(This mainly discusses the original implementation (chrysn's). --smcv)
This is a nice idea, I do have my gripes about the imeplementation.
Assuming that the index's list is in mdwn format is not ideal. I guess the other way to do it would be to make the index be a directive, something like: [[!trail pages="foo bar baz"]]. Assuming that a flat trail structure is enough, otherwise you'd have to get more fancy.
The trailinclude seems a bit redundant with inline, and wanting to inline together all pages in a trail for printing or whatever seems like an unusual use case anyway?
The !trail directive could be simplified to just [[!trail my_indexpage]]. But I wonder if needing to add this directive to every page is the best approach. Alternate approach would be to make the trail index cause breadcrums to be automatically inserted at the top of every page on the trail. (You'd have to use a directive to define the index for that to work.)
Revisiting this, after effectively reimplementing a small version of it in album: it occurs to me that might be a more "ikiwiki-like" way we could get this functionality.
In the index page, you either want an inline, or a list of links. In the former case, maybe we could extend inline like this:
[[!inline ... blah blah ... trail=yes]]
to make it remember the pages it inlined, in order, in the pagestate; in the latter case, we could replace the wikilinks with a directive, an operation something like this in diff notation:
- [[one]] - the unit - [[two]] - the base of binary - [[three|3]] - is a crowd + [[!trailitem one]] - the unit + [[!trailitem two]] - the base of binary + [[!trailitem three|3]] - is a crowd
and have that directive remember the pages in order.
In both cases, a scan() hook could clear the list before starting to scan, then the inline or trailitem preprocessor directive could run in the scan stage as well as the render stage (in the case of inline, there'd be a very early return if trail=yes was not given, and an early return after collecting and sorting the pages if not actually rendering).
This would mean that the contents of the trail, and a list of trails in which each page can be found, would already be in the pagestate by the time any page was rendered, so we'd be able to use them for output, either in a pagetemplate() hook or a [[!trail ]] preprocessor directive.
This way, my album plugin could be turned inside out: instead of precomputing the pages to be inlined, then using pagenames to get them into the inline, it could just do the inline, then incorporate the output of [[!trail ]] into the template rendered for [[!albumimage ]] on each viewer page. (Also, the viewers wouldn't necessarily need to reference the album, only the other way round.)
Using a pagetemplate() hook to stuff the next/previous links into page.tmpl would actually be a bit unfortunate for [[!album ]], because that plugin definitely wants to style the next/previous links as a thumbnail, which means there'd have to be a way to affect the style - perhaps by arranging for album's pagetemplate hook to run after trail's, or perhaps by having trail's pagetemplate hook disable itself for pages that contain a [[!trail ]] directive.