perf: serialize JSON-RPC responses once by Wodann · Pull Request #1698 · NomicFoundation/edr

Wodann · 2026-08-31T08:54:08Z

Follow-up to #1486, and stacked on it.

ResponseWithCallTraces.result was a serde_json::Value. Handlers built it with serde_json::to_value, then marshal_response_data walked the tree again with serde_json::to_string. Holding the serialized bytes instead makes the second pass a copy.

The performance gain is not significant but it is a prerequisite for further JSON-RPC round-trip optimisations.

What changes

result becomes a Box<serde_json::value::RawValue>, so each handler serializes once. Readers go through RawValue::get or the new deserialize_result.
marshal_response_data sizes its buffer from the result's length rather than growing from 128 bytes. That same length also decides the string limit up front, so an oversized response no longer serializes a 250 MB string only to discard it.
Batch responses are assembled from the already-serialized elements instead of being re-serialized through a Value::Array.
serde_json's raw_value feature is enabled workspace-wide, alongside the features already declared there.

Compatibility

The JSON reaching callers is byte-for-byte unchanged. A Box<RawValue> inside the untagged jsonrpc::ResponseData serializes exactly as the Value did, including through #[serde(flatten)], and arbitrary-precision numbers survive. marshal_response_data_matches_serde_for_success and its error counterpart pin both arms against serde's own output.

No #[napi] signature changes, so index.d.ts is untouched.

Benchmark Results

JS scenario runner benchmark (CI, vs latest main):

Scenario	main	#1486	#1698	#1486 / main	#1698 / main	#1698 / #1486
All Scenarios	210,629	153,020	148,833	0.73	0.71	0.97
`synthetix_9a3a109f`	156,751	111,860	108,024	0.71	0.69	0.97
`neptune-mutual-blue-protocol_8db6480`	21,999	17,240	16,885	0.78	0.77	0.98
`rocketpool_6a9dbfd8`	12,998	8,940	8,838	0.69	0.68	0.99
`openzeppelin-contracts_0a5fba7a`	9,110	7,250	7,337	0.80	0.81	1.01
`seaport_585b2ef8`	5,041	4,050	4,047	0.80	0.80	1.00
`uniswap-v3-core_d8b1c63`	4,068	3,190	3,205	0.78	0.79	1.00
`safe-contracts_914d0f8`	662	490	496	0.74	0.75	1.01

HH3 regression benchmark (CI, vs latest main):

Benchmark suite	#1698	#1486	Ratio
`ens-contracts / test vitest`	`3.9861710216199997` s (`± 0.09548535425823759`)	`4.06399684068` s (`± 0.012688302868408073`)	`0.98`
`ens-contracts / test vitest (peak RSS)`	`637` MB	`638` MB	`1.00`
`ens-contracts / test vitest (cpu)`	`81.23737028` s (`± 0`)	`80.07114505999999` s (`± 0`)	`1.01`
`lidofinance-core / test mocha`	`10.451533232740001` s (`± 0.11958584155861957`)	`10.36830099496` s (`± 0.020690816987286226`)	`1.01`
`lidofinance-core / test mocha (peak RSS)`	`795` MB	`804` MB	`0.99`
`lidofinance-core / test mocha (cpu)`	`379.8732149` s (`± 0`)	`378.19546216` s (`± 0`)	`1.00`
`openzeppelin-contracts / test mocha`	`57.92275289228` s (`± 0.019179380485141086`)	`57.84666742306` s (`± 0.010488213657848557`)	`1.00`
`openzeppelin-contracts / test mocha (peak RSS)`	`3531` MB	`3498` MB	`1.01`
`openzeppelin-contracts / test mocha (cpu)`	`62.59357925999999` s (`± 0`)	`62.39426504` s (`± 0`)	`1.00`

Validation

New unit tests for the envelope and for batch assembly.
Scenario replay is unchanged on every fixture.

Notes for review

raw_value is declared workspace-wide deliberately. Four crates name RawValue, and edr_rpc_client sits upstream of edr_provider, so it never inherits the feature.
The envelope is built through serde_json::to_writer into a pre-sized buffer rather than by hand. Concatenating {"result": measures the same but duplicates the wire format of a type in another crate.

changeset-bot · 2026-08-31T08:54:17Z

⚠️ No Changeset found

Latest commit: b6f0be3

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

This PR includes no changesets

When changesets are added to this PR, you'll see the packages that this PR includes changesets for and the associated semver types

Click here to learn what changesets are, and how to add one.

