Continuing mindspace UI work with performance optimizations and conversation logic improvements.
Don't do work twice!
When I was doing some performance tuning a few weeks ago I added code to defer the creation of various widget in conversation message sections, but also to defer syntax highlighting. A few days ago, I realized the syntax highlighting part was only expensive because I was doing it twice!
This hadn't always been true, but a few months ago I modified the markdown parser to make use of the syntax parsers for code blocks. This allows them to catch scenarios where markdown code fences (three backticks) appear in the wrong place within a code block. This would cause the markdown to mis-render, although this is an edge case that pretty-much every markdown parser I know of, except this one, has problems with.
While this worked fine, it did now mean that I ended up parsing each code block twice. Once to check it for markdown, and then a second time to syntax highlight it.
This was pretty wasteful, so now the markdown pass caches the syntax parsing and makes that available for syntax highlighting.
This change did also force me to separate out conversation and wiki syntax highlighting. The wiki code shouldn't have been relying on something in the conversation code.
Cleaning up conversation logic
I recently cleaned up the AI conversation transcript handling to make it more generic. While I was looking at how to handle trees of conversations, however, I realized it had a weird asymmetric interface and some very odd method names.
It doesn't now! The new interface has read and write of AIConversationHistory objects, and supports append of a single AIMessage.