[docs] Add comprehensive architecture documentation #1062
Conversation
Adds ARCHITECTURE_README.md with detailed technical overview including: - Core component descriptions and relationships - System diagrams using Mermaid - Event flow and execution model documentation - Extension points with code examples - Performance considerations and best practices
- Remove execution model details (keeping only node ordering/sorting) - Remove WebGL references - Fix widget system details (add KnobWidget, correct type names) - Update node lifecycle to focus on callbacks instead of execution - Clarify that computeExecutionOrder is for ordering/layout, not execution
|
@webfiltered Anything look wrong? This doc can
|
|
First impression is that it needs a bit of work. If this is LLM output, can you please share the workflow / prompts / etc? Unless there is a good reason not to do so publicly, I think this should be done for all our generated documents. |
| ```mermaid | ||
| graph TD | ||
| LiteGraph[LiteGraph Global] --> LGraph[Graph Data Structure] | ||
| LGraph --> LGraphNode[Node System] | ||
| LGraph --> LLink[Link System] | ||
| LGraph --> LGraphGroup[Groups] | ||
| LGraph --> Reroute[Reroutes] | ||
| LGraphCanvas[Canvas Renderer] --> LGraph | ||
| LGraphNode --> Widgets[Widget System] | ||
| LGraphNode --> Slots[Slot System] | ||
| Subgraph --> LGraph | ||
| ``` |
There was a problem hiding this comment.
Can review this properly but this graph seems a bit weird to me; intentionally hiding the names of the components with things like "Node System". To me at least, it's confusing - Isn't this entire library a node system?
|
The prompt was essentially:
Ran with claude code launched from the root of this repo |
|
You can leave any further review comments if you get a chance and I can have it read and iterate on them, then check manually. |
|
This is the exact type of thing I'm trying to avoid. Unless I am missing something, it seems like it is pushing the burden of effort onto me, only with an inefficient, multi-step process. It is extremely valuable to know how much effort has already been spent reviewing, before jumping in to review myself. I assume you did more than a prompt and a cursory glance, but even if you had only done that - no problem, that's a good start, let me know and I can just take it from there. Reasoning is simply that you already have my trust; if you've checked something already I will skim for more obvious things, rather than analyse (e.g. reading at spoken-word speed, sometimes even pausing to consume from a few different viewpoints). |
|
I see what you're saying, will adjust going forward. |
Summary