See comment: https://news.ycombinator.com/item?id=10095990. Thanks for taking the time to looking through the benchmark. I think you are misinterpreting things. Yes we do caching, and yes IF your app created a million tasks every second you would need to manually dereference them at some point. But you could change the caching (as mentioned in the comment) to never cache additional tags (instead of disabling all caching - which would be like going into the code of Mithril and React and remove all checks in the virtual dom).
I'm really glad to hear that it is still faster even when removing caching. That, to me, is quite insane. But if you wrote a TodoMVC app in Imba, you would do it _exactly_ like the one that is benchmarked. We have even considered adding asserts to warn users if they use uncached nodes in rendering that happens very frequently.
You are actually right that Imba does not dereference these automatically (see other comment re purging by judofyr: https://news.ycombinator.com/item?id=10092454). We could dereference unused tags once every second or so, but for now we think it is better to have manual control of this. Again, how much is usually happening during a render-cycle? Basically nothing. This benchmark is trying to look at how expensive it is to rerender the whole view. Maybe I should make the "Unchanged render" benchmark the main attraction (as it is the most important) -- but it is quite boring to look at.
I'm really glad to hear that it is still faster even when removing caching. That, to me, is quite insane. But if you wrote a TodoMVC app in Imba, you would do it _exactly_ like the one that is benchmarked. We have even considered adding asserts to warn users if they use uncached nodes in rendering that happens very frequently.
You are actually right that Imba does not dereference these automatically (see other comment re purging by judofyr: https://news.ycombinator.com/item?id=10092454). We could dereference unused tags once every second or so, but for now we think it is better to have manual control of this. Again, how much is usually happening during a render-cycle? Basically nothing. This benchmark is trying to look at how expensive it is to rerender the whole view. Maybe I should make the "Unchanged render" benchmark the main attraction (as it is the most important) -- but it is quite boring to look at.