6.1 KiB
6.1 KiB
QFT v0.4.16 implementation status
Windows → Android optical
- QFT2 manifest/data/end frame format
- QFT2 Base45 text layer; Android Result.text decode
- QFT1 legacy receive compatibility
- CRC32 per frame + SHA-256 whole file
- single large QR only; 2×2 / 4-channel removed
- default 450-byte DATA payload
- QR-L + complete 4-module quiet zone
- Windows target display 5~60 FPS
- draggable control/QR split pane
- focus-QR mode and QR zoom
- Rayon background pre-encoding
- reused egui GPU TextureHandle / NEAREST sampling
- positioned source reads; no whole-file mmap
- bounded pre-render cache; 1100 MiB process design budget
- Android centered acquisition + finder-point ROI lock
- smoothed moving ROI + progressive miss expansion/reacquire
- dynamic overlay: search/green lock/yellow miss/outside dim
- lock-centered periodic AF/AE
- Android 640×480 ImageAnalysis + KEEP_ONLY_LATEST
- Camera2Interop advertised AE FPS range request up to 60
- automatic fallback when an OEM rejects explicit FPS/stream combination
- reusable Y-plane
GrayFrameBuffer - GlobalHistogram hot path + sparse Hybrid fallback
- duplicate decoded optical frames suppressed before assembler
- runtime camera FPS / decoded/s / distinct-new-frame/s / decode-ms UI
- synthetic QFT2 v15 full-frame + ROI visual decode test
- physical 30/45/60Hz display × Android camera throughput matrix
Android → Windows
- QFTPAIR2 compact pairing QR
- persistent per-PC Direct AP identity + per-session token
- Windows Wi-Fi Direct Legacy AP implementation
- Android 10+ WifiNetworkSpecifier path
- Android 7~9 legacy Wi-Fi path
- wait for actual IPv4/link route before discovery
- UDP QFTDISC2 receiver discovery
- QR endpoint + gateway + directed broadcast + subnet fallback
- dedicated Wi-Fi immediate parallel gateway + actual-subnet probe
- HTTP bound to selected Android Network
- local cleartext network security config
- local/private IPv4 + session-token validation
- Parallel v2 init/chunk/finish
- adaptive 64 KiB–16 MiB chunks, targeting eight flows per file
- up to 8 Android workers per file
- Windows positioned concurrent writes
- temporary
.qftpart, final SHA-256 and rename - accepted sockets return to timeout-bound blocking I/O before file transfer
- inline Android file selection + removable, pre-transfer sortable queue
- crash-safe drag commit without detaching the active touched view
- loopback UDP + HTTP + out-of-order parallel integration test
- physical Windows Wi-Fi Direct adapter / driver / firewall matrix
Windows → Android local direct
- direction-specific pairing QR starts automatic local download after scan
- HTTP
Rangeresponses return the exact requested range (206+Content-Range) - Android parallel Range downloader targets eight workers per file
- adaptive 64 KiB–16 MiB chunks
.qftpartcompletion ledger resumes only verified completed ranges- five retries per range, final SHA-256, then save to
Downloads/QFT - regression test for delayed large download after nonblocking accept
Android compatibility
- minSdk 24 (Android 7.0)
- WRITE_EXTERNAL_STORAGE <= Android 9
- MediaStore Downloads >= Android 10
- ACCESS_FINE_LOCATION compatibility <= Android 12
- NEARBY_WIFI_DEVICES >= Android 13
- no
java.util.Base64dependency in app source - host Kotlin QFT2/Base45 + LockRect runtime smoke
- host Kotlin QrAnalyzer type-check with CameraX/ZXing-shaped stubs
- real Android Gradle APK build in this container (SDK unavailable here)
- install/runtime matrix on physical Android 7→latest devices
iOS (v0.5.0)
Optical receive (PC screen → iPhone camera)
- Swift QFT2 + Base45 + CRC32 frame decode, QFT1 legacy read
- session assembler: dedupe, positioned writes, whole-file SHA-256, discard on mismatch
- AVCaptureVideoDataOutput + Vision
VNDetectBarcodesRequest - format picked for highest resolution among ≥60 FPS formats; fixed frame rate, HDR/stabilisation off
alwaysDiscardsLateVideoFrames(no stale camera frames)- centered search → ROI lock → progressive expansion → reacquire
- periodic AF/AE at ROI center after lock
- separate decoded/s and distinct-new-frame/s counters
- physical throughput matrix (30/45/60 FPS × device)
iPhone ↔ Windows dedicated network
- accept Wi-Fi-only QFTPAIR2/QFTSEND2 (no endpoints, by design)
NEHotspotConfigurationjoin withjoinOnce- honest fallback to guided manual join when the entitlement is unavailable
- default gateway from
sysctlroute table; gateway-first host candidates - join verified by authenticated
/qft/ping, not by the system callback - physical Windows Wi-Fi Direct AP join (both entitlement paths)
iPhone ↔ Mac / Linux LAN
QFTBROWSE3node browse client- desktop connect code (Crockford Base32 + CRC-8), cross-verified against Rust vectors
POST /qft/paircode → token, no retry on failure- eight-way parallel Range download, 64 KiB–16 MiB adaptive chunks
.qftpart+ completion ledger resume (parallel chunks land out of order)- explicit failure + single-stream fallback when the peer ignores Range
- parallel upload (init/chunk/finish), fixed 4 MiB chunks × 6 workers
iPhone ↔ iPhone (AirDrop-style)
- AWDL
includePeerToPeer+ TLS-PSK, key carried by the QR - multi-file queue (QFTP2P2); manifest digest in the QR, manifest itself on the wire
- per-file SHA-256, partial files removed on any failure
- physical AWDL run with the router switched off
Compatibility
- deployment target iOS 15.0 (iPhone 6s and later)
- iOS 16/17 APIs behind
App/Compatibility.swiftforks - no
loadUnaligned(Swift 5.7 runtime) in the core layer - 180 core-layer checks green via
swiftcon macOS, no Xcode required - Xcode build of the App layer (no Xcode on the authoring machine)
See TEST_REPORT.md and TEST_RESULTS.txt for executed evidence and limitations.