[LeeO86]

Fixes interlaced handling (audit 2026-10-16, sections A, B and D) per the shared convention (MXL 218ddaa): an interlaced flow holds one field per grain at twice the frame rate; grain 2k = the first field in time of TAI frame k, 2k + 1 = the second; tff first field = even lines, bff = odd lines; ANC at the field rate with the video field's index. Spec: new SPECIFICATION §3.2.2.

Capture (SDI → MXL)

• Frame index k from the frame's start (one period before its timestamp) at the frame rate; grain 2k gets the first field's lines, 2k + 1 the other lines, by row (v210, and 8-bit YUV after the expansion; v210a: fill lines). Before: one grain per frame at the field-rate index holding frame lines 0–539, odd indexes invalid.
• Field order from the mode table: NTSC is interlaced_bff (DeckLink bmdLowerFieldFirst) in the flow definition and NMOS; PAL and 1080i stay tff. Before: tff hard-coded.
• ANC flow of an interlaced mode declared at 2× the frame rate; field 1's packets at 2k, field 2's at 2k + 1 (field from the SMPTE line number; an empty grain for a field without packets).

Playout (MXL → SDI)

• VideoReader reads an interlaced flow as frames: grains 2k and 2k + 1 woven with line parity (idx & 1) ^ bff; its rate, head, ring size and indexes are frames, so the read position, preroll, read offset, catch-up, slips and the audio per frame (1920 samples at 1080i50, was 960) all use the frame rate. The sync-group wait waits for the second field's time. A frame with an invalid field repeats (source_invalid). A re-created flow that switches progressive ↔ interlaced restarts the channel.
• The output does not read ANC, so there is no ANC reader to change.

Other

• Web UI MXL tab maps interlaced flows to HD1080i50/HD1080i5994/PAL/NTSC (it looked for HD1080i25, which does not exist).
• Mock: odd lines' counter band is grey in interlaced modes (field marker); MOCK_PLAYBACK_FRAME_FILE dumps played frames for lab checks.
• Version 1.0.5 (CMakeLists, version.hpp, deploy manifest), CHANGELOG, README.

Tests

• Unit: field split/weave tff + bff (every line back in place), frame index from the frame start and field timestamps (1080i50, 1080i59.94), frame head from field head, ANC line → field, field order table, flow written field by field and read back woven (PAL tff, NTSC bff, real MXL domain), output channel playing an interlaced flow (fake card): frame rate, woven frame exact, 1920-sample audio blocks. 116/116 pass locally.
• Mock lab (HD1080i50, CH0 capture with ANC + CH1 output of that flow): grains of 540 lines, both parities valid, grain 2k = even lines, ANC flow 50/1 with the video's indexes (packet at 2k, empty at 2k+1); played frame woven exactly. Own tff 1080i25 writer → output: every frame row in place, 25 frames/s, no repeats. HD1080p50: unchanged (1080-line grains, ANC 50/1 one packet per grain, same behaviour as 1.0.4).

Round 2 (4786562)

• RFC 8331 F bits: interlaced ANC grain 2k = 0b10, 2k+1 = 0b11; progressive 0b00.
• NTSC ANC line 0 (unknown) counts as field 1.
• NMOS ANC flow grain_rate and ANC receiver caps at the field rate for interlaced modes (video keeps the frame rate). Lab: 1080i50 data flow 50/1, video 25/1.
• VideoReader warns mxl_flow_def_unreadable instead of silently reading as progressive; an invalid first field returns the frame invalid at once.
• Docs: empty ANC grain only when the other field has packets; grain counters count 2 per interlaced frame.
• Tests: F bits, line 0, invalid first field, InputChannel interlaced capture with a fake card.

Not changed / found

• Outputs at 29.97/59.94 (progressive and interlaced) read audio at frame index × floor(48000/rate) samples, which is off the TAI sample grid (1601 vs 1601.6 per frame), so their audio is likely silent. Pre-existing, not part of this PR.

🤖 Generated with Claude Code

https://claude.ai/code/session_01St7QwmonpoG9EEEqaLFXLi