Skip to content

RELEASE.md

Latest commit

 

History

History
70 lines (57 loc) · 3.2 KB

RELEASE.md

File metadata and controls

70 lines (57 loc) · 3.2 KB
Raw

Research release checklist

This project may be released only as a research-use-only viewer for inputs that contain no patient data. It has no clinical, diagnostic, or de-identification claim.

Automated gates

Run from a clean checkout with the locked dependency graph:

cargo fmt --all -- --check
cargo clippy --workspace --all-targets --locked -- -D warnings
cargo test --workspace --all-targets --locked
cargo check --workspace --all-targets --locked --features cuda
cargo build --workspace --release --locked
cargo machete
cargo audit --deny unsound
cargo deny check advisories bans licenses sources

All three desktop CI targets must pass. The macOS Metal tests and a CUDA-host runtime validation must pass when those backends are included in a release.

Reproducible source gate

The locked graph must resolve wsi-rs 0.6.0 at revision b940ea94 and J2K 0.10.0 at revision 57b6af89 from their upstream Git repositories. CI and packaging must build without sibling codec checkouts or local-only source overrides. Run cargo metadata --locked --format-version 1 from a clean checkout and confirm that it leaves Cargo.lock unchanged.

Interactive performance gate

Do not infer end-to-end frame pacing from app_ui_cpu_ms or tile_probe. Before describing a build as real-time or interactively responsive:

  1. Record the release commit, Rust version, operating system, CPU, GPU, memory, display resolution and refresh rate, power mode, and SHA-256 checksums for every acceptance fixture.
  2. Predeclare numeric acceptance limits for end-to-end frame intervals, the longest allowed stall, interaction-to-first-sharp latency, full-target coverage latency, peak resident memory, and decode or upload errors.
  3. Run the optimized viewer with DICOM_VIEWER_DEBUG_STATS=1, retain its JSONL stderr, and externally capture presentation frame times. Exercise sustained pan and zoom plus repeated fit, level transitions, facts pagination, measurement, annotation, and GeoJSON replacement.
  4. Test representative SVS, DICOM VL WSI, and raw JPEG 2000 research inputs for at least 30 minutes in total. Include cold starts, warm revisits, malformed input, rapid direction reversals, and memory pressure.
  5. Report p50, p95, p99, and maximum observations alongside every predeclared limit. A missing format, backend, metric, or required runtime is a failed gate, not a pass.

Acceptance and packaging

  • Open representative SVS, DICOM VL WSI, and raw JPEG 2000 research fixtures; exercise fit, pan, zoom, facts pagination, measurement, annotation, and atomic GeoJSON replacement.
  • Exercise malformed metadata, oversized geometry, and decoder failures and confirm the viewer reports errors without terminating.
  • Build packages on their native operating systems. Sign and notarize the macOS application, Authenticode-sign the Windows package, and publish SHA-256 checksums and an SBOM with every artifact.
  • Install each packaged artifact on a clean machine and repeat the critical open, render, and GeoJSON export flow.

Do not describe a release as production-ready or real-time while the reproducible source, interactive performance, native packaging, signing, or representative acceptance gates remain open.