There's been a lot of work on contrib syntax highlighting plugins. One should be picked and added to ikiwiki core.

We want to support both converting whole source files into wiki pages, as well as doing syntax highlighting as a preprocessor directive (which is either passed the text, or reads it from a file). But, the format directive makes this easy enough to do if the plugin only supports whole source files. So, syntax plugins do no really need their own preprocessor directive, unless it makes things easier for the user.

The big list of possibilities

  • highlightcode uses Syntax::Highlight::Engine::Kate, operates on whole source files only, has a few bugs (see here, and needs to be updated to support multiple pages with same name. (Currently a 404 :-( )
  • IkiWiki-Plugin-syntax only operates as a directive. Interestingly, it supports multiple highlighting backends, including Kate and Vim.
  • syntax only operates as a directive (not on source code files), and uses Text::VimColor.
  • sourcehighlight uses source-highlight, and operates on whole source files only. Needs to be updated to support multiple pages with same name.
  • sourcecode also uses source-highlight, and operates on whole source files. Updated to work with the fix for multiple pages with same name. Untested with files with no extension, e.g. Makefile.
  • jasonblevins's code plugin uses source-highlight, and supports both whole file and directive use.

  • hlsimple is a wrapper for the the perl module Syntax::Highlight::Engine::Simple. This is pure perl, pretty simple, uses css. It ought to be pretty fast (according to the author, and just because it is not external). On the other hand, there are not many predefined languages yet. Defining language syntaxes is about as much work as source-highlight, but in perl. I plan to package the base module for debian. Perhaps after the author releases the 5 or 6 language definitions he has running on his web site, it might be suitable for inclusion in ikiwiki. DavidBremner

  • highlight uses highlight via its swig bindings. It optionally supports whole files, but also integrates with the format directive to allow formatting of any of highlight's supported formats. (For whole files, it uses either keepextension or noextension, as appropriate for the type of file.)

General problems / requirements

  • Using non-perl syntax highlighting backends is slower. All things equal, I'd prefer either using a perl module, or a multiple-backend solution that can use a perl module as one option. (Or, if there's a great highlighter python module, we could use an external plugin..)

    Of course, some perl modules are also rather slow.. Kate, for example can only process about 33 lines of C code, or 14 lines of debian/changelog per second. That's 30 times slower than markdown!

    By comparison, source-highlight can do about 5000 lines of C code per second... And launching the program 100 times on an empty file takes about 5 seconds, which isn't bad. And, it has a C++ library, which it seems likely perl bindings could be written for, to eliminate even that overhead.

    highlight has similar features to source-highlight, and swig bindings that should make it trivial in principle to call from perl. I like highlight a bit better because it has a pass-through feature that I find very useful. My memory is unfortunately a bit fuzzy as to how well the swig bindings work. DavidBremner

  • Engines that already support a wide variety of file types are of course preferred. If the engine doesn't support a particular type of file, it could fall back to doing something simple like adding line numbers. (IkiWiki-Plugin-syntax does this.)

  • XHTML output.
  • Emitting html that uses CSS to control the display is preferred, since it allows for easy user customization. (Engine::Simple does this; Kate can be configured to do it; source-highlight can be made to do it via the switches --css /dev/null --no-doc)
  • Nothing seems to support wiki-formatted comments inside source files. Doing this probably means post-processing the results of the highlighting engine, to find places where it's highlighted comments, and then running them through the ikiwiki rendering pipeline. This seems fairly doable with Syntax::Highlight::Engine::Kate, at least.
  • The whole-file plugins tend to have a problem that things that look like wikilinks in the source code get munged into links by ikiwiki, which can have confusing results. Similar problem with preprocessor directives. One approach that's also been requested for eg, mediawiki is to allow controlling which linkification types a page type can have on it.

    The previous two points seem to be related. One thought: instead of getting the source from the content parameter, the plugin could re-load the page source. That would stop directives/links from being processed in the source. As noted above, comments could then be parsed for directives/links later.

    Would it be worth adding a nodirectives option when registering an htmlize hook that switches off directive and link processing before generating the html for a page?

  • The whole-file plugins all get confused if there is a foo.c and a foo.h. This is trivially fixable now by passing the keepextension option when registering the htmlize hooks, though. There's also a noextension option that should handle the case of source files with names that do not contain an extension (ie, "Makefile") -- in this case you just register the while filename in the htmlize hook.

  • Whole-file plugins register a bunch of htmlize hooks. The wacky thing about it is that, when creating a new page, you can then pick "c" or "h" or "pl" etc from the dropdown that normally has "Markdown" etc in it. Is this a bug, or a feature? Even if a feature, plugins with many extensions make the dropdown unusable..

    Perhaps the thing to do here is to use the new longname parameter to the format hook, to give them all names that will group together at or near the end of the list. Ie: "Syntax: perl", "Source code: c", etc.


I'm calling this done since I added the highlight plugin. There are some unresolved issues touched on here, but they either have the own other bug reports, or are documented as semi-features in the docs to the plugin. --Joey