+
+> This plugin is pretty ambitious, and is unfinished, so I'd recommend
+> playing with a normal IkiWiki installation for a bit, then trying
+> out this plugin when you've mastered the basics of IkiWiki. --[[smcv]]
+
+----
+
+You had wanted my feedback on the design of this. I have not looked at the
+code or tried it yet, but here goes. --[[Joey]]
+
+* Needing to create the albumimage "viewer" pages for each photo
+ seems like it will become a pain. Everyone will need to come up
+ with their own automation for it, and then there's the question
+ of how to automate it when uploading attachments. -J
+
+> There's already a script (ikiwiki-album) to populate a git
+> checkout with skeleton "viewer" pages; I was planning to make a
+> specialized CGI interface for albums after getting feedback from
+> you (since the requirements for that CGI interface change depending
+> on the implementation). I agree that this is ugly, though. -s
+
+>> Would you accept a version where the albumimage "viewer" pages
+>> could be 0 bytes long, at least until metadata gets added?
+>>
+>> The more I think about the "binaries as first-class pages" approach,
+>> the more subtle interactions I notice with other plugins. I
+>> think I'm up to needing changes to editpage, comments, attachment
+>> and recentchanges, plus adjustments to img and Render (to reduce
+>> duplication when thumbnailing an image with a strange extension
+>> while simultaneously changing the extension, and to hardlink/copy
+>> an image with a strange extension to a differing target filename
+>> with the normal extension, respectively). -s
+
+>>> Now that we have `add_autofile` I can just create viewer pages
+>>> whenever there's an image to view. The current version of the
+>>> branch does that. -s
+
+* With each viewer page having next/prev links, I can see how you
+ were having the scalability issues with ikiwiki's data structures
+ earlier! -J
+
+> Yeah, I think they're a basic requirement from a UI point of view
+> though (although they don't necessarily have to be full wikilinks).
+> -s
+
+>> I think that with the new dependency types system, the dependencies for
+>> these can be presence dependencies, which will probably help with
+>> avoiding rebuilds of a page if the next/prev page is changed.
+>> (Unless you use img to make the thumbnails for those links, then it
+>> would rebuild the thumbnails anyway. Have not looked at the code.) --[[Joey]]
+
+>>> I do use img. -s
+
+* And doesn't each viewer page really depend on every other page in the
+ same albumsection? If a new page is added, the next/prev links
+ may need to be updated, for example. If so, there will be much
+ unnecessary rebuilding. -J
+
+> albumsections are just a way to insert headings into the flow of
+> photos, so they don't actually affect dependencies.
+>
+> One non-obvious constraint of ikiwiki's current design is that
+> everything "off-page" necessary to build any page has to happen
+> at scan time, which has caused a few strange design decisions,
+> like the fact that each viewer controls what album it's in.
+>
+> It's difficult for the contents of the album to just be a
+> pagespec, like for inline, because pagespecs can depend on
+> metadata, which is gathered in arbitrary order at scan time;
+> so the earliest you can safely apply a pagespec to the wiki
+> contents to get a concrete list of pages is at rebuild time.
+>
+> (This stalled my attempt at a trail plugin, too.) -s
+
+>> Not sure I understand why these need to look at pagespecs at scan time?
+>> Also, note that it is fairly doable to detect if a pagespec uses such
+>> metadata. Er, I mean, I have a cheezy hack in `add_depends` now that does
+>> it to deal with a similar case. --[[Joey]]
+
+>>> I think I was misunderstanding how early you have to call `add_depends`?
