::FailReason->new("page does not contain data key '$key'");
    	}
    	
    	if ($IkiWiki::pagestate{$page}{data}{$key}{data}{$value}) {
    		return IkiWiki::SuccessReason->new("value matches");
    	} else {
    		return IkiWiki::FailReason->new("value does not match");
    	}
    }
    
    sub match_data_link ($$;@) {
    	my $page=shift;
    	my $argSet=shift;
    	my @params=@_;
    	my @args=split(/,/, $argSet);
    	my $key=shift @args;
    	my $value=shift @args;
    
    	if (! exists $IkiWiki::pagestate{$page}{data}) {
    		return IkiWiki::FailReason->new("page $page does not contain any data directives and so cannot match a link");
    	}
    	
    	if (! exists $IkiWiki::pagestate{$page}{data}{$key}) {
    		return IkiWiki::FailReason->new("page $page does not contain data key '$key'");
    	}
    	
    	foreach my $link (keys %{ $IkiWiki::pagestate{$page}{data}{$key}{link} }) {
    		# print STDERR "Checking if $link matches glob $value\n";
    		if (match_glob($link, $value, @params)) {
    			return IkiWiki::SuccessReason->new("Data link on page $page with key $key matches glob $value: $link");
    		}
    	}
    
    	return IkiWiki::FailReason->new("No data link on page $page with key $key matches glob $value");
    }
    
    1