Click here if you're a maintainer who wants to add a changeset to this PR

Wodann · 2026-08-31T08:54:50Z

/bench scenarios="test mocha,test vitest"

Copilot

Pull request overview

Optimizes provider JSON-RPC serialization by storing serialized results as RawValue and reusing those bytes through batching and N-API marshaling.

Changes:

Replaces intermediate serde_json::Value results with Box<RawValue>.
Pre-sizes response envelopes and directly assembles serialized batches.
Updates consumers, tests, workspace features, and release notes.

Reviewed changes

Copilot reviewed 31 out of 31 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
`Cargo.toml`	Enables serde_json `raw_value`.
`.changeset/tidy-jars-serialize.md`	Records the serialization optimization.
`crates/edr_rpc_client/src/jsonrpc.rs`	Documents `RawValue` deserialization constraints.
`crates/edr_provider/src/lib.rs`	Stores raw results and adds typed deserialization.
`crates/edr_provider/src/requests/dispatch.rs`	Directly assembles serialized batch arrays.
`crates/edr_provider/src/test_utils.rs`	Adapts provider test helpers.
`crates/edr_provider/tests/common/provider.rs`	Adapts transaction helper deserialization.
`crates/edr_provider/tests/integration/timestamp.rs`	Uses typed result deserialization.
`crates/edr_provider/tests/integration/rip7212.rs`	Uses typed result deserialization.
`crates/edr_provider/tests/integration/issues/issue_407.rs`	Uses typed result deserialization.
`crates/edr_provider/tests/integration/issues/issue_356.rs`	Uses typed result deserialization.
`crates/edr_provider/tests/integration/issues/issue_325.rs`	Uses typed result deserialization.
`crates/edr_provider/tests/integration/issues/issue_324.rs`	Uses typed result deserialization.
`crates/edr_provider/tests/integration/interval_mining.rs`	Uses typed result deserialization.
`crates/edr_provider/tests/integration/eth_max_priority_fee_per_gas.rs`	Uses typed result deserialization.
`crates/edr_provider/tests/integration/eth_get_proof.rs`	Uses typed result deserialization.
`crates/edr_provider/tests/integration/estimate_gas.rs`	Uses typed result deserialization.
`crates/edr_provider/tests/integration/eip7843.rs`	Uses typed result deserialization.
`crates/edr_provider/tests/integration/eip7778.rs`	Uses typed result deserialization.
`crates/edr_provider/tests/integration/eip7702.rs`	Uses typed result deserialization.
`crates/edr_provider/tests/integration/eip7691.rs`	Uses typed result deserialization.
`crates/edr_provider/tests/integration/eip4844.rs`	Uses typed result deserialization.
`crates/edr_provider/tests/integration/eip2537.rs`	Uses typed result deserialization.
`crates/edr_provider/tests/integration/coverage.rs`	Uses typed result deserialization.
`crates/edr_provider/tests/integration/calldata_floor.rs`	Uses typed result deserialization.
`crates/edr_provider/tests/integration/block_timestamp_in_logs.rs`	Uses typed result deserialization.
`crates/edr_op/tests/integration/provider.rs`	Adapts OP provider tests.
`crates/edr_op/tests/integration/isthmus_operator_fee.rs`	Adapts OP fee tests.
`crates/edr_napi/src/mock.rs`	Stores mock responses as raw JSON.
`crates/edr_napi/src/context.rs`	Propagates mock serialization errors.
`crates/edr_napi_core/src/spec.rs`	Optimizes envelope serialization and sizing.

Suppressed comments (1)

crates/edr_napi_core/src/spec.rs:197

This function's buffer can grow for error responses because their capacity is fixed at 128 bytes while the serialized error may be larger. Limit the claim to the success path so the documentation matches the implementation.

/// Serializes a JSON-RPC response into a buffer that never has to grow.
///
/// `serde_json::to_string` starts at 128 bytes and doubles, so it copies a
/// large result several times over.

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

codecov · 2026-08-31T08:59:37Z

Codecov Report

❌ Patch coverage is 89.42308% with 22 lines in your changes missing coverage. Please review.
✅ Project coverage is 80.01%. Comparing base (a65c6b1) to head (b6f0be3).

