Dedup auto-memoize passthrough definitions during compile by Alek99 · Pull Request #6735 · reflex-dev/reflex

Alek99 · 2026-07-10T19:31:43Z

All Submissions:

Have you followed the guidelines stated in CONTRIBUTING.md file?
Have you checked to ensure there aren't any other open Pull Requests for the desired changed?

Type of change

Performance improvement (non-breaking change, identical compile output)

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?

What this does

The auto-memoize plugin (MemoizeStatefulPlugin) built a full memo definition for every stateful call site and only deduped by (tag, source_module) afterwards. In profiling of a 500-duplicated-row page this was ~56% of compile CPU: 1,001 definitions were built and then collapsed to 3 unique ones. Per call site it:

ran _analyze_params twice on a constant signature — with from __future__ import annotations, get_type_hints eval-compiles the stringized annotations each time (2,472 builtins.compile calls in the profile), and
evaluated the memo body twice (once for the tag preview, once inside _create_component_definition).

Fix (behavior-preserving)

Hoist the params analysis to an import-time constant (_PASSTHROUGH_PARAMS): the passthrough wrapper signature is always (children: Var[Component]) -> Component.
Reuse the evaluated preview as the definition body (_LazyBody.ready(_lift_rest_props(preview))) instead of re-running the memo function.
Check CompileContext.auto_memo_components for the computed (tag, source_module) key first (new optional existing_definitions parameter): repeated identical subtrees reuse the first definition instead of rebuilding an equivalent one per call site. The cache stays per-CompileContext, so definitions are still never shared across compiles (covered by test_passthrough_memo_definitions_are_not_shared_globally).

Compile output is unchanged: the surviving definition is structurally identical to the one that previously overwrote the registry entry last.

Measured results (CodSpeed instrumentation, this PR vs `main` @ `9a5c4d3`)

Overall: +21.96% compile performance.

Benchmark	`BASE`	`HEAD`	Efficiency
`test_compile_all_artifacts[_stateful_page]`	27.2 ms	21.7 ms	+25.74%
`test_compile_page[_stateful_page]`	30.8 ms	25.3 ms	+21.94%
`test_compile_page_full_context[_stateful_page]`	35.8 ms	30.3 ms	+18.3%
`test_compile_page_duplicated_rows` (new)	N/A	828.1 ms	baseline for follow-ups

Verification

New unit tests: cache-hit returns the cached definition object (and a different source_module still misses); the memo body is evaluated exactly once per definition build.
New CodSpeed benchmark test_compile_page_duplicated_rows (200 identical stateful rows) stresses this path so follow-up compiler-perf PRs are measured on every run.

Part of a compiler-performance series; tracked in Linear as ENG-10101.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Jy8uHH11KircGa2MbTrR8g

greptile-apps · 2026-07-10T19:34:11Z

Greptile Summary

This PR improves auto-memoized component compilation and update emission performance. The main changes are:

Reuses duplicate passthrough memo definitions within a compile.
Computes passthrough memo tags without rebuilding the memo body.
Expands component hash coverage for dynamic imports.
Speeds up deterministic hash walking for common render data.
Replaces an awaited emit task wrapper with a direct await and loop yield.
Adds memo dedup tests and a duplicated-row compile benchmark.

Confidence Score: 5/5

This looks safe to merge.

No blocking issues found in the changed code.

Important Files Changed

Filename	Overview
packages/reflex-base/src/reflex_base/components/memo.py	Adds shared passthrough parameter analysis, precomputed passthrough tag generation, registry reuse, and single body evaluation for passthrough memo definitions.
packages/reflex-base/src/reflex_base/components/component.py	Optimizes deterministic hashing and includes dynamic imports in component hash inputs.
reflex/compiler/plugins/memoize.py	Moves passthrough memo definition reuse into the memo creation helper and runs event-trigger preparation only when a new body is built.
reflex/compiler/utils.py	Updates passthrough memo compilation to render the already-holed memo body directly.
reflex/app.py	Awaits websocket emits directly and keeps an explicit loop yield after queuing updates.
tests/units/compiler/test_memoize_plugin.py	Adds tests for memo registry reuse, source-module isolation, body evaluation count, event-trigger rewriting, and dynamic import tag coverage.
tests/benchmarks/test_compilation.py	Adds a duplicated-row compile benchmark for the auto-memoize dedup path.

