ikiwiki/doc/todo/allow_site-wide_meta_definitions.mdwn

[[!tag plugins/meta patch]]
[[!template id=gitbranch branch=jon/defaultmeta author="[[Jon]]"]]

I'd like to define [[plugins/meta]] values to apply across all pages
site-wide unless the pages define their own: default values for meta
definitions essentially.

    <snip old patch, see below for latest>

-- [[Jon]]

> This doesn't support multiple-argument meta directives like
> `link=x rel=y`, or meta directives with special side-effects like
> `updated`.
>
> The first could be solved (if you care) by a syntax like this:
>
>     meta_defaults => [
>         { copyright => "© me" },
>         { link => "about:blank", rel => "silly", },
>     ]
>
> The second could perhaps be solved by invoking `meta::preprocess` from within
> `scan` (which might be a simplification anyway), although this is complicated
> by the fact that some (but not all!) meta headers are idempotent.
> 
> --[[smcv]]

>> Thanks for your comment. I've revised the patch to use the config syntax
>> you suggest. I need to perform some more testing to make sure I've
>> addressed the issues you highlight.
>> 
>> I had to patch part of IkiWiki core, the merge routine in Setup, because
>> the use of `possibly_foolish_untaint` was causing the hashrefs at the deep
>> end of the data structure to be converted into strings. The specific change
>> I've made may not be acceptable, though -- I'd appreciate someone providing
>> some feedback on that hunk!

>>> Well, re that hunk, taint checking is currently disabled, but
>>> if the perl bug that disallows it is fixed and it is turned back on,
>>> the hash values will remain tainted, which will probably lead to
>>> problems.
>>>
>>> I'm also leery of using such a complex data structure in config.
>>> The websetup plugin would be hard pressed to provide a UI for such a
>>> data structure. (It lacks even UI for a single hash ref yet, let alone
>>> a list.)
>>> 
>>> Also, it seems sorta wrong to have two so very different syntaxes to 
>>> represent the same meta data. A user without a lot of experience will
>>> be hard pressed to map from a directive to this in the setup file.
>>>
>>> All of which leads me to think the setup file could just contain
>>> a text that could hold meta directives. Which generalizes really to
>>> a text that contains any directives, and is, perhaps appended to the
>>> top of every page. Which nearly generalizes to the sidebar plugin,
>>> or perhaps something more general than that... 
>>>
>>> However, excessive generalization is the root of all evil, so 
>>> I'm not necessarily saying that's a good idea. Indeed, my memory
>>> concerns below invalidate this idea pretty well. --[[Joey]] 

    <snip old patch>

-- [[Jon]]

>> Ok, I've had a bit of a think about this. There are currently 15 supported
>> meta fields. Of these: title, licence, copyright, author, authorurl,
>> and robots might make sense to define globally and override on a per-page
>> basis.
>> 
>> Less so, description (due to its impact on map); openid (why would
>> someone want more than one URI to act as an openid endpoint to the same
>> place?); updated. I can almost see why someone might want to set a global
>> updated value. Almost.
>> 
>> Not useful are permalink, date, stylesheet (you already have a global
>> stylesheet), link, redir, and guid.
>>
>> In other words, the limitations of my first patch that [[smcv]] outlined
>> are only relevant to defined fields that you wouldn't want to specify a
>> global default for anyway.
>>
>>> I generally agree with this. It is *possible* that meta would have a new
>>> field added, that takes parameters and make sense to use globally.
>>> (Indeed, this later happened to some extent with the sortas parameters
>>> being added to some metas.)
>>> --[[Joey]] 
>>
>> Due to this, and the added complexity of the second patch (having to adjust
>> `IkiWiki/Setup.pm`), I think the first patch makes more sense. I've thus
>> reverted to it here.
>>
>> Is this merge-worthy?

    <snip old patch>

-- [[Jon]]

>>> Merry Christmas/festive season/happy new year folks. I've been away from
>>> ikiwiki for the break, and now I've returned to watching recentchanges.
>>> Hopefully I'll be back in the mix soon, too. In the meantime, Joey, have
>>> you had a chance to look at this yet? -- [[Jon]]

>>>> Ping :) Hi.  [[Joey]], would you consider this patch for the next
>>>> ikiwiki release? -- [[Jon]]

>>> For this to work with websetup and --dumpsetup, it needs to define the
>>> `meta_*` settings in the getsetup function.
>>>> 
>>>> I think this will be problematic with the current implementation of this
>>>> patch. The datatype here is an array of hash references, with each hash
>>>> having a variable (and arbitrary) number of key/value pairs.  I can't
>>>> think of an intuitive way of implementing a way of editing such a
>>>> datatype in the web interface, let alone registering the option in
>>>> getsetup.
>>>> 
>>>> Perhaps a limited set of defined meta values could be exposed via
>>>> websetup (the obvious ones: author, copyright, license, etc.) -- [[Jon]]
>>>
>>> I also have some concerns about both these patches, since both throw
>>> a lot of redundant data at meta, which then stores it in a very redundant
>>> way. Specifically, meta populates a per-page `%metaheaders` hash
>>> as well as storing per-page metadata in `%pagestate`. So, if you have
>>> a wiki with 10 thousand pages, and you add a 1k site-wide license text,
>>> that will bloat the memory usage of ikiwiki by in excess of 2
>>> megabytes. It will also cause ikiwiki to write a similar amount more data
>>> to its state file which has to be loaded back in each
>>> run.
>>>
>>> Seems that this could be managed much more efficiently by having
>>> meta special-case the site-wide settings, not store them in these
>>> per-page data structures, and just make them be used if no per-page
>>> metadata of the given type is present. --[[Joey]]
>>>> 
>>>> that should be easy enough to do. I will work on a patch. -- [[Jon]]
>>>>> Hi — I've written a new patch which I hope addresses the concerns raised
>>>>> with the previous ones. The new approach is to hard-code in `scan()`
>>>>> which of the meta types are supported in the setup file.  If one is
>>>>> defined, then `scan()` calls `preprocess()`, as [[smcv]] suggested,
>>>>> rather than stuffing redundant data into ikiwiki's data structures.
>>>>> 
>>>>> Two types supported in the setup file have optional arguments: `author`
>>>>> and `title`.  These are supported by having special-cased setup keys
>>>>> `meta_author_sortas` and `meta_title_sortas`.  Future expansion of the
>>>>> number of supported types, or addition of arguments to existing ones,
>>>>> can similarly be special-cased.
>>>>> 
>>>>> The setup data structure is no longer complicated with an
>>>>> array-of-hashes, which means this is suitable for exposing via websetup.
>>>>> `getsetup()` has been adjusted accordingly.
>>>>> 
>>>>> The patch can be found at the git branch described above.
>>>>>  — [[Jon]]

>>>>>> I wish I could take pity on you and just merge this, but 
>>>>>> AFAICS it still suffers from the memory bloat described above.
>>>>>> Specifically, when `scan` calls `preprocess`, it
>>>>>> stores the metadata in `%pagestate` etc. --[[Joey]]

>>>>>>> No pity required — but whoops, yes, that was a bit of a mistake
>>>>>>> ☺ I guess I'll have to split the current `preprocess` in half.
>>>>>>> — [[Jon]]

>>>>>>>> I've been taking another look at this today, as I'm very keen to
>>>>>>>> close various open loops of mine in IkiWiki to move on and do some
>>>>>>>> other stuff.  However, I'm not actually *using* this at the moment,
>>>>>>>> so whilst I think it's a good idea, I can't really motivate myself
>>>>>>>> to fix it anymore. I guess for now, this should just rot. — [[Jon]]