+ > It looks like 80% of the job would be accomplished by hooking `htmlize` for
+ > the `.xml` extension. That would satisfy the `pagetype` test that causes
+ > the edit CGI to say `not an editable page`. (That happens too early for a
+ > `canedit` hook.) The `htmlize` hook could just
+ > copy in to out unchanged (this is an internal wiki, I'm not thinking hard
+ > about evil XML content right now). For extra credit, an `editcontent` hook
+ > could validate the XML. (Can an `editcontent` hook signal a content error?)
+
+ > The tricky bit seems to be to register the fact that the target file should
+ > have extension `.xml` and not `.html`. Maybe what's needed is a generalized
+ > notion of an `htmlize` hook, one that specifies its output extension as well
+ > as its input, and isn't assumed to produce html? --ChapmanFlack 17July2008
+
+ > Belay that, there's nothing good about trying to use `htmlize` for this; too
+ > many html-specific assumptions follow. For now I'm back to an embarrassing quick
+ > hack that allows editing my xml file. But here's the larger generalization I
+ > think this is driving at:
+
+ > IkiWiki is currently a tool that can compile a wiki by doing two things:
+ > 1. Process files of various input types _foo_ into a single output type, html, by
+ > finding suitable _foo_->html plugins, applying various useful transformations
+ > along the way.
+ > 1. Process files of other input types by copying them with no useful transformations at all.
+
+ > What it could be: a tool that compiles a wiki by doing this:
+ > 1. Process files of various input types _foo_ into various output types _bar_, by
+ > finding suitable _foo_->_bar_ plugins, applying various useful transformations along
+ > the way, but only those that apply to the _foo_->_bar_ conversion.
+ > 1. The second case above is now just a special case of 1 where _foo_->_foo_ for any
+ > unknown _foo_ is just a copy, and no other transformations apply.
+
+ > In some ways this seems like an easy and natural generalization. `%renderedfiles`
+ > is already mostly there, keeping the actual names of rendered files without assuming
+ > an html extension. There isn't a mechanism yet to say which transformations for
+ > linkification, preprocessing, etc., apply to which in/out types, but it could be
+ > easily added without a flag day. Right now, they _all_ apply to any input type for
+ > which an `htmlize` hook exists, and _none_ otherwise. That rule could be retained
+ > with an optional hook parameter available to override it.
+
+ > The hard part is just that right now the assumption of html as the one destination
+ > type is in the code a lot. --ChapmanFlack
+