It has been clear for a while that some kind of higher level documentation is needed for LV2. The reference documentation is much better than it used to be, but this isn't very good at giving an overview, and is very imposing to new developers since LV2 is a bit different than, say, a library API due to the code/data split.
Overview documentation can't really go in the extension documentation, since many tasks plugin developers want to achieve involve several extensions (or no extensions at all). I have been wondering for a while what form this documentation should take: FAQ? Wiki pages? Just examples that illustrate specific things? For whatever reason I had never thought of a book format, but now I think that is the best approach. Too many LV2 ideas live in my head alone, especially the umbrella ideas, which is not good. A book seems like the best way to convey an overall understanding of LV2. This is important - LV2 is what I call a "click" technology (like LISP or RDF), it seems insane until it "clicks" and you get it. You really need that little enlightenment experience to be able to work with it well (and especially to be able to work on it). I need to convey that to more people. Of course it would also serve as a simple "How To" manual for plugin developers, which is the primary goal.
By "book" I don't necessarily mean paper based. While that might be nice a web version definitely should exist. I mean a largish document with several chapters and sections, with a clear overall arc that starts with the most basic plugin possible, and each section is about a specific type of functionality, like "receiving MIDI events" or "transport synchronization". That way, when a developer has a broad question like "how do I implement state, and dynamically control it?" there is a nice book chapter on exactly that topic with explanatory text.
The other thing that is definitely good is examples. We still have only a few official examples in the LV2 tree, more would be nice. Making this book very example centric would be a good idea, solving several problems at once. Perhaps a Literate Programming like approach, where each chapter is really just the comments of a working plugin that demonstrates the relevant functionality. This way, even people who won't read the documentation (i.e. most) still have working examples to build from, and these examples would also serve as nice test cases for hosts.
The early sections could also serve as a more proper specification for things like discovery, module loading, and so on, which thus far we haven't documented so well, but honestly I don't see this as a very high priority. It's a nice idea to have an authoritative LV2 document, but in practice hosts should just use a library (like Lilv) for that stuff anyway, and most do.
In terms of publishing tech, I am not sure. I would like the examples to be actual C source code without duplicating everything, hence the Literate Programming idea. Literate Programming in philosophy only, not actually using a weird format like CWEB. I'd just parse out the comments and render those as text, and pretty-print the code in-between. This is simple, and with Pygments it can easily output both HTML and LaTeX for both an online and a typeset version. This way, for "just show me the code" types, the example source is just normal C code, but heavily documented with explanatory prose. As it happens I banged out a little Python script to do just that this evening. The tricky part is things like links, references, and so on, that require special syntax. I'd really rather not have to invent and implement a new comment syntax. However, perhaps something so directly tied to code can just avoid the need for any of that anyway, and the master document can just include all the generated bits. Might have to manually make a separate HTML and LaTeX master document but that's not too bad. This would all be trivial if I only use LaTeX, but since this is documentation, being able to link to a chapter online and such is important. Anyway, this is beside the point, the important thing is the general idea of writing an LV2 book.
I think this is a really good idea, and though I don't exactly have all the time in the world to start a book-level project, I might start building it in the LV2 repository so it is part of the official distribution. As the spec grows, the book can grow along with it. I envision future LV2 releases adding specific functionality (like transport sync, which will be the upcoming one), along with a new book chapter on the topic. On a personal note, this meshes well with desire to stick to non-realtime communication (e.g. email), since I can't have the constant distraction of IRC in my life any more. I've spewed enough text on to the Internet to fill volumes, might as well put that into something of lasting value.
To me it seems clear that this is the best way to better document LV2 and make it more accessible to newbies, but I'm notoriously bad at predicting what others want, so any feedback would be appreciated. If I wrote an LV2 book, would you read it, or at least use it as a reference?
Is it worth my time to do this, or should I just stick to hacking when I have time to do LAD stuff? If there's a good chance that this will increase the Bus Factor of LV2, I think it's probably worth it.