v1.3.3

[1.3.3] - 2026-08-30

Added

• Model-lock tests (tests/model_lock/): frozen 8200i SE driver-path oracles (geometry, crop mapping, dummy trim, slope-table feed order, ME configure, Scan Lab window kwargs). Other-model fixes must specialize that model rather than retarget these asserts. Run with uv run pytest -m model_lock.

Fixed

• ME short/long width mismatch: content-aware ENDPIXEL dummy trim could drop a different column count on short vs long (and IR) because edge DN differs — especially at 3600 dpi where the cap is larger. Multi-pass now discovers the drop on the short/first colour plane and reuses it for long and IR assemble so merge/align see matching widths. Single-pass stays content-aware.

v1.3.2

[1.3.2] - 2026-08-30

Fixed

• Scan Lab Prescan crop: a click or too-small rubber-band no longer clears a committed crop (right-click to clear). Scan builds the hardware window on the GUI thread and logs crop applied vs full window so a dropped crop cannot silently scan the safe window (including ME/IR passes).
• Scan Lab Prescan crop X mapping: the rubber-band is in mirrored display space. Mapping now flips X back to TA so Color short / ME / IR match the box (cropping the right film base no longer cuts the left).
• Left-edge frame clip: a fixed ENDPIXEL dummy drop was removing 16/24 px of film gate after mirror_x. Assemble now trims only columns that look like USB dummy, capped at the old width.
• 8200i SE horizontal framing: image STR/END are shifted by 96 native clocks (optical_end_inactive_native) toward ENDPIXEL so dummy-trim + mirror_x no longer clips the film gate on the left while showing extra holder on the right. USB span / line_bytes stay the same; shading still tracks the image window.
• 8200i SE positioning slope tables: SilverFast uploads SLOPE_TABLE_SLOW for the first (reference) feed and SLOPE_TABLE_FAST for the second (final positioning) feed — 39/39 capture pairs. pyopticfilm previously used the fast ramp for both. The V2-only use_slow_final_positioning_feed flag is no longer on Model8200iSE (8100 V2 still uses fast-then-slow).

Changed

• 8200i SE priming default: default_gl128_prime is now False (same as the 8100 V2). Scanner.scan() with gl128_prime unset skips the discarded first-scan AGOHOME-park pass. Explicit gl128_prime=True still primes.

Contributors

• @TobbyTravel for default priming off on the 8100 V2 and for scoping the slow-second-feed slope-table fix to V2.

v1.3.1

[1.3.1] - 2026-08-29

Added

• Manual exposure overrides (GL128): Scanner.scan(single_pass_exposure=..., me_short_exposure=..., me_long_exposure=...) send an exact REG_EXPOSURE for the retained single-pass, ME short, or ME long pass, bypassing adaptive selection, PPI clamp, and me_hardware_max_exposure. me_long_exposure takes precedence over me_exposure_mode. Values are validated for the 24-bit register range (1–0xFFFFFF); out-of-range values raise ValueError. All three default to None (unchanged behavior) and never apply to the discarded GL128 priming pass. Scan Lab exposes matching Manual exposure overrides fields.
• GL128 priming-pass override: Scanner.scan(gl128_prime=False) skips the discarded first-scan AGOHOME-park pass for that call (debug/testing only — expect ~30 px of first-scan position drift). Skipping does not mark the scanner as primed, so a later call without the override still primes normally. ScanStatus / on_status gains "prime_skipped". Defaults to True (unchanged behavior). Scan Lab Disable priming pass (debug) applies to Prescan and Scan.

