Question about re-evaluation strategy

This looks like a very nice library, congratulations!

I was wondering, suppose a component reads a.b.c from tracked state a. Suppose a new state a is reduced, and a.b also “changes”, but a.b.c remains the same. I guess in this case the component will not re-render? In other words, to determine if a component should re-render, you always compare the leaf nodes of the dependency tracking graph?

Is this reasoning correct?

[ziriax]

Thanks for your feedback! I’ve been thinking about it, and I my gut feeling says your approach is valid.

[dai-shi]

New ideas/insights are generally welcome. 😉

But I guess you can’t know this.

Yeah.

One possible misunderstanding you and maybe others might have is that it doesn’t stop tracking for child components. This is very unique to this library (more precisely the family of my libraries: This, [1] and [2]). Technically, we can’t know when render function ends. React doesn’t provide such a feature. So, I rather keep tracking forever… This means it will keep tracking even with the inside of <Foo x={abc}/> or <Bar x={abc} y={ab}/>. In other words, most likely the leaf nodes are primitives like strings and numbers. The exception is probably memoization and maybe some others. It does best to work with memoization but there must be pitfalls. I’d love to get feedback from everyone.

The current approach and implementation should be pretty good. I spent quite a long time to reach this point. I can say it’s almost optimal, meaning a good trade-off balance between the usability and the performance.

If there’s a practical use case this approach doesn’t work, I will reconsider and maybe tweak it (or gave up such a use case). So again, any example to break this tracking approach is welcome, preferably practical ones instead of hypothetical ones.