>>> to work -- see <http://ikidev.betacantrips.com/blah.html> for an example.
>>> --Ethan
-[[tag patch]]
+## Alternative solution
+
+[Here](http://www.jk.fr.eu.org/ikiwiki/format-escapes-2.diff) is a patch
+largely inspired from the one below, which is up to date and written with
+[[todo/multiple_output_formats]] in mind. "htmlize" hooks are generalized
+to "convert" ones, which can be registered for any pair of filename
+extensions.
+
+Preprocessor directives are allowed to return the content to be inserted
+as a hash, in any format they want, provided they provide htmlize hooks for it.
+Pseudo filename extensions (such as `"_link"`) can also be introduced,
+which aren't used as real extensions but provide useful intermediate types.
+
+--[[JeremieKoenig]]
+
+> Wow, this is in many ways a beautiful patch. I did notice one problem,
+> if a link is converted to rst and then from there to a hyperlink, the
+> styling info usially added to such a link is lost. I wonder if it would
+> be better to lose _link stuff and just create link html that is fed into
+> the rst,html converter. Other advantage to doing that is that link
+> creation has a rather complex interface, with selflink, attrs, url, and
+> content parameters.
+>
+> --[[Joey]]
+
+>> Thanks for the compliment. I must confess that I'm not too familiar with
+>> rst. I am using this todo item somewhat as a pretext to get the conversion
+>> stuff in, which I need to implement some other stuff. As a result I was
+>> less careful with the rst plugin than with the rest of the patch.
+>> I just updated the patch to fix some other problems which I found with
+>> more testing, and document the current limitations.
+
+>> Rst cannot embed raw html in the middle of a paragraph, which is why
+>> "_link" was necessary. Rst links are themselves tricky and can't be made to
+>> work inside of words without knowledge about the context.
+>> Both problems could be fixed by inserting marks instead of the html/link,
+>> which would be replaced at a later stage (htmlize, format), somewhat
+>> similiar to the way the toc plugin works. When I get more time I will
+>> try to fix the remaining glitches this way.
+
+>> Also, I think it would be useful if ikiwiki had an option to export
+>> the preprocessed source. This way you can use docutils to convert your
+>> rst documents to other formats. Raw html would be loosed in such a
+>> process (both with directives and marks), which is another
+>> argument for `"_link"` and other intermediate forms. I think I can
+>> come up with a way for rst's convert_link to be used only for export
+>> purposes, though.
+
+>> --[[JeremieKoenig]]
+
+> Another problem with this approach is when there is some html (say a
+> table), that contains a wikilink. If the link is left up to the markup
+> lamguage to handle, it will never convert it to a link, since the table
+> will be processed as a chunk of raw html.
+> --[[Joey]]
+
+### Updated patch
+
+I've created an updated [patch](http://www.idletheme.org/code/patches/ikiwiki-format-escapes-rlk-2007-09-24.diff) against the current revision. No real functionality changes, except for a small test script, one minor bugfix (put a "join" around a scalar-context "map" in convert_link), and some wrangling to get it merged properly; I thought it might be helpful for anyone else who wants to work on the code.
+
+(With that out of the way, I think I'm going to take a stab at Jeremie's plan to use marks which would be replaced post-htmlization. I've also got an eye towards [[todo/multiple_output_formats]].)
+
+--Ryan Koppenhaver
+
+## Original patch
+[[!tag patch]]
<pre>
Index: debian/changelog
-* reStructuredText does not allow raw html to be inserted into
- documents, but ikiwiki does so in many cases, including
-- [[WikiLinks|WikiLink]] and many
-- [[PreprocessorDirectives|PreprocessorDirective]].
+- [[WikiLinks|ikiwiki/WikiLink]] and many
+- [[Directives|ikiwiki/Directive]].
+* Some bits of ikiwiki may still assume that markdown is used or embed html
+ in ways that break reStructuredText. (Report bugs if you find any.)
* It's slow; it forks a copy of python for each page. While there is a