Fixed

• Cropped 8200i SE ENDPIXEL dummy suffix: STR/END are computed in native 7200 dpi clocks (session 03 origin STR 242 / END 10610; same crop keeps STR/END at 1800 and 3600). The last ~24 output columns at 1800 (96 native clocks; image left after mirror_x) are a per-line USB dummy suffix plus the invert-white transition — shrinking ENDPIXEL does not remove them. Decode drops that fixed width after shading; USB line_bytes stays locked to the programmed span. Heuristic post-decode edge trim is not used (pcap 7200 URB-wider-than-span trim stays). IR column flatten still uses inset edge levels and a gain cap so residual edge samples cannot blow up to ~56k. Pass registration fills shift borders from interior columns instead of replicating padding (which widened the IR bright band and caused NegPy ICE over-repaint). Scan Lab surfaces IR align shift in the status bar / IR caption.
• OpticFilm 8100 (V2) geometry/timing: five constants inherited from the 8200i SE without V2 capture proof are now V2-specific — feed_to_scan_steps 13128 (was 13704; fixes ~1 mm clipped at the top of full-frame scans), 7200 dpi LPERIOD 16035 (was 15963), 7200 white-shading strip dummy clock 0x10 (was 0x17), max_image_lincnt_by_feed2 {13128: 29012} (was SE preview entry 4836), and ladder_feed2_steps 13128 (was SE 13560).
• GL128 consecutive-scan AFE strip timeout: after an image pass, stationary AFE START could hang for 5s at status 0xcd (home, buffer empty, lamp on, motor still enabled). Scan Lab reported AFE strip: no stationary data ready within 5s (last status=0xcd) — including 1800→3600 with the same crop (DPI cache miss remesures calib). Stationary acquires now abort START, confirm SCAN/motor-off from hardware (park leaves 0x02 armed), program session-03 dark-strip clocks, write SCAN absolutely, and retry START once. Colour shading remesure reuses existing AFE codes when the last search was colour.

Changed

• GL128 AFE zero-collapse recovery logs now name which offset/gain components collapsed, which fallback was used, and warn when image colour may be degraded.

Contributors

• @TobbyTravel for OpticFilm 8100 (V2) capture-derived geometry/timing fixes and for GL128 manual exposure overrides, priming-pass skip, and improved AFE zero-collapse diagnostics.

v1.3.0

[1.3.0] - 2026-08-27

Added

• GL128 first-scan priming: before the first retained scan after open, run a discarded small pass (default 600 dpi, area (0, 0, 1, 0.12)) so image-pass AGOHOME parks the carriage for repeatable start position. Override with POF_GL128_PRIME (full or <dpi>:x0,y0,x1,y1).
• Adaptive ME long exposure (GL128): after the short colour pass, choose a frame-specific long REG_EXPOSURE from RGB dense percentiles, then clamp through a separate safety envelope (42k–85k, max ratio 7×). Failures fall back to fixed 42000. Opt out with me_exposure_mode="fixed".
• Scan Lab Fixed 42k long (A/B) checkbox; USB/status log shows proposed / selected long exposure and clamp reason.
• MeScanDebug.exposure_proposed / exposure_reason for lab observability.
• Optional on_status callback on Scanner.scan and exported ScanStatus ("priming" / "scanning") so hosts can show GL128 priming; Scan Lab status bar and USB log surface it.

Fixed

• GL128 discarded priming pass always forces geometry=None, apply_calib=False, and mode="color", so hosts that pass bring-up geometry (e.g. Scan Lab) no longer stretch the prime into a full request-PPI shading+scan cycle.
• GL128 adaptive quiet USB drain now sleeps the full LPERIOD deficit (plus a small lag) instead of capping at 3 ms/line, so 7200 dpi image creep is no longer ~15% ahead of the ASIC line clock.

Changed

• GL128 multi-exposure colour-long REG_EXPOSURE is clamped by PPI (8200i SE + 8100 V2): max 42000 at 7200 dpi (SilverFast known-good); 14000–85000 at other resolutions.
• PyPI development status classifier is now 4 - Beta (was 3 - Alpha).
• README now points the "Latest release" link at v1.3.0.
• README and docs/ now document the hardware-tested set as 8200i SE + 8100 (V2) (and distinguish GL845 8100 07b3:130c from GL128 8100 V2 07b3:1824).
• Comments, docstrings, and user-facing errors now describe the two-model scan-ready set; Scan Lab README matches. Internal bring-up helper renamed is_opticfilm_8200i_se → is_gl128_opticfilm.
• ME long-pass pixel clocks are selected via an explicit long-pass flag (not only exposure >= 42000); REG_EXPOSURE is hard-clamped to me_hardware_max_exposure at configure time.

