+[Jekyll](http://jekyllrb.com/)'s implementation looks rather neat:
+
+* <https://github.com/mojombo/jekyll/wiki/template-data>
+* <https://github.com/mojombo/jekyll/blob/master/lib/jekyll/generators/pagination.rb>
+
+
+
+> This is a perfect use for [[todo/wikitrails]], of which my
+> [[plugins/contrib/trail]] plugin is an implementation. Code review on that
+> plugin would be welcome; it might even get merged one day.
+>
+>> The trail plugin is very likely to be merged soon, and is already
+>> available. So, closing this bug report [[done]] --[[Joey]]
+>
+> Unfortunately, IkiWiki blogs use a [[ikiwiki/PageSpec]] to define the set of
+> "posts" in the blog (through which the next/prev trail should range), and
+> the current implementation of [[plugins/contrib/trail]] in terms of typed
+> links would have a circular dependency if used with a PageSpec: typed links
+> have to be added before PageSpecs are evaluated, because "A links to B" is
+> something that can be in a PageSpec; but if you want to add typed links
+> ("A is part of trail B" in this case) based on a PageSpec, then the PageSpec
+> must be evaluated before the typed links can be added. Chicken/egg.
+>
+> One solution would be to make the trail plugin use its own data
+> structure, like [[plugins/contrib/album]] used to do, instead of typed
+> links: at scan time, the trail plugin would just record what the PageSpec
+> was, and delay actually *evaluating* the PageSpec until the beginning
+> of the `render` stage (after all pages have been scanned). This
+> reduces the generic usefulness of typed links, though - in particular
+> you can no longer use "is part of trail A" in a PageSpec. --[[smcv]]
+
+>> Version 3 of [[plugins/contrib/trail]] now does this. For `traillink`
+>> and `trailitem` it additionally adds a typed link, which it does not
+>> itself consume; for `trailinline` and `trail` it doesn't. --[[smcv]]
+
+>>> Indeed, I know the problem; I ran into the same kind of thing with my [[plugins/contrib/report]] plugin and its `trail` concept.
+>>> I simply had to declare that one couldn't use "trail" and "maketrail" options within the same report. That way, "maketrail" will add links in the "scan" pass, and "trail" will test for links in the "build" pass. That usually works. --[[KathrynAndersen]]
+
+>>>> I'm not sure that even that is *quite* right: if your `trail` takes
+>>>> pagespecs as arguments, then it's potentially evaluating those pagespecs
+>>>> before all pages have been scanned, which could mean it lists pages
+>>>> which matched the spec before a recent change, or doesn't list pages
+>>>> which didn't previously match the spec but do now.
+>>>>
+>>>> In version 3 of [[plugins/contrib/trail]] I ended up storing
+>>>> uninterpreted pagespecs and links at scan time, and evaluating them the
+>>>> first time a page is built. I *think* that's sufficiently lazy to give
+>>>> the right answer... --[[smcv]]
+
+>> Do you have an example? --[[hendry]]
+
+>>> Now linked on the plugin's page - it doesn't pretend to be a blog, but
+>>> [the second demo](http://demo.hosted.pseudorandom.co.uk/trail2/)
+>>> is a `trailinline`, so you could do that with blog posts just as well.
+>>> Making [[plugins/contrib/album]] require `trail` v3, and trying it out
+>>> on my blog, are next on the list.
+>>> --[[smcv]]
+
+>>>> Sorry thank link <http://demo.hosted.pseudorandom.co.uk/trail2/> doesn't work. I get a forbidden. --[[hendry]]
+
+
+[[wishlist]]