Files with missing lines	Patch %	Lines
crates/edr_napi_core/src/spec.rs	93.60%	6 Missing and 2 partials ⚠️
crates/edr_napi/src/mock.rs	0.00%	4 Missing ⚠️
crates/edr_provider/src/requests/dispatch.rs	92.98%	0 Missing and 4 partials ⚠️
crates/edr_provider/src/lib.rs	70.00%	0 Missing and 3 partials ⚠️
crates/edr_provider/src/test_utils.rs	81.81%	0 Missing and 2 partials ⚠️
crates/edr_napi/src/context.rs	0.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1698      +/-   ##
==========================================
+ Coverage   79.97%   80.01%   +0.04%     
==========================================
  Files         462      462              
  Lines       80509    80685     +176     
  Branches    80509    80685     +176     
==========================================
+ Hits        64388    64564     +176     
+ Misses      13893    13891       -2     
- Partials     2228     2230       +2

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

github-actions · 2026-08-31T09:10:09Z

🚀 Starting regression benchmark for c76fd1d65ec2 against Hardhat main (projects matching test mocha,test vitest, benchmarks matching test solidity,test mocha,test vitest).

github-actions · 2026-08-31T18:30:04Z

❌ Regression benchmark failed for c76fd1d65ec2 against Hardhat main. This is either a detected performance regression or an infrastructure failure — see the run for details.

View workflow run

Wodann · 2026-08-31T19:40:03Z

/bench benchmarks="test mocha,test vitest"

github-actions · 2026-08-31T19:40:22Z

⏳ EDR CI for this commit hasn't passed yet, so the regression benchmark was not started. Comment /bench again once CI is green.

Wodann · 2026-08-31T21:03:02Z

/bench benchmarks="test mocha,test vitest"

github-actions · 2026-08-31T21:33:41Z

⏳ EDR CI for this commit hasn't passed yet, so the regression benchmark was not started. Comment /bench again once CI is green.

Wodann · 2026-09-01T06:36:28Z

/bench benchmarks="test mocha,test vitest"

github-actions · 2026-09-01T06:36:40Z

🚀 Starting regression benchmark for be36bade231c against Hardhat main (benchmarks matching test mocha,test vitest).

github-actions · 2026-09-01T06:49:47Z

✅ Regression benchmark passed for be36bade231c against Hardhat main.

View workflow run

`ResponseWithCallTraces.result` was a `serde_json::Value`, built by `to_json` and walked again by `marshal_response_data`. Holding the serialized bytes instead makes the second pass a copy, and lets a future response cache hand out clones for the price of one memcpy. The wire format is unchanged: a `Box<RawValue>` inside the untagged `jsonrpc::ResponseData` serializes byte-for-byte as the `Value` did, through `#[serde(flatten)]` as well. `result` and `marshal_response_data`'s parameter change type. Readers go through `RawValue::get` or the new `deserialize_result`. Enabling serde_json's `raw_value` feature is workspace-wide, as the crate's other features already are. `MockProvider` now converts its mocked response once, at construction, rather than on every request.

`serde_json::to_string` starts at 128 bytes and doubles, so it copies a large result several times over. A `RawValue` result makes the envelope's length known before it is built, so the buffer can be sized once. That same length also settles the string-limit question up front. An oversized response no longer serializes a discarded 250 MB string before falling back to a `serde_json::Value`. Both arms are pinned against serde's output by `marshal_response_data_matches_serde_for_success` and its error counterpart.

`ResponseData<Box<RawValue>>` compiles and always fails at runtime, and only on the `Success` variant, so a test covering the error path passes. `client::parse_response_str` is generic over `SuccessT`, so anyone forwarding a remote result without re-serializing it would hit this with an error message that points nowhere near the cause.

The batch tests reached every branch, but only through a provider fixture, and neither asserted the serialized text. Nothing covered a value containing a separator, which is what hand-building the array risks getting wrong. Also fixes the reservation for an empty array, which asked for one byte and needed two. `json_array_len` is now a named function so a test can assert the reservation equals the output length.

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

The limit becomes a parameter of a private helper so a test can reach it without allocating 250 MB. Both `Either::B` sites are now exercised: the oversized-success preflight and the post-serialization fallback. `envelope_len_matches_the_serialized_success` pins the hard-coded envelope overhead against serde, which the limit decision depends on. The precision test records that the fallback re-parses the raw value and that `arbitrary_precision` keeps the digits intact. Mutating the preflight to `>=`, the fallback check to `<`, or the overhead to 10 each fails at least one of these tests.

The name promised more than the test checks. `napi` panics marshalling a `serde_json::Value` number wider than `u64`, so the digits this test pins never reach JS through the `Value` arm. Verified against napi 3.12.1: `as_u64().unwrap()` at serde.rs:180 has no arm for an `arbitrary_precision` integer that fits neither `i64`, `u64` nor `f64`.

popescuoctavian

LGTM, thanks! Just a small nit

The three `map_err` arms in `marshal_response_data_with_limit` built the same `napi::Error` from a `serde_json::Error`.