Contributors

• @TobbyTravel for GL128 first-scan priming and for OpticFilm 8100 (V2) exposure-ladder research showing REG_EXPOSURE has no hardware ceiling and that the multi-exposure long bin should be raised toward film-dependent ~56–64k targets (not a 42k limit).

v1.2.0

[1.2.0] - 2026-08-24

Added

• OpticFilm 8100 (V2) (07b3:1824) support: GL128 sibling of the 8200i SE, hardware-validated for single-pass and multi-exposure colour scans. The 8100 V2 has no infrared channel / iSRD, so infrared=True and mode="infrared" raise a clear ScanError (multi-exposure short+long bracket remains available).
• Model8100V2 / MODEL_8100_V2 (subclass of Model8200iSE; register tables, geometry and motor clamps inherited from the capture-validated SE).
• IR capability guard in Gl128ScanSession.run for models with supports_infrared=False.
• examples/scan.py: barebones interactive Python CLI example for enumerating scanners, selecting a device, choosing PPI, optionally enabling ME / IR where supported, and saving 16-bit TIFF output.

Changed

• Scan validation set is now 8200i SE + 8100 (V2); tests updated from "SE only" to the two-model validated set.
• README hardware support docs now cover the newly validated 8100 V2 path.

Contributors

• @TobbyTravel for the OpticFilm 8100 (V2) enablement, validation updates, version bump, and release documentation polish.

v1.1.2

[1.1.2] - 2026-08-22

Added

• Multi-exposure (ME) scanning for OpticFilm 8200i SE (GL128): short and long colour passes with host SNR/IVW merge into the deliverable ScanImage.rgb.
• MeScanDebug and Scanner.last_me_debug for lab-only bracket planes, fusion stats, and pass alignment shifts (not part of the public ScanImage type).
• Scan Lab (tools/scanlab/): PyQt6 bring-up GUI (mock or real SE), prescan crop, IR pass, ME tabs, USB log, pcap decode, calib cache controls, and adaptive quiet-drain toggle.
• docs/negpy-integration.md: integrator guide for scan orientation, ScanImage contract, and NegPy migration notes.
• Adaptive quiet USB drain default on GL128 (DEFAULT_IMAGE_USB_PACE_S, line-aligned throttle in bulk acquire).
• tools/audit_me_bracket: bracket TIFF audit; --align reports estimated pass shift and post-align luma residual.
• Mock scanner tests and expanded geometry / pipeline test coverage.
• opencv-python-headless in the lab dependency group for ME/IR pass registration in Scan Lab.

Fixed

• Scan orientation on mirror_x models (8200i SE): left–right correction in ImagePipeline.assemble(); prescan crop no longer applies a compensating X flip.
• ME shadow colour fringes from short/long misregistration: improved OpenCV phase-correlation alignment (AREA downsampling, ROI refinement), merge guard at misaligned edges, and chunked IVW merge to avoid float32 OOM at 3600+ dpi.
• High-PPI memory use in the image pipeline (integer oversample sums, chunked host calib / film-base makeup / border clamp).
• 8200i SE IR scans: correct colour DVDSET handling, flattened green IR plane on ScanImage.ir, grayscale preview in Scan Lab.
• Scan Lab: crop UI limited to Prescan tab; missing busy_changed worker signal; effective crop status in status bar.

Changed

• ScanImage slimmed to NegPy-facing fields only (rgb, dpi, device_model, optional ir). ME bracket data moved to Scanner.last_me_debug.
• ME short/long planes stay linear on the bracket; film-base makeup runs once on the merged deliverable.
• Non-SE bring-up: short Y travel clamps, USB planar RGB toggle, and SANE-aligned session improvements for experimental models.

Notes for integrators

• Forward-only orientation fix: scans saved before 1.1.2 may still appear mirrored on disk; rescan if orientation matters.
• NegPy / other apps: consume ScanImage.rgb / ir as returned; do not apply an extra [:, ::-1] flip or prescan crop mirror compensation. See docs/negpy-integration.md.
• Breaking: removed ScanImage ME fields (rgb_short, rgb_long, exposure_*, merge_*, align_shift_*). Use Scanner.last_me_debug for bracket inspection.

v1.1.1

cancel mid-scan USB wedge fix

pyopticfilm 1.1

Minor changes

pyopticfilm 1.0

Initial release