Ha! I just implemented this, and I _generally_ like what I came up with. I'm thinking of writing it up as a document or releasing it if work and time will allow me to. I can share my basic implementation here.
The overview: First, I created an ITaggable interface: class ITaggable(Interface): """ Tags are a field of keywords """ tags = zope.schema.Tuple( title=u"Tags", description=_(u"Keyword Tags"), value_type=zope.schema.TextLine(title=u"Tag", required=True) ) (I don't know if the 'required=True' needs to be there in the value_type). Then for my particular implementation I made a property for my object that mapped to the dublin core 'subjects' property. I wrote a custom 'DublinCoreProperty' for this. You can do your own thing if you like. My reasoning - I want tags to be thought of as 'tags', but since I'm working with a Zope 3 content object that has dublin core, I may as well take advantage of what's there, so I basically adapt 'subjects' to 'tags'. This could also be done with an adapter, and you could make everything taggable, using something like this: from zope.app import zapi from zope.interface import implements from zope.app.dublincore.interfaces import IZopeDublinCore from zope.app.annotation.interfaces import IAnnotatable class TaggableSubjects(object): """ Adapts annotatable objects to expose the 'subjects' field as 'tags' """ implements(ITaggable) zapi.adapts(IAnnotatable) def __init__(self, context): self.dc = IZopeDublinCore(context) def _getTags(self): return self.dc.subjects def _setTags(self, value): self.dc.subjects = value tags = property(_getTags, _setTags) That's a pretty simplistic adapter - you'd probably want more error checking. In any case, I just want to point out that in my implementation, tags are just a tuple of strings on any ITaggable object. Now - how about querying them? This was hard until recently. I used zc.catalog out of the Zope subversion repository, because it has what's known as a SetIndex. A SetIndex allows for querying against, um, sets of data. So with a catalog or extent catalog (the 'extent catalog' is part of the 'zc.catalog' package), I create an index, 'tags', as a SetIndex, which uses the 'tags' attribute of the ITaggable interface. The catalog will try to adapt an indexed object to ITaggable, so you can use something like the above adapter to give 'tags' to existing objects. Or implement it directly. In any case, this is a really cool feature, adaptation. There are already event listeners in place for the catalog / indexes to respond to object modification and deletion events, so you don't have to remove it from all objects it was tagged with. In my scenario, tags are not smart objects (I did have an earlier implementation that had them as such, before catalogs and the SetIndex were available). To query them, you just need to search the catalog. I use hurry.query, and with that it looks like this: from zope.app import zapi from hurry.query.interfaces import IQuery from hurry.query import set as setquery (inside whatever class / function / method you have for querying): tags = self.tags # a tuple of strings: ('zope','tags','tagging','catalog') tagindex = ('catalog', 'tags') # (ICatalog name, index name) - its how hurry.query works query = setquery.AllOf(tagindex, tags) queryUtility = zapi.getUtility(IQuery) results = self._results = queryUtility.searchResults(query) With hurry.query you can easily add other queries on to this, such as full text indexing, date sets, and more, depending on your indexes. Anyways, there's the heart of what I use in this knowledge base application I've been working on for the past couple of weeks. It's an application that stores all of its articles in one big container, generates names automatically (yay INameChooser!). You just add an article and tag it, and use tags for navigation. Although... that's where things get a little more tricky. I do have a system in place that lets me have URLs like: http://kbase.example.com/@@tags/zope/catalog/setindex and that will turn into a query for all documents that have the tags ('zope', 'catalog', 'setindex'). I have a solution in place that I like that actually generates objects on the fly for each point in the URL. These aren't Tag objects (since tags are just keyword strings), but TagMatchers. A TagMatcher is a simple interface and object that holds a list of tags and runs a catalog query for the matching tags. This query has to be called explicitly. I then have some IBrowserPublisher adapters and views that are used to manage the traversal. Basically every time a name is traversed, a new tagmatcher is made, with the understanding that its parent is a TagMatcher. The parents tags plus the new name are passed into the new matcher. At the end of the path, a default view for ITagMatcher calls 'process' which causes the query to be run and then it gets and sorts the results for display. One of these days I'm going to make a kick-ass site where I can write documents about how I did this. It's cool when you get it working, but figuring out the traversal options and all of that just to have cool URLs like 'tags/foo/bar/baz' when 'foo', 'bar', and 'baz' are all objects being created on the fly was not easy. Anyways, the summary is: I used Catalog with a SetIndex (zc.catalog, look at http://svn.zope.org/Sandbox/zc/catalog/ ) for ``ITaggable.tags``. I use publishTraverse and views to do tag traversal. I have an AllTags view that when called renders a page showing tags in the 'cloud' style (showing popularity by size and with a number of matching), and when that's traversed through the TagMatcher sequence starts which creates these objects for each point along the URL that collect each tag name and when the end is reached, the default view runs the query and presents the results. You don't delete 'tags', since they're not first class objects. The set index deals with everything and seems quite efficient. You can do your option #2. My previous implementation, done last summer, worked that way.. kindof. Tags were more of a first class object and intids and all of that were used. But this new implementation I'm using seems a lot faster and more efficient. There were some hard things (namely the traversal stuff for nice URLs), but as far as indexing and querying goes, the Catalog/SetIndex combination works great. On 1/8/06, Igor Stroh <[EMAIL PROTECTED]> wrote: > Hi there, > > I'm trying to figure out the best way to implements a general > solution for tags (like tags on flickr or del.icio.us). So far > I've two different approaches, maybe someone could comment on > them: > > 1) Store the tags as attributes of the particular content types. > Searching for tagged content is handled by a catalog index. > New tags are created by simply adding them to the object, > a collection of tags is the set of all tags from the catalog. > Drawbacks: centralized tag management is inefficient, e.g. > deleting a tag implies searching for the particular tag and > removing it from all objects it was tagged with. > > 2) Store tags in a local utility. The utility manages a > tag -> intids mapping (similar to a catalog index). > Searching for tags is easy - just query the mapping > for appropriate keys. New tags are created by adding > a new key to the mapping, a (weighted) collection of > tags is the list if mapping keys. > Drawbacks: I can't think of any right now > > Has anyone done that before? Maybe I'm just reinventing the wheel... > > Thanks in advance, > Igor > _______________________________________________ > Zope3-users mailing list > Zope3-users@zope.org > http://mail.zope.org/mailman/listinfo/zope3-users > -- -- Jeff Shell _______________________________________________ Zope3-users mailing list Zope3-users@zope.org http://mail.zope.org/mailman/listinfo/zope3-users