Only propagate newly-dirty vars in _mark_dirty_computed_vars

[Alek99]

Linear: ENG-10095

Description

Every setattr and every proxied append/__setitem__ triggered a full dirty-propagation pass: a needs_update scan over all computed vars (interval expiry) plus a dependency re-walk of the entire accumulated dirty_vars set, not just the newly dirtied var.

• Each class now precomputes which of its computed vars carry an update interval (_interval_computed_vars, refreshed whenever computed_vars changes), so the expiry scan is skipped entirely for the overwhelmingly common case of none.
• Each instance tracks a transient "propagated frontier": vars whose dependency closure has already been walked. Subsequent mutations only process newly-dirty vars.

Correctness: re-walking the dirty set is only observably needed after a cached computed var is recomputed mid-cycle (the recompute re-materializes a cache that a later mutation of its dependencies must invalidate again — including across states). Recomputes therefore bump a process-wide generation counter, and a generation mismatch resets the frontier, forcing the same full re-walk as today. The frontier is cleared by _clean and excluded from pickling, so restored/copied states always start with a full walk.

Benchmarks (GitHub Actions runner, run, 2 passes each; state with 2 cached computed vars)

Case	main	this PR	speedup
proxied list append	16.72 / 15.97 us	8.51 / 8.64 us	~1.9x
int var setattr	21.00 / 20.32 us	9.33 / 9.27 us	~2.2x

cProfile (2000 proxied appends): total function calls drop 216k -> 78k; _mark_dirty_computed_vars cumulative time drops 0.102s -> 0.019s and _expired_computed_vars disappears from the profile. The benchmark script also spot-checks that computed vars still recompute correctly after the mutation loop.

Type of change

• Bug fix (non-breaking change which fixes an issue)

Changes To Core Features:

• Have you added an explanation of what your changes do and why you'd like us to include them?
• Have you written new tests for your core changes, as applicable?
  • test_computed_var_recompute_after_mid_cycle_read: mutate -> read (recompute) -> mutate again -> the cache is invalidated again (the exact staleness hazard a naive frontier would introduce).
  • test_computed_var_recompute_after_mid_cycle_read_across_states: same hazard across parent/child states.
  • test_interval_computed_vars_precomputed: per-class interval var precomputation.
• Have you successfully ran tests with your changes locally?

Note: this touches the same ComputedVar.__get__ region as #6738; whichever merges second needs a trivial rebase.

🤖 Generated with Claude Code

https://claude.ai/code/session_01G8cXh3TUbNtbjm2ERqE62X

---

Generated by Claude Code

[linear-code]

ENG-10095

[greptile-apps]

Greptile Summary

This PR makes computed-var dirty propagation do less repeated work. The main changes are:

• Tracks already-propagated dirty vars during a mutation cycle.
• Resets that tracking when cached computed vars are recomputed.
• Precomputes interval-based computed vars per state class.
• Adds tests for mid-cycle recompute behavior and interval-var tracking.
• Adds release notes for the performance improvement.

Confidence Score: 5/5

This looks safe to merge.

• No blocking issues found in the changed code.

Important Files Changed

Filename	Overview
reflex/state.py	Adds the dirty-propagation frontier, interval computed-var cache, cleanup handling, and serialization handling.
packages/reflex-base/src/reflex_base/vars/base.py	Adds a shared recompute generation counter and bumps it when cached computed vars recompute.
packages/reflex-base/src/reflex_base/utils/types.py	Marks the new transient propagation fields as reserved backend var names.
tests/units/test_state.py	Adds tests for recomputation after mid-cycle reads, cross-state recomputation, and interval-var precomputation.
news/6743.performance.md	Adds a release note for the dirty-propagation performance change.
packages/reflex-base/news/6743.performance.md	Adds the package release note for the dirty-propagation performance change.

_{Reviews (4): Last reviewed commit: "Write propagation generation via __dict_..." | Re-trigger Greptile}

[codspeed-hq]

Merging this PR will improve performance by 7.01%

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 1 improved benchmark
✅ 25 untouched benchmarks
⏩ 8 skipped benchmarks¹

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
⚡	test_process_event	7.6 ms	7.1 ms	+7.01%

Tip

Curious why this is faster? Comment @codspeedbot explain why this is faster on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing claude/reflex-perf-optimizations-01l7a3-eng-10095 (0c44576) with main (32cc257)}

Footnotes

1. 8 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[masenf]

if we already have the instance_dict, why not just shove the value in there?

[masenf]

why are we treating these new fields like this and not defining them as normal fields in a BaseState, like all the other fields