As noted in tag pagespec function, there is a "misbehavior" of a tagged() pagespec: it matches even pages which have plain links to the tag page.

And in general, it would be quite useful to be able to distinguish different kinds of links: one more kind, in addition to "tag", is "bug dependency" noted in structured page data and tracking bugs with dependencies.

It could distinguish the links by the rel= attribute. (Tags already receive a special rel-class.) This means there is a general need for a syntax to specify user-defined rel-classes on wikilink (then bug deps would simply use their special rel-class, either directly, or through a special directive like [[!depends ]]), and to refer to them in pagespecs (in forward and backward direction).

Besides pagespecs, the rel= attribute could be used for styles. --Ivan Z.

FWIW, the add_link function introduced in a recent release adds an abstraction that could be used to get part of the way there to storing data about different types of links. That function could easily be extended to take an optional third parameter specifying the link type.

Then there's the question of how to store and access the data. %links does not offer a good way to add additional information about links. Now, we could toss %links entirely and switch to an accessor function, but let's think about not doing that..

The data that seems to be needed is basically a deep hash, so one could check $linktype{$page}{tag}{$link} to see if the page contains a link of the given type. (Note that pages could contain links that were duplicates except for their types.)

There would be some data duplication, unfortuantly, but if %linktype is not populated for regular wikilinks, it would at least be limited to tags and other unusual link types, so not too bad.

%linktype could be stored in %pagestate.. if so the actual use might look like $pagestate{$page}{linktype}{tag}{$link}. That could be implemented by the tag plugin right now with no core changes. (BTW, then I originally wrote tag, pagestate was not available, which is why I didn't make it differentiate from normal links.) Might be better to go ahead and add the variable to core though. --Joey

I've implemented this with the data structure you suggested, except that I called it %typedlinks instead of %linktype (it seemed to make more sense that way). I also ported tag to it, and added a tagged_is_strict config option. See below! --smcv

I saw somewhere else here some suggestions for the wiki-syntax for specifying the relation name of a link. One more suggestion---the syntax used in Semantic MediaWiki, like this:

... the capital city is [[Has capital::Berlin]] ...

So a part of the effect of [[\[[!taglink TAG\]\]|plugins/tag]] could be represented as something like [[tag::TAG]] or (more understandable relation name in what concerns the direction) [[tagged::TAG]].

I don't have any opinion on this syntax (whether it's good or not)...--Ivan Z.


Available in a git repository branch.
Branch: smcv/ready/link-types
Author: Simon McVittie

Documentation for smcv's branch

added to pagespec

  • "typedlink(type glob)" - matches pages that link to a given page (or glob) with a given link type. Plugins can create links with a specific type: for instance, the tag plugin creates links of type tag.

added to tag

If the tagged_is_strict config option is set, tagged() will only match tags explicitly set with tag or taglink; if not (the default), it will also match any other WikiLinks to the tag page.

added to write

%typedlinks

The %typedlinks hash records links of specific types. Do not modify this hash directly; call add_link(). The keys are page names, and the values are hash references. In each page's hash reference, the keys are link types defined by plugins, and the values are hash references with link targets as keys, and 1 as a dummy value, something like this:

$typedlinks{"foo"} = {
    tag => { short_word => 1, metasyntactic_variable => 1 },
    next_page => { bar => 1 },
};

Ordinary WikiLinks appear in %links, but not in %typedlinks.

add_link($$;$)

This adds a link to %links, ensuring that duplicate links are not added. Pass it the page that contains the link, and the link text.

An optional third parameter sets the link type (undef produces an ordinary WikiLink).

Review

Some code refers to oldtypedlinks, and other to oldlinktypes. --Joey

Oops, I'll fix that. That must mean missing test coverage, too :-( --s

A test suite for the dependency resolver would be nice. --Joey

Bug fixed, I think. A test suite for the dependency resolver seems more ambitious than I want to get into right now, but I added a unit test for this part of it... --s

I'm curious what your reasoning was for adding a new variable rather than using pagestate. Was it only because you needed the old version to detect change, or was there other complexity? --J

You seemed to be more in favour of adding it to the core in your proposal above, so I assumed that'd be more likely to be accepted :-) I don't mind one way or the other - %typedlinks costs one core variable, but saves one level of hash nesting. If you're not sure either, then I think the decision should come down to which one is easier to document clearly - I'm still unhappy with my docs for %typedlinks, so I'll try to write docs for it as pagestate and see if they work any better. --s

On reflection, I don't think it's any better as a pagestate, and the contents of pagestates (so far) aren't documented for other plugins' consumption, so I'm inclined to leave it as-is, unless you want to veto that. Loose rationale: it needs special handling in the core to be a dependency type (I re-used the existing link type), it's API beyond a single plugin, and it's really part of the core parallel to pagestate rather than being tied to a specific plugin. Also, I'd need to special-case it to have ikiwiki not delete it from the index, unless I introduced a dummy typedlinks plugin (or just hook) that did nothing... --s

I have not convinced myself this is a real problem, but.. If a page has a typed link, there seems to be no way to tell if it also has a separate, regular link. add_link will add to @links when adding a typed, or untyped link. If only untyped links were recorded there, one could tell the difference. But then typed links would not show up at all in eg, a linkmap, unless it was changed to check for typed links too. (Or, regular links could be recorded in typedlinks too, with a empty type. (Bloaty.)) --J

I think I like the semantics as-is - I can't think of any reason why you'd want to ask the question "does A link to B, not counting tags and other typed links?". A typed link is still a link, in my mind at least. --s

Me neither, let's not worry about it. --Joey

I suspect we could get away without having tagged_is_strict without too much transitional trouble. --Joey

If you think so, I can delete about 5 LoC. I don't particularly care either way; Jon expressed concern about people relying on the current semantics, on one of the pages requesting this change. --s

Removed in a newer version of the branch. --s

I might have been wrong to introduce typedlink(tag foo). It's not very user-friendly, and is more useful as a backend for other plugins that as a feature in its own right - any plugin introducing a link type will probably also want to have its own preprocessor directive to set that link type, and its own pagespec function to match it. I wonder whether to make a typedlink plugin that has the typedlink pagespec match function and a new [[!typedlink to="foo" type="bar"]] though... --smcv

I agree, per-type matchers are more friendly and I'm not enamored of the multi-parameter pagespec syntax. --Joey

Removed in a newer version of the branch. I re-introduced it as a plugin in smcv/typedlink, but I don't think we really need it. --s


I am ready to merge this, but I noticed one problem -- since match_tagged now only matches pages with the tag linktype, a wiki will need to be rebuilt on upgrade in order to get the linktype of existing tags in it recorded. So there needs to be a NEWS item about this and the postinst modified to force the rebuild.

Done, although you'll need to plug in an appropriate version number when you release it. Is there a distinctive reminder string you grep for during releases? I've used UNRELEASED for now. --smcv

Also, the ready branch adds typedlink() to pagespec, but you removed that feature as documented above. --Joey

Done. --s