conflux · v0.1.0-preview · evidence-backed pre-release

Linux-first C++ runtime and HTTP, scoped to attached evidence.

Conflux is a modules-first, Linux-only C++23-baseline runtime, JSON, and HTTP library built around io_uring. This preview is pre-v1: APIs may still change, and public support is scoped to the release SKUs and evidence lanes below.

C++23 baseline · C++26-gated experiments Linux hard requirement · requires liburing for runtime/http MODULE_INTERFACE primary preview mode f0f511b tagged source · evidence a3fa450
01 / preview scope

First public preview: narrow surface, attached evidence.

The release-facing contract is the documented SKU set, not every internal or experimental component in the tree.

release-json

Dependency-light JSON lane

core, json, and file_io_sync. The liburing-free generated-header install publishes only dependency-light components.

Evidence: full ASAN/UBSAN/TSAN closure, generated headers, package smoke.

release-http-api

HTTP API surface

core, json, and http. Real-liburing producer installs may publish runtime-facing package components such as work and http.

Use: find_package(conflux REQUIRED COMPONENTS http json).

release-web-server

Web server quickstarts

HTTP framework behavior, lifecycle, static/SSE/WebSocket examples, observability docs, and deployment-facing configuration.

Evidence: closure artifacts plus Docker conformance lane.

not advertised

PostgreSQL / pg

pg is intentionally outside this tag's advertised package component set because PostgreSQL evidence is not attached.

Rule: advertise pg only when DB evidence is attached and successful.

02 / examples

The shown examples are typed and macro-free.

These panels use quickstart code from examples/quickstart/; the JSON CRUD panel is excerpted.

Five concise files

The JSON CRUD panel is excerpted from the longer quickstart; the other panels show the full concise files.

examples/quickstart/hello.cxx16 lines · full file
import conflux;import std;int main() {	namespace http = conflux::http;	auto app = http::app();	app.get("/", [] { return http::text("hello from conflux\n"); });	app.get<"/hello/{name}">(		[](http::Path<"name"> name) { return http::html(std::format("<h1>Hello, {}!</h1>", name.get())); });	return http::exit_code(std::move(app).run({.port = 9090}));}
03 / public surface

Modules first; generated headers as release artifacts.

The primary source-consumption mode is MODULE_INTERFACE. Header mode exists for staged compatibility artifacts and is scoped to the components and toolchains covered by package-smoke evidence.

APIcurated app façadeimport conflux;, http::app()preview
HTTPserver/client building blocksconflux::http, typed path/body/state helpersliburing
JSONparser, DOM/view, serde boundaryconflux.json, native provider, JSONTestSuite laneliburing-free
FILEsync file and map surfacesfile_io_sync, file_mapliburing-free
RUNwork/runtime primitiveswork, uring, socket/file async layersadvanced
OSLinux + io_uringcapability probes, seccomp/container caveatsplatform
Install contract
Normal HTTP/JSON consumers request http and json; the package imports the required dependency closure without requiring users to request work directly.
Header artifacts
Generated headers are staged release artifacts from module sources. They are not hand-maintained source and not the design center for new API work.
Toolchain evidence
Preview support is limited to the attached lanes: Clang 21.1.8, GCC 15.2.1, GCC 16.1.0, CMake/Ninja from the evidence manifest.
Pre-v1 honesty
Public names may still change before 1.0 when a change removes compatibility clutter or fixes an incorrect contract.
04 / release evidence

Actual data for v0.1.0-preview.

The evidence repo records the source commit, proof-suite commit, host, compilers, raw benchmark logs, conformance reports, checksums, and rerun commands.

Benchmark policy
6 runs

Displayed benchmark rows use Conflux worst-of-6 against external best-of-6.

values: exact raw evidence · ranking: higher is better
HTTP evidence
4 workloads

Plaintext, small-JSON, route-param, and echo-POST wrk lanes at 256 connections.

scope: source-equivalent supporting measurement · compiler: clang 21.1.8
JSON evidence
3 compilers

Typed-struct deserialization lanes for Clang, GCC 15, and GCC 16 import-std.

commit: f0f511b · rounds: 6/compiler
h2spec 2.6.0
145 pass

HTTP/2 Docker conformance lane: 146 tests, 145 passed, 1 skipped, 0 failed.

skip: window-size-negative · failed: 0
Autobahn wstest
0 wrong

WebSocket Docker conformance lane: 301 OK/informational/non-strict, 216 optional unimplemented.