Wodann requested a balanced review from Copilot August 31, 2026 08:54

Wodann self-assigned this Aug 31, 2026

Wodann had a problem deploying to github-action-benchmark August 31, 2026 08:54 — with GitHub Actions Error

Copilot started reviewing on behalf of Wodann August 31, 2026 08:54 View session

Copilot AI reviewed Aug 31, 2026

View reviewed changes

Comment thread crates/edr_rpc_client/src/jsonrpc.rs Outdated

Comment thread crates/edr_napi_core/src/spec.rs

Wodann temporarily deployed to github-action-benchmark August 31, 2026 09:31 — with GitHub Actions Inactive

Wodann temporarily deployed to github-action-benchmark August 31, 2026 09:48 — with GitHub Actions Inactive

Wodann had a problem deploying to github-action-benchmark August 31, 2026 09:48 — with GitHub Actions Error

Wodann temporarily deployed to github-action-benchmark August 31, 2026 11:42 — with GitHub Actions Inactive

Wodann had a problem deploying to github-action-benchmark August 31, 2026 12:21 — with GitHub Actions Failure

Wodann temporarily deployed to github-action-benchmark August 31, 2026 12:21 — with GitHub Actions Inactive

Wodann requested a review from a team August 31, 2026 13:28

Wodann temporarily deployed to github-action-benchmark August 31, 2026 19:48 — with GitHub Actions Inactive

Wodann force-pushed the perf/raw-value-response branch from 0cc3ee1 to 6509d71 Compare August 31, 2026 20:43

Wodann had a problem deploying to github-action-benchmark August 31, 2026 20:43 — with GitHub Actions Error

Wodann force-pushed the perf/raw-value-response branch from 6509d71 to e074e5c Compare August 31, 2026 21:14

Base automatically changed from fix/interval-mining-validation to main August 31, 2026 21:34

Wodann force-pushed the perf/raw-value-response branch from e074e5c to be36bad Compare August 31, 2026 21:34

Wodann temporarily deployed to github-action-benchmark August 31, 2026 21:34 — with GitHub Actions Inactive

Wodann temporarily deployed to github-action-benchmark August 31, 2026 22:39 — with GitHub Actions Inactive

Wodann added the no changeset needed This PR doesn't require a changeset label Sep 1, 2026

Wodann force-pushed the perf/raw-value-response branch from be36bad to d77f283 Compare September 1, 2026 09:31

Wodann had a problem deploying to github-action-benchmark September 1, 2026 09:31 — with GitHub Actions Error

Wodann and others added 7 commits September 1, 2026 11:33

misc: apply review suggestion

49dc549

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

Wodann force-pushed the perf/raw-value-response branch from d77f283 to c73070d Compare September 1, 2026 09:33

Wodann had a problem deploying to github-action-benchmark September 1, 2026 09:33 — with GitHub Actions Error

popescuoctavian approved these changes Sep 1, 2026

View reviewed changes

Comment thread crates/edr_napi_core/src/spec.rs Outdated

Comment thread crates/edr_napi_core/src/spec.rs

refactor(edr_napi_core): hoist the napi error conversion into a closure

b6f0be3

The three `map_err` arms in `marshal_response_data_with_limit` built the same `napi::Error` from a `serde_json::Error`.

Wodann temporarily deployed to github-action-benchmark September 1, 2026 10:06 — with GitHub Actions Inactive

Wodann temporarily deployed to github-action-benchmark September 1, 2026 10:16 — with GitHub Actions Inactive

Uh oh!

Conversation

Wodann commented Aug 31, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What changes

Compatibility

Benchmark Results

Validation

Notes for review

Uh oh!

changeset-bot Bot commented Aug 31, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

⚠️ No Changeset found

Uh oh!

Wodann commented Aug 31, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

codecov Bot commented Aug 31, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

github-actions Bot commented Aug 31, 2026

Uh oh!

github-actions Bot commented Aug 31, 2026

Uh oh!

Wodann commented Aug 31, 2026

Uh oh!

github-actions Bot commented Aug 31, 2026

Uh oh!

Wodann commented Aug 31, 2026

Uh oh!

github-actions Bot commented Aug 31, 2026

Uh oh!

Wodann commented Sep 1, 2026

Uh oh!

github-actions Bot commented Sep 1, 2026

Uh oh!

github-actions Bot commented Sep 1, 2026

Uh oh!

popescuoctavian left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Wodann commented Aug 31, 2026 •

edited

Loading

changeset-bot Bot commented Aug 31, 2026 •

edited

Loading

codecov Bot commented Aug 31, 2026 •

edited

Loading