Shared verification and benchmarks for Campfire and its Django, Laravel, Express, Elixir, Go, Rust, C and C++ implementations.
Every measured HTTP response must match its route contract: status, headers, complete decoded body, expected messages and content. Every acknowledged message write must match its exact persisted ID, body, room and search-index entry. Any invalid response or failed write audit fails the run. Browser flows check installation, live messages, editing, search, permissions, settings, invitations and session transfer.
See the current results and verification report.
Requires Ruby with Minitest, Rust 1.98.1, Node 22.18+, SQLite CLI, FFmpeg, Git and Docker. Keep this checkout alongside the implementation checkouts.
npm ci
npx playwright install chromium
bin/check
bin/seed
cargo build --release --locked --manifest-path loadgen/Cargo.toml
bin/benchmark --apps rails,elixir,go,rustBuild the implementations' production images first. Override their image names with RAILS_IMAGE, DJANGO_IMAGE, LARAVEL_IMAGE, EXPRESS_IMAGE, ELIXIR_IMAGE, GO_IMAGE, RUST_IMAGE, C_IMAGE and CPP_IMAGE. The C++ implementation is not in the default --apps list; select it with --apps rust,cpp. --help lists the benchmark options, including CPU affinity, seed path, route selection and output directory. The default is three alternating rounds with 16 concurrent clients. A process lock prevents overlapping benchmark runs. The fixture builder pins public Rails revision 90b3300 and generates real attachments and variants; it generates disposable signing, push and login credentials locally and refuses to overwrite an existing seed.
An optional cache-churn profile runs the same validated reads alongside one paced writer:
bin/benchmark --apps rails,elixir,go,rust --mixed-write-rate 10It posts at most ten messages per second to a separate room, without catch-up bursts.
Warmup and timed writes use the same response and persisted-write audits as the normal
POST benchmark. Reader and writer measurements stay separate in mixed-summary.json;
they do not enter the headline table. This tests cache invalidation under writes, not
sustained chat capacity. See the source architecture inventory.
Against a fresh, disposable running app:
bin/browser --base http://127.0.0.1:3000Allow several hundred megabytes of free space in /tmp for browser runs. Playwright's default --disable-dev-shm-usage makes Chromium use temporary files for shared memory; this harness sets TMPDIR under /tmp. Low space can cause asset loads to fail with ERR_INSUFFICIENT_RESOURCES even when RAM is available.
A dependency-free frontend regression can also check an implementation’s room-list controller:
node test/sidebar-reload.mjs ../once-campfire/app/javascript/controllers/rooms_list_controller.js
node test/editor-preservation.mjs ../once-campfire/app/javascript/controllers/rooms_list_controller.jsFor frontends that negotiate JSON autocomplete responses:
node test/autocomplete-json.mjs ../once-campfire-elixir/assets/overrides/lib/autocomplete/base_autocomplete_handler.jsThe browser flow creates and modifies accounts, rooms and messages. Run each implementation's own complete test suite as well: shared checks complement framework-specific tests and screenshot inventories.
The HTTP client sends Fetch Metadata for browser writes and includes legacy CSRF tokens only when an older implementation provides them.
loadgen cable --sessions FILE --refresh-secs 50 exercises separate session subscriptions and staggered presence refreshes. Each line contains a cookie followed by tab-separated Action Cable identifiers; clients cycle through the rows. Generate sessions for independently verified fixture users when testing distinct people. Results report assigned rows, distinct cookie strings, subscription count ranges, unread notices and refreshes sent; cookie counts alone do not establish user counts. Refreshes start after presence subscription confirmation and skip missed ticks. This profile requires the ordinary WebSocket client; --deflate does not support refresh commands.
Raw results, seeds and browser artifacts stay in ignored directories. Raw benchmark receipts are not committed. An HTTP throughput result does not measure concurrent users or WebSocket capacity.
Extracted from the Rust and Elixir verification work, with browser flows adapted from the Go implementation. The fixtures are built by the public Rails app. Original 37signals copyright and MIT license are retained.