perf(compile): read only set props and cache literal Var dispatch

[FarhanAliRaza]

Summary

First of three stacked compile-performance PRs. This one is pure hot-path work with no change to generated output.

• Component._iter_set_props: _render, _get_vars, and the prop-component scan walked every declared prop (34 to 70 per Radix component) through the field descriptor to find the few that are set. They now iterate the instance dict plus class-level defaults, in declaration order. Literal defaults, @property overrides, and factory defaults still surface.
• _literal_var_for: caches the literal Var class per exact value type instead of copying and walking the twelve-entry isinstance ladder on every LiteralVar.create. Cleared when a literal subclass registers.
• insert_app_wraps short-circuits on identity before the field-by-field component comparison.
• CommonTag now holds the fields and generic render protocol shared by every tag class, with render(children=None) defined as dict(self.set(children=children)). Tag subclasses it and overrides render with a direct fast path that produces the same dictionary (tested). CondTag, MatchTag, IterTag, and Tagless inherit CommonTag, so Component._render and its overrides are annotated -> CommonTag. render_prop passes strings and dicts straight through.
• cached_property keys use a plain object() instead of uuid4.
• The memoize plugin's component imports are hoisted to module level (no import cycle exists).

Measurements

Docs site (docs/app, 511 routes), App._compile(dry_run=True) in a fresh process, warm runs:

state	compile
main	46.5 to 47.1 s
this PR	39.7 to 40.0 s

cProfile of the same compile: field-descriptor reads fell from 14.9M to 3.1M calls; _render -6.2 s, _get_vars -3.1 s, _create_literal_var -5.1 s, Tag.add_props -2.5 s (profiled time).

Benchmark pages (tests/benchmarks/fixtures.py, full-context compile): complicated page 29.3 ms to 25.0 ms.

Stack

1. this PR
2. perf(events): share one chain per handler and trigger across call sites #7122 perf(events): share one chain per handler and trigger across call sites
3. ENG-10101: perf(memo): evaluate passthrough bodies once and reuse their analysis #7123 perf(memo): evaluate passthrough bodies once and reuse their analysis

Each later PR's diff includes the earlier ones; merge in order.

Test plan

• New tests: set-prop iteration with defaults, plain-tag render protocol, custom-tag render protocol, render_prop passthrough, literal dispatch invalidation on late registration, cached_property identity, pickle, and release.
• uv run pytest tests/units/components tests/units/reflex_base tests/units/compiler tests/units/test_event.py green apart from failures that reproduce on clean main in this environment.
• ruff, pyright, and pre-commit clean.

https://claude.ai/code/session_01PmizE1eQhtYZyVs1RK2ke3

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-11T23:20:07.816636Z	870787a	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[cubic-dev-ai]

All reported issues were addressed across 8 files

<sub>Reply with feedback, questions, or to request a fix.

Re-trigger cubic</sub>

[greptile-apps]

Confidence Score: 5/5

The PR appears safe to merge; no actionable correctness, security, or repository-rule violations remain.

Summary

This PR optimizes several compilation hot paths while preserving the existing render protocol and generated output.

• Iterates only explicitly set or defaulted component props.
• Caches literal Var dispatch by exact Python value type and invalidates the cache when literal subclasses register.
• Introduces CommonTag and gives ordinary Tag rendering a direct fast path.
• Replaces UUID-based cached-property keys with identity objects and hoists memoization imports.
• Adds focused regression tests and changelog fragments for every affected package.

_{Reviews (7) · Last reviewed commit: "Merge branch 'main' into farhan/compile-..."}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 870787ab22

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[codspeed]

Merging this PR will improve performance by 13.16%

⚡ 9 improved benchmarks
✅ 31 untouched benchmarks
⏩ 8 skipped benchmarks¹

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
⚡	test_compile_page[_complicated_page]	275.8 ms	215.5 ms	+28.03%
⚡	test_compile_page_full_context[_complicated_page]	321.6 ms	255.9 ms	+25.66%
⚡	test_compile_all_artifacts[_complicated_page]	154.5 ms	135.8 ms	+13.77%
⚡	test_compile_page[_stateful_page]	69 ms	61.3 ms	+12.56%
⚡	test_compile_page_full_context[_stateful_page]	78.4 ms	69.9 ms	+12.26%
⚡	test_compile_all_artifacts[_stateful_page]	60.6 ms	55.8 ms	+8.61%
⚡	test_evaluate_page_with_hooks[_complicated_page]	62.9 ms	58.7 ms	+7.2%
⚡	test_evaluate_page[_complicated_page]	62.4 ms	58.2 ms	+7.17%
⚡	test_console_log	811.7 µs	770.1 µs	+5.41%

Tip

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

---

_{Comparing FarhanAliRaza:farhan/compile-prop-hot-paths (f3f0361) with main (2062139)}

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]

i'm still nervous overall about not having a consistent Tag interface.

I think what we need is to move most Tag functionality into a CommonTag base class that has

def render(self, children: Sequence[Any] | None = None) -> dict[str, Any]:
    if children is not None:
        return dict(self.set(children=children))

Have the specialty tags inherit from this. And have Tag inherit from CommonTag and overwrite the render(...) method with its more efficient implementation.

That way we don't have weird code duplication or ignored params, so CommonTag has a stable interface that is correct top to bottom, but the frequent case can still be performant.

The whole reason Tag and friends exist is to create another layer of indirection between the component and the rendered output. It's mostly an internal implementation detail, but I want to make sure that it remains consistent so we can still reason about it as a concept.