ICCery/AGENTS.md

ICCery Agent Notes

Stage 5 Verification / Profcheck

• profcheck output is parsed from both JSON summaries (preferred) and legacy plain-text report formats.
• If no delta-E values can be detected, the report cards show 0.00 and a warning is appended to the process log.
• The .gam file for the 3D viewer is parsed using parseGamutFile, which supports multiple BEGIN_DATA blocks (some Argyll files use a separate block per surface section), inline # comments, and out-of-bounds vertex warnings.
• Manual parser tests live in src/js/gamut_viewer.test.js.

3D Gamut Viewer

• The viewer renders the measured/derived .gam volume and an optional sRGB reference wireframe in CIELAB.
• Layer controls (profile, sRGB, axes) each have visibility toggles and opacity sliders.
• Click Reset View or press R to return the camera to its default position.
• Full JSDoc is provided on the public API in src/js/gamut_viewer.js.

Stage 4 colprof Options

• colprof options are exposed in the Stage 4 UI with contextual tooltips:
  • FWA / OBA Compensation (-f): choose D50, None, D65, or a custom .sp spectrum file.
  • Standard Illuminant (-i) and Observer (-o): override default D50 / 1931 2° for CIE colourimetric calculations.
  • Viewing Conditions (-c input, -d output): set CIE viewing-condition transforms for the profile table and output intent.
• The backend build_colprof_args in commands.rs maps these UI values to colprof CLI flags. Unit tests cover all combinations.

Stage 3 Swatch Grid

• Swatch patches render a 135° diagonal split: top-left is the intended/expected colour, bottom-right is the measured colour.
• Argyll chartread emits is_pad for boundary/spacer patches. White reference patches (e.g. -e white steps) may also carry is_pad, but they have valid expected.Lab or non-zero device data. The guard skips only pads with no measurement and all-zero device values.
• Row/patch order from chartread (rows A→Z, patches 1→N within each row) is rendered left-to-right / top-to-bottom to match the printtarg output.
• Tooltip shows intended Lab (or device %), measured Lab, and ΔE₀₀ with a Good/Warning/Bad classification.

UI Button Conventions

The frontend uses a tiered button sizing system defined in src/styles/main.css. Prefer these utility classes over inline style attributes.

Class	Size	Use for
.btn-sm	28px	Toolbar actions, header icons, minor toggles
.btn-md (default for button.secondary/button.danger)	36px	Standard dialog/form actions and browse buttons
.btn-lg	40px	Primary stage actions (Generate, Create Layout, Create Profile, Verify, etc.)
.btn-icon-sq	36×36px	Square icon-only buttons (refresh, settings, etc.)
.icon-btn	28×28px	Small header icon-only buttons (settings, about, save preset, manage presets)
.btn-properties	36px	Printer driver preferences button

• Primary action buttons use button.primary plus .btn-lg for major stage actions.
• Danger actions use button.danger (36px).
• All action rows use one of: .stage-actions, .modal-actions, .chartread-actions, .print-actions-row, .btn-row, .btn-row-sm, .btn-row-end, or .input-row-sm.
• Avoid inline style on <button> elements or their immediate parent rows.

Settings & Preferences

• Settings are persisted to settings.json in the app data directory and include the Stage 3 ΔE₀₀ traffic-light thresholds.
• Valid threshold values must be non-negative and delta_e_good_max < delta_e_warning_max; both the frontend and backend enforce this.
• Saving settings dispatches a settings-saved custom event so live components (e.g. the swatch grid) can re-classify on the fly.

Build Commands

• Rust backend: cd src-tauri && CARGO_INCREMENTAL=0 cargo check (the project lives on a network filesystem that doesn't support file locking, so CARGO_INCREMENTAL=0 is required)
• Rust tests: cd src-tauri && CARGO_INCREMENTAL=0 cargo test
• Frontend: cd src-tauri && npm run build (or npm run dev for development)

Architecture Overview

• Backend: Rust + Tauri v2 (src-tauri/)
• Frontend: Vanilla JS modules (src/js/), HTML (src/index.html)
• Print subsystem: Platform-specific code under src-tauri/src/print/
  • mod.rs: Shared types (PrintOptions, Printer, PrinterCapabilities, PrinterDevModeStore)
  • macos.rs: macOS-specific lp spooling and native NSPrintPanel integration
  • unix.rs: Generic Unix/CUPS utilities (printer enumeration, PPD parsing, lp args)
  • windows.rs: Windows-specific printing via Win32 API and DEVMODE

Cross-Platform Printer Field Notes

When adding fields to Printer in src-tauri/src/print/mod.rs, update every platform-specific constructor in src-tauri/src/print/windows.rs, src-tauri/src/print/macos.rs, and src-tauri/src/print/unix.rs to avoid build regressions on any target. Use ..Default::default() where possible, or explicitly provide values (e.g. display_name: None on Windows).

macOS Print Properties (Issue #188)

The "Preferences" button opens the native macOS NSPrintPanel (not CUPS web UI or System Settings).