FAILED: 0 · WRONG: 0
Release closures
3 SKUs

release-json, release-http-api, and release-web-server closure artifacts were produced.

full suite sanitizer lanes · tests, benchmarks, examples
HTTP throughputrequests/second · Conflux worst-of-6, external best-of-6 · exact raw values
HTTP workloadConflux moduleConflux headeruWebSocketsBoost.BeastlibmicrohttpdlibeventScope
plaintext GET438,316.40req/s · -0.5%435,906.08req/s · -1.1%440,593.37req/s · best observed287,025.64req/s · -34.9%434,435.53req/s · -1.4%215,490.61req/s · -51.1%raw log · source-equivalent
small-JSON GET439,486.75req/s · best observed434,456.33req/s · -1.1%429,631.12req/s · -2.2%289,236.60req/s · -34.2%434,850.46req/s · -1.1%207,187.39req/s · -52.9%raw log · source-equivalent
route-param GET450,973.13req/s · -0.5%453,313.94req/s · best observed431,765.30req/s · -4.8%285,824.91req/s · -36.9%431,706.97req/s · -4.8%205,093.91req/s · -54.8%raw log · source-equivalent
echo-POST448,748.10req/s · -0.8%452,561.28req/s · best observed429,958.50req/s · -5.0%285,120.44req/s · -37.0%415,388.62req/s · -8.2%191,949.16req/s · -57.6%raw log · source-equivalent
JSON compiler viewchoose the compiler lane; all values are exact evidence values
JSON typed-struct throughputMB/s · clang 21.1.8 · LTO · Conflux worst-of-6, external best-of-6
JSON workloadConfluxGlazesimdjsonyyjsonBoost.JSONnlohmannScope
typed struct decode925.837MB/s · best observed860.298MB/s · -7.1%712.855MB/s · -23.0%724.754MB/s · -21.7%258.687MB/s · -72.1%79.926MB/s · -91.4%round logs · current tag
typed struct decode + reuse1,161.14MB/s · -13.0%1,334.84MB/s · best observed932.596MB/s · -30.1%968.903MB/s · -27.4%294.196MB/s · -78.0%80.642MB/s · -94.0%round logs · current tag
HTTP: exact requests/sec from raw wrk output wrk · 4 threads · 256 connections HTTP: Conflux worst-of-6 vs external best-of-6 JSON: exact MB/s from compiler round logs JSON: Conflux worst-of-6 vs external best-of-6 JSONTestSuite y/n cases Node + Python differential smoke h2spec 2.6.0 Autobahn wstest spectral 6.16.0

Claim scope

HTTP numbers are exact raw values selected from source-equivalent supporting measurements recorded under commit 6466c3cb8b638e87b849adb5433723fb39a86df7; the release/evidence README states the source delta to f0f511b is release/evidence/docs and proof-script-only. Treat them as supporting measurements unless an exact-commit HTTP rerun is attached.

JSON throughput values are exact raw values selected from the f0f511b compiler round logs; JSON conformance, SKU closures, and Docker conformance are attached for the f0f511b preview evidence set. These claims do not extend to HTTP/2 multiplexing, TLS, WebSocket throughput, static-file serving, large-body streaming, other hardware, other client tools, or production deployment configurations.

// evidence produced 2026-06-09T09:20:49Z · AMD Ryzen 7 5800X · Linux 6.18.31 · clang 21.1.8 · g++ 15.2.1 · g++-16 16.1.0

Claim audit ledgerevery public claim category points to an attached artifact or stays out of scope
Claim categoryAttached artifactPublic wording allowedExplicitly not claimed
HTTP/1.1 throughputfour raw wrk logssource-equivalent same-machine supporting measurementexact-release rerun, TLS, HTTP/2 performance, production deployment
JSON typed decode throughputcompiler round logscurrent-tag supporting measurement by compiler laneall JSON operations, arbitrary schemas, production latency guarantees
Protocol conformanceDocker conformance reportsh2spec, Autobahn, and spectral results for the attached lanenative-tool lane, optional WebSocket extensions, broader protocol performance
Release packagingrelease-json, release-http-api, release-web-serverthree attached SKU closuresunlisted components such as PostgreSQL
05 / measured comparisons

Scoped same-machine measurements.

