Skip to content

BUILD_LOG.md

Latest commit

 

History

History
85 lines (76 loc) · 4.61 KB

BUILD_LOG.md

File metadata and controls

85 lines (76 loc) · 4.61 KB

NexView Build Log

Build 6: (uncommitted) - Multi-radar synced animation + composite GIF

  • Branch: perf-operational
  • Date: 2026-03-10
  • Status: ✅ Builds clean (ML + non-ML)
  • Changes:
    • RadarInstance expanded: anim_frames, anim_textures, anim_timestamps_ms, per-radar opacity
    • RadarApp: sync_timeline, sync_frame_map for unified multi-radar animation
    • load_multi_radar_animation(): parallel S3 download for all active radars
    • build_sync_timeline(): matches secondary frames to primary by closest timestamp (5-min window)
    • sync_secondary_to_frame(): swaps all secondary radar files/textures on frame advance
    • export_composite_gif(): renders all radars onto 1920x1080 canvas with alpha compositing
    • composite_radar_to_canvas(): geographic projection + alpha blending per radar
    • Timeline: multi-radar loading progress, radar count indicator, frame scrub syncs all
    • Sidebar: per-radar opacity sliders, group radar list with remove buttons, loading progress bars
    • Toolbar: "Group GIF" button label when multi-radar active
    • Level2File::unix_timestamp_ms() for timestamp matching

Build 5: (uncommitted) - Custom color table loading UI + GR2/RadarScope format

  • Branch: all
  • Date: 2026-03-10
  • Status: ✅ Builds clean (ML + non-ML)
  • Changes:
    • rfd file dialog added for native file picking
    • from_pal_string: full GRLevelX/RadarScope .pal parser (Color:, Color4:, SolidColor:, SolidColor4:, Scale:, Offset:, gradients, semicolons)
    • Sidebar: "Load Custom..." button with file dialog, auto-selects loaded table, "Delete" for custom
    • Supports .pal, .pal3, .wctpal, .csv, .txt extensions

Build 4: a8f0f63 - Custom color tables with LUT, presets, manager

  • Branch: all (ml-bugfixes, ml-prediction, perf-operational, master)
  • Date: 2026-03-10
  • Status: ✅ Builds clean (ML + non-ML)
  • Changes (cherry-picked from agent work on perf-feat-colortables):
    • color_table.rs: Complete rewrite with O(1) LUT-based lookup (4096 entries)
    • 6 built-in presets: NWS Default, GR2Analyst, NSSL, Classic, Dark, Colorblind
    • ColorTableManager with JSON persistence (%APPDATA%/nexview/colortables.json)
    • .pal and .csv custom color table import
    • Sidebar UI: preset selector dropdown, custom file loader, delete custom tables
    • panels.rs + sidebar.rs updated to use ColorTableManager

Build 3: f3972cb - Color table + PHI→KDP fixes

  • Branch: ml-bugfixes, ml-prediction, perf-operational, master
  • Date: 2026-03-10
  • Status: ✅ Builds clean (ML + non-ML)
  • Changes:
    • color_table.rs: O(log n) binary search for color lookup
    • convert.rs: reusable PHI scratch buffer, KDP edge backward difference
  • Build commands:
    • ML: cargo build --release --features tornado-predict
    • Perf: cargo build --release

Build 2: f19f8a2 - Parser safety + UX bugfixes

  • Branch: ml-bugfixes
  • Date: 2026-03-10
  • Status: ✅ Builds clean
  • Changes:
    • level2.rs: byte comparison for block type (avoids String alloc per radial)
    • level2.rs: scale != 0 guard prevents division by zero
    • app.rs: drain() instead of remove(0) for prediction history
    • app.rs: Shift+M for measure mode (M alone = mosaic, fixes key conflict)
    • app.rs: clear prediction state on station switch

Build 1: 332c85c - ML tornado prediction platform

  • Branch: master (original)
  • Date: 2026-03-09
  • Status: ✅ Builds clean
  • Features: Swin3D support, auto-infer, Ctrl+Click targeting, 8-frame buffer

Fork Architecture

  • ml-prediction: ML tornado prediction platform for model testing (Swin3D, ResNet3D, future models)
    • Build: cargo build --release --features tornado-predict
  • perf-operational: Performance-focused pure radar for operational meteorologists
    • Build: cargo build --release (no ML features)
  • Both forks share the same codebase; ML is behind tornado-predict feature flag

Known Issues (from audit, not yet fixed)

  • Level2File cloning (~25-50MB each) in animation/prediction paths - biggest perf win would be Arc/index-based approach
  • Smooth rendering algorithm not yet defaulted on
  • Multi-radar loading needs hardening

Agent work available but not yet integrated

  • ml-feat-uipolish (a257210): Model name overlay, risk gauge, inference time, Escape dismiss, "Clear All" button
  • perf-feat-optimize (dd9f938, 1ebc88b, 1adbc5f): Parallel smooth render (rayon par_iter + AtomicU32), smooth default on, #[cfg] guards on ML UI, LRU cache with 2GB budget, save-on-switch, station status tracking
  • ml-feat-multiradar (f43eb35): Per-station prediction buffers (HashMap), preserves prediction state across station switches, merged auto-infer results