+>>> The critical thing I missed was that if you're scanning a page, you're
+>>> going to rebuild it in a moment anyway, so it doesn't matter if you
+>>> have no idea what it depends on until the rebuild phase. -s
+
+* One thing I do like about having individual pages per image is
+ that they can each have their own comments, etc. -J
+
+> Yes; also, they can be wikilinked. I consider those to be
+> UI requirements. -s
+
+* Seems possibly backwards that the albumimage controls what album
+ an image appears in. Two use cases -- 1: I may want to make a locked
+ album, but then anyone who can write to any other page on the wiki can
+ add an image to it. 2: I may want an image to appear in more than one
+ album. Think tags. So it seems it would be better to have the album
+ directive control what pages it includes (a la inline). -J
+
+> I'm inclined to fix this by constraining images to be subpages of exactly
+> one album: if they're subpages of 2+ nested albums then they're only
+> considered to be in the deepest-nested one (i.e. longest URL), and if
+> they're not in any album then that's a usage error. This would
+> also make prev/next links sane. -s
+
+>> The current version constrains images to be in at most one album,
+>> choosing one arbitrarily (dependent on scan order) if albums are
+>> nested. -s
+
+> If you want to reference images from elsewhere in the wiki and display
+> them as if in an album, then you can use an ordinary inline with
+> the same template that the album would use, and I'll make sure the
+> templates are set up so this works. -s
+
+>> Still needs documenting, I've put it on the TODO list on the main
+>> page. -s
+
+> (Implementation detail: this means that an image X/Y/Z/W/V, where X and
+> Y are albums, Z does not exist and W exists but is not an album,
+> would have a content dependency on Y, a presence dependency on Z
+> and a content dependency on W.)
+>
+> Perhaps I should just restrict to having the album images be direct
+> subpages of the album, although that would mean breaking some URLs
+> on the existing website I'm doing all this work for... -s
+
+>> The current version of the branch doesn't have this restriction;
+>> perhaps it's a worthwhile simplification, or perhaps it's too
+>> restrictive? I fairly often use directory hierarchies like
+>> `a_festival/saturday/foo.jpg` within an album, which makes
+>> it very easy to write `albumsection` filters. -s
+
+* Putting a few of the above thoughts together, my ideal album system
+ seems to be one where I can just drop the images into a directory and
+ have them appear in the album index, as well as each generate their own wiki
+ page. Plus some way I can, later, edit metadata for captions,
+ etc. (Real pity we can't just put arbitrary metadata into the images
+ themselves.) This is almost pointing toward making the images first-class
+ wiki page sources. Hey, it worked for po! :) But the metadata and editing
+ problems probably don't really allow that. -J
+
+> Putting a JPEG in the web form is not an option from my point of
+> view :-) but perhaps there could just be a "web-editable" flag supplied
+> by plugins, and things could be changed to respect it.
+
+>> Replying to myself: would you accept patches to support
+>> `hook(type => 'htmlize', editable => 0, ...)` in editpage? This would
+>> essentially mean "this is an opaque binary: you can delete it
+>> or rename it, and it might have its own special editing UI, but you
+>> can never get it in a web form".
+>>
+>> On the other hand, that essentially means we need to reimplement
+>> editpage in order to edit the sidecar files that contain the metadata.
+>> Having already done one partial reimplementation of editpage (for
+>> comments) I'm in no hurry to do another.
+>>
+>> I suppose another possibility would be to register hook
+>> functions to be called by editpage when it loads and saves the
+>> file. In this case, the loading hook would be to discard
+>> the binary and use filter() instead, and the saving conversion
+>> would be to write the edited content into the metadata sidecar
+>> (creating it if necessary).
+>>
+>> I'd also need to make editpage (and also comments!) not allow the
+>> creation of a file of type albumjpg, albumgif etc., which is something
+>> I previously missed; and I'd need to make attachment able to
+>> upload-and-rename.