• The CUPS destination ID is bound to the panel via Core Printing PMPrinterCreateFromPrinterID and PMSessionSetCurrentPMPrinter
• A Printer.display_name (from CUPS printer-info) is cached at enumeration as a fallback for NSPrinter::printerWithName
• Pre-configured with both AP_ColorMatchingMode=AP_ApplicationColorMatching and AP.ColorMatchingMode=AP_ApplicationColorMatching (dot-notation) as a locked PMPrintSettings value and in the NSPrintInfo job ticket
• Uses the private Core Printing PMSessionSetColorMatchingMode / PMSessionSetColorMatchingModeLock / PMSessionSetColorMatchingModeNoLock SPI (resolved at runtime via dlsym) to gray out and lock the Color Matching controls; all three symbols use the 2-argument (PMPrintSession, *const CFString) signature; PMSessionSetColorMatchingModeLock sets and locks in one call; NoLock sets the mode without locking; falls back to the public PMPrintSettingsSetValue setting if the SPI is absent
• Pre-selects the driver-specific "no color adjustment" PPD option (Canon CNIJIntent2=4, Epson EPIJ_CMat=3, etc.) in the native panel and on the lp command line
• Captures user's media type / quality selections as a CUPS options string with PMPrintSettingsToOptions
• Returns a PrintPropertiesResult with the effective selected_printer and captured PrintOptions
• Cancellation is returned as None, not an error
• Captured options are stored in frontend capturedCupsOptions map and passed via PrintOptions.cups_options
• build_lp_args in macos.rs always adds both -o AP_ColorMatchingMode=AP_ApplicationColorMatching and -o AP.ColorMatchingMode=AP_ApplicationColorMatching, and forwards captured options
• Only AP_ApplicationColorMatching and ApplicationColorMatching are passed to the private SPI; AP_ColorSyncMatching and AP_VendorColorMatching are intentionally avoided because they would enable color management on profiling targets

Key Dependencies (macOS only)

• objc2 0.6 — MainThreadMarker, rc
• objc2-app-kit 0.3.2 — NSPrintPanel, NSPrintInfo, NSPrinter
• objc2-foundation 0.3.2 — NSString
• objc2-core-foundation 0.3.2 — CFString, CFType
• objc2-application-services 0.3.2 — PMCore (PMPrintSettings, PMPrinter, PMSession, etc.)

PPD Option Detection

• Epson media type key: EPIJ_Medi (in addition to CNIJMediaType, MediaType, StpMediaType)
• Epson color bypass: EPIJ_CMat=3 (Off / No Color Adjustment)
• Canon color bypass: CNIJIntent2=4 or CNIJIntent=4
• Gutenprint: StpColorCorrection=Uncorrected

Verification History & Printer Drift Tracking (#95)

• Historical verification runs are stored in verification_history.json in the app data directory.
• Record schema (VerificationRecord in src-tauri/src/quality_store.rs):
  • id: unique record identifier in the format vr-<epoch_millis>-<seq>.
  • profile_name: target profile filename.
  • printer: device name captured at print spooling (wizardState.printerName), or "Unknown".
  • avg_de, max_de, rms_de: CIEDE2000 metrics from profcheck (using -u JSON summary).
  • patch_count: number of test patches evaluated.
  • status: classified status using ICCery verification bands (issue #95):
    • < 1.0: "Excellent" (badge-excellent)
    • < 2.0: "Good" (badge-good)
    • < 3.5: "Acceptable" (badge-acceptable)
    • >= 3.5: "Warning" (badge-poor)
  • timestamp: ISO-8601 UTC string.
• Max capacity is 500 records; oldest records evicted on overflow.
• Atomic file writes (.tmp write followed by rename) prevent data corruption.
• Tauri IPC command casing:
  • Nested struct fields (VerificationRecord) serialize with snake_case.
  • Top-level Tauri command arguments use camelCase (e.g. savePath, record, profileName).
• Drift history UI in Stage 5 features an interactive SVG trend chart with ICCery verification reference bands, consecutive-breach alert card (requires \ge 2 consecutive runs \ge 3.5 on distinct calendar days or \ge 1 hour apart), and RFC-4180 compliant CSV export.

Stage 3 XY Automated Scanning Tables (#93)

• Supports automated XY scanning tables (GretagMacbeth SpectroScan, X-Rite i1iO) in Stage 3 chartread.
• Hardware detection in instlist flags devices matching /spectro\s?scan|i1io/i with data-xy="1" and · XY Table label suffix.
• Runtime auto-detection activates when any XY-specific prompt is classified from chartread stdout (supporting i1iO units reporting as i1Pro).
• XY State Machine additions:
  • STATE.TABLE_PLACE_SHEET: Prompts user to place sheet on table; button displays "✓ Sheet Placed — Continue".
  • STATE.TABLE_ALIGN: Prompts user to align measurement head with target fiducial patches (locate patch <ID> with sight); button displays "✓ Aligned — Continue".
• Two-line prompt handling & sticky state:
  • Argyll chartread.c splits XY prompts across two lines (prompt line followed by hit return to continue...).
  • While in TABLE_PLACE_SHEET or TABLE_ALIGN, subsequent continuation lines remain sticky in that table state, preserving the custom button label and preventing regression to generic PROMPT_CONTINUE.
• Button behaviors:
  • btnAccept: in TABLE_* states, sends \n without forcing STATE.READING; the state machine advances naturally when Argyll emits the next prompt.
  • btnCancel: in TABLE_* states or when an XY table is active, sends q\n first to allow the hardware to park its measurement head gracefully before terminating the process.
• Multi-sheet and final sheet notice:
  • Multi-sheet targets are measured within a single chartread process lifecycle; sheet changes transition through TABLE_PLACE_SHEET without opening the Stage 3 multi-pass averaging panel.
  • Please remove last sheet from table is emitted by Argyll right before writing .ti3 and exiting; it is classified as an info-only notice (isRemoveSheetNotice: true) and does not prompt for user input.
• Testing:
  • Pure line classification unit tests live in src/js/chartread.test.js (executable directly in Node or browser console).
  • Unix/macOS mock script src-tauri/argyll/mocks/chartread.mock supports --xy flag (or MOCK_XY_TABLE=1) with blocking read calls simulating calibration, sheet placement, fiducial alignment, and scanning.