Using pre-built lunr indexes
I've implemented pre-built indexes vs. on demand i.e. generating them when the search page is being loaded. I'm not entirely happy with the solution yet for the following reasons:
- Additional requirements - pre-built indexes require node.js to generate the indexes offline.
- Documents collection - the original search page still depends on the
reffield (the link to the relevant blog post) and title because the lunr index doesn't provide this. The
contentfield is now surplus to requirements once the site has been built, but it is needed during index generation.
- Added complexity - I know adding node.js isn't a big ask, but it has resulted in a new makefile. And until I update my travis config, pre-built indexes only get created when I've written a new article.
One novel approach to solving both these problems is to utilise a V8 crate and run lunr index natively. This is the approach adopted by the lunr indexer for middleman (a static site generate written in ruby).
Since we're not looking at native Cobalt solutions i.e. using liquid templates,
another approach would be to modify the node.js script used to pre-build the
index, to parse the content natively and remove the need for the
template. The script would also need to generate the documents collection
that's still used by the search page minus the content field.
The code changes can be found in the following commits:
- node.js script - used to pre-build the index, mostly cribbed from the lunr guide on that subject.
- makefile and search template - the makefile is cribbed from Matthias Endler's own version used to build his Cobalt site. The search template adds the prebuilt index to the list of assets to be loaded when the page is loaded.