+>> -s
+
+>>> I believe the current branch meets your requirements, by having
+>>> first-class wiki pages spring into existence using `add_autofile`
+>>> to be viewer pages for photos. -s
+
+> In a way, what you really want for metadata is to have it in the album
+> page, so you can batch-edit the whole lot by editing one file (this
+> does mean that editing the album necessarily causes each of its viewers
+> to be rebuilt, but in practice that happens anyway). -s
+
+>> Replying to myself: in practice that *doesn't* happen anyway. Having
+>> the metadata in the album page is somewhat harmful because it means
+>> that changing the title of one image causes every viewer in the album
+>> to be rebuilt, whereas if you have a metadata file per image, only
+>> the album itself, plus the next and previous viewers, need
+>> rebuilding. So, I think a file per image is the way to go.
+>>
+>> Ideally we'd have some way to "batch-edit" the metadata of all
+>> images in an album at once, except that would make conflict
+>> resolution much more complicated to deal with; maybe just
+>> give up and scream about mid-air collisions in that case?
+>> (That's apparently good enough for Bugzilla, but not really
+>> for ikiwiki). -s
+
+>>> This is now in the main page's TODO list; if/when I implement this,
+>>> I intend to make it a specialized CGI interface. -s
+
+>> Yes, [all metadata in one file] would make some sense.. It also allows putting one image in
+>> two albums, with different caption etc. (Maybe for different audiences.)
+>> --[[Joey]]
+
+>>> Eek. No, that's not what I had in mind at all; the metadata ends up
+>>> in the "viewer" page, so it's necessarily the same for all albums. -s
+
+>> It would probably be possible to add a new dependency type, and thus
+>> make ikiwiki smart about noticing whether the metadata has actually
+>> changed, and only update those viewers where it has. But the dependency
+>> type stuff is still very new, and not plugin friendly .. so only just
+>> possible, --[[Joey]]
+
+----
+
+'''I think the "special extension" design is a dead-end, but here's what
+happened when I tried to work out how it would work. --[[smcv]]'''
+
+Suppose that each viewer is a JPEG-or-GIF-or-something, with extension
+".albumimage". We have a gallery "memes" with three images, badger,
+mushroom and snake.
+
+> An alternative might be to use ".album.jpg", and ".album.gif"
+> etc as the htmlize extensions. May need some fixes to ikiwiki to support
+> that. --[[Joey]]
+
+>> foo.albumjpg (etc.) for images, and foo._albummeta (with
+>> `keepextension => 1`) for sidecar metadata files, seems viable. -s
+
+Files in git repo:
+
+* index.mdwn
+* memes.mdwn
+* memes/badger.albumjpg (a renamed JPEG)
+* memes/badger/comment_1._comment
+* memes/badger/comment_2._comment
+* memes/mushroom.albumgif (a renamed GIF)
+* memes/mushroom._albummeta (sidecar file with metadata)
+* memes/snake.albummov (a renamed video)
+
+Files in web content:
+
+* index.html
+* memes/index.html
+* memes/96x96-badger.jpg (from img)
+* memes/96x96-mushroom.gif (from img)
+* memes/96x96-snake.jpg (from img, hacked up to use totem-video-thumbnailer :-) )
+* memes/badger/index.html (including comments)
+* memes/badger.jpg
+* memes/mushroom/index.html
+* memes/mushroom.gif
+* memes/snake/index.html
+* memes/snake.mov
+
+ispage("memes/badger") (etc.) must be true, to make the above rendering
+happen, so albumimage needs to be a "page" extension.
+
+To not confuse other plugins, album should probably have a filter() hook
+that turns .albumimage files into HTML? That'd probably be a reasonable
+way to get them rendered anyway.
+
+> I guess that is needed to avoid preprocess, scan, etc trying to process
+> the image, as well as eg, smiley trying to munge it in sanitize.
+> --[[Joey]]
+
+>> As long as nothing has a filter() hook that assumes it's already
+>> text... filters are run in arbitrary order. We seem to be OK so far
+>> though.