_{Reviews (6): Last reviewed commit: "Update moment pyi hash: memo dedup chang..." | Re-trigger Greptile}

codspeed-hq · 2026-07-10T19:36:16Z

Merging this PR will improve performance by 27.63%

⚠️

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

⚡ 3 improved benchmarks
✅ 23 untouched benchmarks
🆕 1 new benchmark
⏩ 8 skipped benchmarks¹

Performance Changes

	Benchmark	`BASE`	`HEAD`	Efficiency
⚡	`test_compile_page[_stateful_page]`	30.2 ms	23 ms	+31.24%
⚡	`test_compile_page_full_context[_stateful_page]`	34.7 ms	27.5 ms	+26.24%
⚡	`test_compile_all_artifacts[_stateful_page]`	26.6 ms	21.2 ms	+25.47%
🆕	`test_compile_page_duplicated_rows`	N/A	647.2 ms	N/A

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-compiler-perf-t8ztc9-11-memoize-dedup (dea5c0c) with claude/reflex-perf-optimizations-21kg8y-emit-update (34610c2)}

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. ↩

linear-code · 2026-07-10T19:59:29Z

ENG-10101

The auto-memoize plugin built a full memo definition for every stateful call site and only deduped by tag afterwards. Per component this ran _analyze_params twice on a constant signature (get_type_hints eval-compiles the stringized annotations each time) and evaluated the memo body twice. - Hoist the passthrough params analysis to an import-time constant; the wrapper signature (children: Var[Component]) never changes. - Reuse the evaluated preview as the definition body instead of re-running the memo function inside _create_component_definition. - Check CompileContext.auto_memo_components for the computed (tag, source_module) key before building, so repeated identical subtrees reuse the first definition instead of rebuilding an equivalent one per call site. Adds a duplicated-rows compile benchmark that stresses this path. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Jy8uHH11KircGa2MbTrR8g

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Jy8uHH11KircGa2MbTrR8g

create_passthrough_component_memo gained the existing_definitions parameter, changing the generated stub. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Jy8uHH11KircGa2MbTrR8g

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Jy8uHH11KircGa2MbTrR8g

…build Hash the call-site component with a constant {children}-hole render to derive the memo tag before any body work happens. On a registry hit the plugin now skips the event-trigger rewrite and the memo body build entirely; on a miss the new prepare_body hook applies the rewrite and the definition self-registers. Soundness of dedup-by-tag requires the component hash to cover every input the memo module emission reads, so dynamic imports are now hashed (with a regression test). The hash walk also gets exact-type fast paths for str/dict/list and a per-class dataclass field-name cache.

Alek99 requested a review from a team as a code owner July 10, 2026 19:31

Alek99 mentioned this pull request Jul 10, 2026

Cache Var.to()/guess_type() registry lookups #6742

Merged

5 tasks

FarhanAliRaza previously approved these changes Jul 10, 2026

View reviewed changes

claude and others added 7 commits July 17, 2026 18:18

Add news fragments for auto-memoize dedup

df953ab

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Jy8uHH11KircGa2MbTrR8g

Update pyi hash for reflex/experimental/memo

a20c564

create_passthrough_component_memo gained the existing_definitions parameter, changing the generated stub. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Jy8uHH11KircGa2MbTrR8g

Annotate existing_definitions dict key type for pyright invariance

1b9e87e

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Jy8uHH11KircGa2MbTrR8g

Regenerate pyi hash for reflex/experimental/memo after rebase

c8ece1f

Update moment pyi hash: memo dedup changes Moment (MemoizationLeaf) pyi

dea5c0c

Alek99 dismissed FarhanAliRaza’s stale review via dea5c0c July 18, 2026 01:45

Alek99 force-pushed the claude/reflex-compiler-perf-t8ztc9-11-memoize-dedup branch from 0390d78 to dea5c0c Compare July 18, 2026 01:45

Alek99 changed the base branch from main to claude/reflex-perf-optimizations-21kg8y-emit-update July 18, 2026 01:46

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Dedup auto-memoize passthrough definitions during compile#6735

Dedup auto-memoize passthrough definitions during compile#6735
Alek99 wants to merge 7 commits into
claude/reflex-perf-optimizations-21kg8y-emit-updatefrom
claude/reflex-compiler-perf-t8ztc9-11-memoize-dedup

Alek99 commented Jul 10, 2026 •

edited

Loading

Uh oh!

greptile-apps Bot commented Jul 10, 2026 •

edited

Loading

Uh oh!

codspeed-hq Bot commented Jul 10, 2026 •

edited

Loading

Uh oh!

linear-code Bot commented Jul 10, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

Alek99 commented Jul 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

All Submissions:

Type of change

Changes To Core Features:

What this does

Fix (behavior-preserving)

Measured results (CodSpeed instrumentation, this PR vs main @ 9a5c4d3)

Verification

Uh oh!

greptile-apps Bot commented Jul 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Greptile Summary

Confidence Score: 5/5

Important Files Changed

Uh oh!

codspeed-hq Bot commented Jul 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Merging this PR will improve performance by 27.63%

Performance Changes

Footnotes

Uh oh!

linear-code Bot commented Jul 10, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Alek99 commented Jul 10, 2026 •

edited

Loading

Measured results (CodSpeed instrumentation, this PR vs `main` @ `9a5c4d3`)

greptile-apps Bot commented Jul 10, 2026 •

edited

Loading

codspeed-hq Bot commented Jul 10, 2026 •

edited

Loading