Merge commit 'upstream/master' into pub/master

[git.ikiwiki.info.git] / doc / todo / Short_wikilinks.mdwn

diff --git a/doc/todo/Short_wikilinks.mdwn b/doc/todo/Short_wikilinks.mdwn
index a6734ac23e07478d627aede6519f603d8fcdd210..b9aec91123652a7cad007f49908b5cfaa76e7f56 100644 (file)
--- a/doc/todo/Short_wikilinks.mdwn
+++ b/doc/todo/Short_wikilinks.mdwn
@@ -1,16 +1,104 @@
-Markdown supports nice short links to external sites within body text by references defined elsewhere in the source:
+Markdown supports nice short links to external sites within body text by
+references defined elsewhere in the source:

 
         foo [bar][ref]
 
         [ref]: http://example.invalid/
 
 
         foo [bar][ref]
 
         [ref]: http://example.invalid/
 

-It would be nice to be able to do this or something like this for wikilinks as well, so that you can have long page names without the links cluttering the body text. I think the best way to do this would be to move wikilink resolving after HTML generation: parse the HTML with a proper HTML parser, and replace relative links with links to the proper files (plus something extra for missing pages).
+It would be nice to be able to do this or something like this for wikilinks
+as well, so that you can have long page names without the links cluttering
+the body text. I think the best way to do this would be to move wikilink
+resolving after HTML generation: parse the HTML with a proper HTML parser,
+and replace relative links with links to the proper files (plus something
+extra for missing pages).

 
 

-A related possibility would be to move a lot of "preprocessing" after HTML generation as well (thus avoiding some conflicts with the htmlifier), by using special tags for the preprocessor stuff. (The old preprocessor could simply replace links and directives with appropriate tags, that the htmlifier is supposed to let through as-is. Possibly the htmlifier plugin could configure the format.)
+> That's difficult to do and have resonable speed as well. Ikiwiki needs to
+> know all about all the links between pages before it can know what pages
+> it needs to build to it can update backlink lists, update links to point
+> to new/moved pages etc. Currently it accomplishes this by a first pass
+> that scans new and changed files, and quickly finds all the wikilinks
+> using a simple regexp. If it had to render the whole page before it was
+> able to scan for hrefs using a html parser, this would make it at least
+> twice as slow, or would require it to cache all the rendered pages in
+> memory to avoid re-rendering. I don't want ikiwiki to be slow or use
+> excessive amounts of memory. YMMV. --[[Joey]]
+
+>> Or you could disk cache the incomplete page containing only the body text,
+>> which should often not need re-rendering, as most alterations consist of
+>> changing the link targets exactly, and we can know pages that exist before
+>> rendering a single page. Then after backlinks have been resolved, it would
+>> suffice to feed this body text from the cache file to the template. However, e.g.
+>> the inline plugin would demand extra rendering after the depended-upon pages
+>> have been rendered, but these pages should usually not be that frequent, or
+>> contain that many other pages in full. (And for 'archive' pages we don't need
+>> to remember that much information from the semi-inlined pages.) It would help
+>> if you could get data structures instead of HTML text from the HTMLizer, and
+>> then simply cache these data structures in some quickly-loadeble form (that
+>> I suppose perl itself has support for). Regexp hacks are so ugly compared
+>> to actually parsing a properly-defined syntax...
+
+A related possibility would be to move a lot of "preprocessing" after HTML
+generation as well (thus avoiding some conflicts with the htmlifier), by
+using special tags for the preprocessor stuff. (The old preprocessor could
+simply replace links and directives with appropriate tags, that the
+htmlifier is supposed to let through as-is. Possibly the htmlifier plugin
+could configure the format.)
+
+> Or using postprocessing, though there are problems with that too and it
+> doesn't solve the link scanning issue.

 
 Other alternatives would be
 
   * to understand the source format, but this seems too much work with all the supported formats; or
 
 
 Other alternatives would be
 
   * to understand the source format, but this seems too much work with all the supported formats; or
 

-  * something like the shortcut plugin for external links, with additional support for specifying the link text, but the syntax would be much more cumbersome then.
+  * something like the shortcut plugin for external links, with additional
+    support for specifying the link text, but the syntax would be much more
+    cumbersome then.
+
+> I agree that a plugin would probably be more cumbersome, but it is very
+> doable. It might look something like this:
+
+       \[[!link bar]]
+
+       \[[!link bar=VeryLongPageName]]
+
+>> This is, however, still missing specifying the link text, and adding that option would seem to me to complicate the plugin syntax a lot, unless support is added for the |-syntax for specifying a particular parameter to every plugin.
+
+>>> Well, the link text in my example above is "bar". It's true that if
+>>> you want to use the same link text for multiple distinct links, or
+>>> different link texts for the same link, this is missing a useful layer of
+>>> indirection; it's optimised for the (probably) more common case. It
+>>> could be done as a degenerate form of the syntax I propose below, BTW.
+>>> --[[Joey]]
+
+>> ... Returning to this, the syntax infact wouldn't be so bad with the |-syntax, given a short name for the plugin:
+
+       [[whatever|ref 1]]
+       \[[!ref 1=page_with_long_name]]
+
+>>> A way to do this that doesn't need hacking at the preprocessor syntax
+>>> follows: --[[Joey]]
+
+       \[[!link bar=1]]
+       \[[!dest 1=page_with_long_name]]
+
+>>>> But this doesn't work so well for links that aren't valid keys. Such
+>>>> as stuff with spaces in it. I'd like to be able to write any kind of
+>>>> links conveniently, not just something that looks like a wikilink.
+
+>>>>> You're right, and to fix that it could be turned around: --[[Joey]]
+
+       \[[!link 1=bar]]
+       \[[!dest 1=page_with_long_name]]
+
+>> It also shouldn't be difficult to support non-wiki links in this same
+>> way, so that you could still link everywhere in an uniform manner, as
+>> the (still preferred by me) HTML processing approach would provide.
+>> Perhaps a plugin call wouldn't even be necessary for the links
+>> themselves: what about aliases for the normal link mechanism? Although
+>> the 'ref' call may infact be cleaner, and adding that |-syntax for
+>> plugins could offer other possibilities for other plugins.

 
 

+>>> I agree, it should be easy to make it support non-wiki links too.
+>>> We seem to have converged at something we can both live with that's
+>>> reasonable to implement.. --[[Joey]]