+>>
+>> If this is the route I take, I propose to have the result of filter()
+>> be the contents of the sidecar metadata file (empty string if none),
+>> with the `\[[!albumimage]]` directive (which no longer requires
+>> arguments) prepended if not already present. This would mean that
+>> meta directives in the metadata file would work as normal, and it
+>> would be possible to insert text both before and after the viewer
+>> if desired. The result of filter() would also be a sensible starting
+>> point for editing, and the result of editing could be diverted into
+>> the metadata file. -s
+
+do=edit&page=memes/badger needs to not put the JPG in a text box: somehow
+divert or override the normal edit CGI by telling it that .albumimage
+files are not editable in the usual way?
+
+> Something I missed here is that editpage also needs to be told that
+> creating new files of type albumjpg, albumgif etc. is not allowed
+> either! -s
+
+Every image needs to depend on, and link to, the next and previous images,
+which is a bit tricky. In previous thinking about this I'd been applying
+the overly strict constraint that the ordered sequence of pages in each
+album must be known at scan time. However, that's not *necessarily* needed:
+the album and each photo could collect an unordered superset of dependencies
+at scan time, and at rebuild time that could be refined to be the exact set,
+in order.
+
+> Why do you need to collect this info at scan time? You can determine it
+> at build time via `pagespec_match_list`, surely .. maybe with some
+> memoization to avoid each image in an album building the same list.
+> I sense that I may be missing a subtelty though. --[[Joey]]
+
+>> I think I was misunderstanding how early you have to call `add_depends`
+>> as mentioned above. -s
+
+Perhaps restricting to "the images in an album A must match A/*"
+would be useful; then the unordered superset could just be "A/*". Your
+"albums via tags" idea would be nice too though, particularly for feature
+parity with e.g. Facebook: "photos of Joey" -> "tags/joey and albumimage()"
+maybe?
+
+If images are allowed to be considered to be part of more than one album,
+then a pretty and usable UI becomes harder - "next/previous" expands into
+"next photo in holidays/2009/germany / next photo in tagged/smcv / ..."
+and it could get quite hard to navigate. Perhaps next/previous links could
+be displayed only for the closest ancestor (in URL space) that is an
+album, or something?
+
+> Ugh, yeah, that is a problem. Perhaps wanting to support that was just
+> too ambitious. --[[Joey]]
+
+>> I propose to restrict to having images be subpages of albums, as
+>> described above. -s
+
+Requiring renaming is awkward for non-technical Windows/Mac users, with both
+platforms' defaults being to hide extensions; however, this could be
+circumvented by adding some sort of hook in attachment to turn things into
+a .albumimage at upload time, and declaring that using git/svn/... without
+extensions visible is a "don't do that then" situation :-)
+
+> Or extend `pagetype` so it can do the necessary matching without
+> renaming. Maybe by allowing a subdirectory to be specified along
+> with an extension. (Or allow specifying a full pagespec,
+> but I hesitate to seriously suggest that.) --[[Joey]]
+
+>> I think that might be a terrifying idea for another day. If we can
+>> mutate the extension during the `attach` upload, that'd be enough;
+>> I don't think people who are skilled enough to use git/svn/...,
+>> but not skilled enough to tell Explorer to show file extensions,
+>> represent a major use case. -s
+
+Ideally attachment could also be configured to upload into a specified
+underlay, so that photos don't have to be in your source-code control
+(you might want that, but I don't!).
+
+> Replying to myself: perhaps best done as an orthogonal extension
+> to attach? -s
+
+> Yet another non-obvious thing this design would need to do is to find
+> some way to have each change to memes/badger._albummeta show up as a
+> change to memes/badger in `recentchanges`. -s
+
+Things that would be nice, and are probably possible:
+
+* make the "Edit page" link on viewers divert to album-specific CGI instead
+ of just failing or not appearing (probably possible via pagetemplate)
+
+* some way to deep-link to memes/badger.jpg with a wikilink, without knowing a
+ priori that it's secretly a JPEG (probably harder than it looks - you'd
+ have to make a directive for it and it's probably not worth it)