add orjson

[benedikt-bartscher]

No description provided.

[codspeed-hq]

Merging this PR will not alter performance

✅ 32 untouched benchmarks
🆕 2 new benchmarks
⏩ 8 skipped benchmarks¹

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
🆕	test_process_event_wire[large_delta]	N/A	18 ms	N/A
🆕	test_process_event_wire[small_delta]	N/A	2.6 ms	N/A

---

_{Comparing benedikt-bartscher:try-orjson (47cd571) with main (7711b20)}

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. ↩

[greptile-apps]

Greptile Summary

This PR adds optional orjson support for faster JSON handling. The main changes are:

• New reflex[orjson] and reflex-base[orjson] extras.
• Shared JSON helpers for orjson dumps, loads, socket payloads, and non-finite floats.
• Updated socket, upload, compiler, template, and utility JSON call sites.
• Stdlib JSON parsing kept for inbound payloads that need exact large integers.
• Tests and news fragments for the new JSON behavior.

Confidence Score: 5/5

This looks safe to merge.

• No blocking issues found in the changed code.

Important Files Changed

Filename	Overview
reflex/app.py	Adds a per-server Socket.IO codec and keeps inbound event decoding on stdlib JSON.
packages/reflex-components-core/src/reflex_components_core/core/_upload.py	Keeps upload event args on stdlib JSON and uses the socket JSON helper for streamed updates.
packages/reflex-base/src/reflex_base/utils/serializers.py	Updates serializer registration behavior and parses Plotly JSON with the engine-aware helper.
packages/reflex-base/src/reflex_base/utils/format.py	Adds the shared orjson helpers and socket-specific handling for unsupported JSON values.
packages/reflex-base/src/reflex_base/.templates/web/utils/helpers/json.js	Adds frontend parsing for non-finite float sentinels and bare backend tokens.

_{Reviews (33): Last reviewed commit: "Merge branch 'main' into try-orjson" | Re-trigger Greptile}

[greptile-apps]

_{5 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[adhami3310]

the issue with orjson is that we lose NaN handling which is relevant for numpy, which is why we were using JSON5 as well

[benedikt-bartscher]

Yeah, that's why i did not update it in all places

[benedikt-bartscher]

related #6339

[cubic-dev-ai]

All reported issues were addressed across 35 files

<sub>Tip: cubic used a learning from your PR history. Let your coding agent read cubic learnings directly with the cubic MCP.

Re-trigger cubic</sub>

[cubic-dev-ai]

2 issues found across 39 files

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="packages/reflex-base/src/reflex_base/.templates/web/utils/state.js">

<violation number="1" location="packages/reflex-base/src/reflex_base/.templates/web/utils/state.js:588">
P2: When a valid socket payload contains a literal `__reflex_nan__` string, this call routes it through the reviver and changes the string to `NaN`. Parse valid JSON without a reviver first, and only use the reviver after detecting and rewriting non-finite tokens or escaped sentinels.</violation>
</file>

<file name="pyproject.toml">

<violation number="1" location="pyproject.toml:68">
P2: The `orjson` optional dependency is declared on `reflex`, but the module that actually imports orjson is `reflex-base` (`packages/reflex-base/src/reflex_base/utils/format.py` does `import orjson` in a try/except fallback). No `orjson` extra is declared in `packages/reflex-base/pyproject.toml`, so reflex-base's published metadata never advertises the optional dependency it owns at runtime. `reflex[orjson]` works because installing reflex pulls reflex-base into the same env, but a consumer who installs reflex-base directly (or who only wants the format helpers) has no supported opt-in and gets a silent stdlib fallback with no way to enable orjson through the package that uses it. Declare the `orjson` extra on reflex-base and have `reflex[orjson]` pull it in (e.g. `reflex-base[orjson]`), so the importing package's metadata is accurate.</violation>
</file>

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

Re-trigger cubic</sub>

[cubic-dev-ai]

P2: When a valid socket payload contains a literal __reflex_nan__ string, this call routes it through the reviver and changes the string to NaN. Parse valid JSON without a reviver first, and only use the reviver after detecting and rewriting non-finite tokens or escaped sentinels.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At packages/reflex-base/src/reflex_base/.templates/web/utils/state.js, line 588:

<comment>When a valid socket payload contains a literal `__reflex_nan__` string, this call routes it through the reviver and changes the string to `NaN`. Parse valid JSON without a reviver first, and only use the reviver after detecting and rewriting non-finite tokens or escaped sentinels.</comment>

<file context>
@@ -603,16 +585,10 @@ export const connect = async (
-      } catch (e2) {
-        return false;
-      }
+      return parseNonFiniteAwareJSON(str);
+    } catch {
+      // socket.io's decoder expects false for an undecodable packet.
</file context>

[benedikt-bartscher]

The premise does not hold, and the suggested order would introduce the bug it is guarding against.

Sentinels do not only arrive as bare tokens. The backend emits them as ordinary JSON strings, and escapes any user string that would collide. For {"user": "__reflex_nan__", "v": float("nan")} the wire format is:

{"user":"__reflex_esc____reflex_nan__","v":"__reflex_nan__"}

That is valid JSON. Parsing it without a reviver first succeeds, so the frontend would get the string "__reflex_nan__" instead of NaN, and __reflex_esc____reflex_nan__ would stay unstripped.

The scenario in the comment — an unescaped __reflex_nan__ in a valid payload — cannot occur: everything emitted on this socket goes through orjson_dumps_socket, which escapes collisions before they reach the wire (covered by test_sentinel_string_value_is_escaped and test_escape_prefixed_string_is_escaped_again). The reviver is only reached when the payload contains the prefix at all, which is what keeps the fast parser on the common path.

[cubic-dev-ai]

P2: The orjson optional dependency is declared on reflex, but the module that actually imports orjson is reflex-base (packages/reflex-base/src/reflex_base/utils/format.py does import orjson in a try/except fallback). No orjson extra is declared in packages/reflex-base/pyproject.toml, so reflex-base's published metadata never advertises the optional dependency it owns at runtime. reflex[orjson] works because installing reflex pulls reflex-base into the same env, but a consumer who installs reflex-base directly (or who only wants the format helpers) has no supported opt-in and gets a silent stdlib fallback with no way to enable orjson through the package that uses it. Declare the orjson extra on reflex-base and have reflex[orjson] pull it in (e.g. reflex-base[orjson]), so the importing package's metadata is accurate.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At pyproject.toml, line 68:

<comment>The `orjson` optional dependency is declared on `reflex`, but the module that actually imports orjson is `reflex-base` (`packages/reflex-base/src/reflex_base/utils/format.py` does `import orjson` in a try/except fallback). No `orjson` extra is declared in `packages/reflex-base/pyproject.toml`, so reflex-base's published metadata never advertises the optional dependency it owns at runtime. `reflex[orjson]` works because installing reflex pulls reflex-base into the same env, but a consumer who installs reflex-base directly (or who only wants the format helpers) has no supported opt-in and gets a silent stdlib fallback with no way to enable orjson through the package that uses it. Declare the `orjson` extra on reflex-base and have `reflex[orjson]` pull it in (e.g. `reflex-base[orjson]`), so the importing package's metadata is accurate.</comment>

<file context>
@@ -65,6 +65,7 @@ db = [
   "pydantic >=2.12.0,<3.0",
   "sqlmodel >=0.0.24,<0.1",
 ]
+orjson = ["orjson >=3.11.1,<4.0"]
 pydantic = ["reflex-base[pydantic]"]
 
</file context>

[benedikt-bartscher]

blocked by #6933

edit: merged/unblocked now