summaryrefslogtreecommitdiff
path: root/doc/bugs/pagetitle_function_does_not_respect_meta_titles.mdwn
blob: 4a83f9ec8ab8d1f77dda156842c436281e1ea85c (plain)

The IkiWiki::pagetitle function does not respect title changes via meta.title. It really should, so that links rendered with htmllink get the proper title in the link text.

--[[madduck]]


It is possible to set a Page-Title in the meta-plugin, but that one isn't reused in parentlinks. This [[patch]] may fix it.


A few quick notes about it:

  • Using inline would avoid the redefinition + code duplication.
  • A few plugins would need to be upgraded.
  • It may be necessary to adapt the testsuite in t/pagetitle.t, as well.

--[[intrigeri]]

It was actually more complicated than expected. A working prototype is now in my meta branch, see my userpage for the up-to-date url. Thus tagging [[patch]]. --[[intrigeri]]

Joey, please consider merging my meta branch. --[[intrigeri]]

So, looking at your meta branch: --[[Joey]]

  • Inter-page dependencies. If page A links to page B, and page B currently has no title, then A will display the link as "B". Now page B is modified and a title is added. Nothing updates "A". The added overhead of rebuilding every page that links to B when B is changed (as the postscan hook of the po plugin does) is IMHO a killer. That could be hundreds or thousands of pages, making interactive editing way slow. This is probably the main reason I had not attempted this whole thing myself. IMHO this calls for some kind of intellegent dependency handler that can detect when B's title has changed and only rebuild pages that link to B in that case.
  • Looks like some plugins that use pagetitle to format it for display were not changed to use nicepagetitle (for example, rename). But most of those callers intend to display the page name as a title, but including the parent directories in the path. (Ie, "renaming foo/page title to bar/page title" -- you want to know it's moved from foo to bar.) nicepagetitle does not allow doing that since it always takes the basename.
  • I don't like the name nicepagetitle. It's not very descriptive, is it? And it seems very confusing to choose whether to use the "nice" or original version. My hope is that adding a second function is unnecessary. As I understand it, you added a new function for two reasons:
    1. It needs the full page name, not basename.

    2. titlepage(pagetitle($page)) reversability.

    3. If you look at all the callers Of pagetitle most of them pass a complete page name, not just the basename. In most cases pagetitle is used to display the full name of the page, including any subdirectory it's in. So why not just make it consitently be given the full name of the page, with another argument specifying if we want to get back just the base name.

    4. I can't find any code that actually uses the reversability like that. The value passed to titlepage always comes from some external source. Unless I missed one.

  • The use of File::Spec->rel2abs is a bit scary.
  • Does it really make sense to call pagetitle on the meta title in meta's nicepagetitle? What if the meta title is something like "foo_bar" -- that would be changed to "foo bar".
  • parentlinks is changed to use nicepagetitle(bestlink($page, $path)). Won't bestlink return "" if the parent page in question does not exist?
  • backlinks() is changed to add an additional title field to the hash returned, but AFAICS this is not used in the template.
  • Shouldn't Render.pm use nicepagetitle when getting the title for the page template? Then meta would not need to override the title in the pagetemplate hook. (Although this would eliminate handling of title_overridden -- but that is little used and would not catch all the other ways titles can be overridden with this patch anyway.)

I'm not a reviewer or anything, but can I chime in on changes to pagetitle? I don't think having meta-titles in wikilinks and the parentlinks path by default is necessarily a good thing. I don't consider the meta-title of a page as used in <title> to be the same thing as the short title you want in those contexts - IMO, the meta-title is the "formal" title of the page, enough to identify it with no other context, and frequently too long to be used as a link title or a parentlink, whereas the parentlinks title in particular should be some abbreviated form that's enough to identify it in context. tbm expressed a similar opinion when I was discussing ikiwiki with him at the weekend.

It's a matter of taste whether wikilinks are "like a parentlink" or "like a <title>"; I could be persuaded either way on that one.

An example from my site: this page is the parent of this page with a title too long to use in the latter's parentlinks; I think the titles of both those pages are too long to use as wikilink text too. Similarly, tbm's page about Debian on Orion devices from Buffalo can simply be called "Buffalo" in context.

Having a \[[!meta abbrev="..."]] that took precedence over title in parentlinks and possibly wikilinks might be a good way to fix this? Or if your preference goes the other way, perhaps a \[[!meta longtitle=""]] could take precedence when generating the <title> and the title that comes after the parentlinks. --[[smcv]]

I think you've convinced me. (I had always had some doubt in my mind as to whether using titles in all these other places would make sense.)

Instead of meta abbrev, you could have a meta pagename that overrides the page name displayed everywhere (in turn overridden by meta title iff the page's title is being displayed). But is this complexity needed? We have meta redir, so if you want to change the name of a page, you can just rename it, and put in a stub redirection page so links still work.

This leaves the [[plugins/contrib/po]] plugin, which really does need a way to change the displayed page name everywhere, and at least a subset of the changes in the meta branch are needed to support that.

(This would also get around my concern about inter-page dependency handling, since po contains a workaround for that, and it's probably acceptable to use potentially slow methods to handle this case.) --[[Joey]]