Do not mark unchanged computed var deps as dirty

[benedikt-bartscher]

part of #4487

[masenf]

one potentially issue i see with this approach is it makes it harder to manually mark a var as dirty, for example if internal change tracking is not working for a certain kind of object, a hackaround was always to assign it to itself to signal to the framework that it should be included in the delta. i know @adhami3310 uses this trick quite a bit.

[benedikt-bartscher]

@masenf I knew this comment was coming 😆

However, I have 2 arguments against it:

• self.var = self.var is not really descriptive and needs a comment to explain what it's doing. self.dirty_vars.add("var"); self._mark_dirty() is pretty descriptive and doesn't need further explanation. We could add a helper which does both in one call.
• Currently, there is no way for users to stop reflex from marking all vars which depend on a computed var as dirty, even if the computed var did not change (_mark_dirty_computed_vars). Maybe I can adjust this PR to only change the computed var behavior?

[benedikt-bartscher]

I just refactored one small repo to always check if the var has changed before assignment. Now I have a diff of 85 lines with repetitive

if self.var != new_var:
  self.var = new_var

[adhami3310]

haven't looked into this too deeply but self.var = self.var is only used to let reflex know about a difference that internal mutability proxy didn't catch

[benedikt-bartscher]

current implementation is not optimal, i have a better one coming
BTW, this PR is now only about unchanged computed vars. I will split #4487 into multiple PRs

[benedikt-bartscher]

need to recheck order of computed var computation

[benedikt-bartscher]

need to recheck order of computed var computation

It changed due to priming of the caches. I disabled it for now in cc6edf1

This broke some of our apps without failing reflex tests. Maybe we can add a test for it (order and count of computed var execution)

Not sure why the last commit broke dynamic route var tests. My test project which I wrote while fixing var deps for dynamic args with @masenf still works fine. Maybe the test only needs adjustment. Need to investigate (I don't just want to change it without fully understanding it)

[benedikt-bartscher]

@masenf can you test if your model recursion issue also occurs with a computed var in this PR?

[masenf]

Thank you for contributing to Reflex. I am closing this old or stale PR for now, but please refresh and reopen it if you believe this issue is still relevant.

With the State refactoring, dirty var status will be stored in an "EventContext" rather than in the class itself. This should make it easier to handle checking whether a variable is actually dirty and should be recalculated. Since the underlying code is going through quite a bit of change, will be closing this one for now, but will incorporate the ideas to reduce delta size where possible.