The preview evidence compares Conflux against uWebSockets, Boost.Beast, libmicrohttpd, and libevent on four same-machine HTTP/1.1 workloads. Values are exact requests/second from raw evidence, selected as Conflux worst-of-6 versus external best-of-6.

HTTP claim
Same-machine HTTP/1.1 measurements

Plaintext, JSON, route-param, and echo-POST using wrk at 256 connections. The table marks the leading framework per workload and shows exact deltas. Conflux rows use the worst run out of six; external rows use the best run out of six.

JSON claim
Typed-struct throughput evidence

Six rounds per compiler across Clang 21.1.8, GCC 15.2.1, and GCC 16.1.0 with import std. The compiler tabs show exact MB/s values selected directly from the attached round logs.

Release posture
Pre-v1, evidence-scoped

APIs may still change before 1.0. PostgreSQL is not advertised for this tag because DB integration/pipeline evidence is not attached.

06 / evidence artifacts

Artifacts, commands, and checksums are published.

The evidence repository keeps bulky runtime and benchmark outputs out of the source tree while keeping raw artifacts, commands, checksums, and environment metadata attached.

Attached for this preview

  • SKU closures. release-json, release-http-api, and release-web-server with source archives, generated headers, package smoke, build costs, and capability reports.
  • Sanitizers. Full ASAN/UBSAN/TSAN lanes across tests, benchmarks, and examples.
  • JSON parser conformance. JSONTestSuite accept/reject cases plus Node/Python differential smoke.
  • Docker protocol conformance. h2spec, Autobahn WebSocket, and spectral reports attached under conformance-docker/.
  • Finalization scan. The final release evidence log scan reported no warning/error/sanitizer/failure markers.
07 / limits & next evidence

Attached scope, limits, and next evidence lanes.

Roadmap items stay out of release claims until raw evidence lands.

Attached now

attached
Three release SKUsrelease-json, release-http-api, release-web-server.
Quickstart examplesHello, JSON CRUD, middleware, static files, SSE, WebSocket.
JSON conformanceJSONTestSuite y/n cases plus differential smoke.
Docker conformanceh2spec, Autobahn wstest, spectral.
HTTP supporting measurementsFour wrk HTTP/1.1 workloads, six rounds per server.
source: f0f511b tag evidence, HTTP under 6466c3cb

Next evidence

missing
Exact-commit HTTP rerunNeeded before calling HTTP numbers exact-release benchmark data.
PostgreSQL lanepg becomes advertised only after DB evidence is attached and green.
Formal JSON benchmark summaryCurrent JSON throughput numbers are supporting measurements.
Broader workload coverageStatic files, large body streaming, WebSocket throughput, HTTP/2 multiplexing.
Lower toolchain floorsAdvertise lower CMake/compiler versions only after attached successful evidence lanes.
policy: support follows evidence

Experimental / gated

opt-in
HTTP/2 and HTTP/3 breadthh2spec is attached; multiplexing and wider protocol performance are not.
C++26 reflectionP2996 lane is compiler-gated and not the default first-contact path.
Standard SIMD targetsC++26-capable toolchain plus matching library feature probe required.
SEND_ZC tuningNon-loopback ZC-capable NIC evidence required for public throughput claims.
Complete aggregate surfaceUseful for local experimentation, not the default public entry point.
status: not advertised as first-contact preview defaults
08 / release links

Use the source tag and evidence together.

The source tag, proof harness, and evidence tree define this preview scope together.

conflux · v0.1.0-preview

Evidence-backed, pre-v1, Linux-only.

Start with the quickstarts, read the release notes, and treat benchmark numbers exactly as scoped: same-machine, attached-artifact evidence, with HTTP called supporting measurement until an exact-commit rerun lands.