#!/usr/bin/env bash
# Requires Bash, curl 7.76+, jq, Python 3, ffprobe and uuidgen.
# Place a video at talk.mp4 and set CERUL_API_KEY before running.
# Processing and search consume credits. OCR is off by default.
set -euo pipefail
: "${CERUL_API_KEY:?Set CERUL_API_KEY before running}"
CERUL_QUICKSTART_RUN_ID=$(uuidgen)
export CERUL_QUICKSTART_RUN_ID
export OCR_ENABLED=false
CERUL_TMP=$(mktemp -d)
trap 'rm -rf "$CERUL_TMP"' EXIT

cerul() {
  curl --fail-with-body --silent --show-error \
    --connect-timeout 15 --max-time 120 \
    -H "Authorization: Bearer $CERUL_API_KEY" \
    -H "Content-Type: application/json" "$@"
}

cerul https://api.cerul.ai/videos | jq '.data'

FILE_SIZE=$(wc -c < ./talk.mp4 | tr -d ' ')
CONTENT_SHA256=$(python3 - <<'PY'
import hashlib
digest = hashlib.sha256()
with open('talk.mp4', 'rb') as source:
    for chunk in iter(lambda: source.read(1024 * 1024), b''):
        digest.update(chunk)
print(digest.hexdigest())
PY
)
DURATION_SECONDS=$(ffprobe -v error -show_entries format=duration -of csv=p=0 ./talk.mp4)

VIDEO=$(cerul https://api.cerul.ai/videos \
  -H "Idempotency-Key: add-video-$CERUL_QUICKSTART_RUN_ID" \
  -d "$(jq -n --argjson size "$FILE_SIZE" \
    '{filename:"talk.mp4",media_type:"video/mp4",byte_size:$size}')")
CERUL_VIDEO_ID=$(printf '%s' "$VIDEO" | jq -er '.data.id')
export CERUL_VIDEO_ID
printf 'Video ID: %s\n' "$CERUL_VIDEO_ID"
UPLOAD_CONTENT_TYPE=$(printf '%s' "$VIDEO" | jq -er '.data.upload.required_headers["Content-Type"]')
printf '%s' "$VIDEO" | jq -e '.data.upload.parts | length > 0' >/dev/null
printf '%s' "$VIDEO" | jq -c '.data.upload.parts[]' > "$CERUL_TMP/parts.jsonl"

# Upload exact byte ranges using only the signed headers, without the API Key.
OFFSET=0
while IFS= read -r PART; do
  PART_URL=$(printf '%s' "$PART" | jq -er '.url')
  PART_SIZE=$(printf '%s' "$PART" | jq -er '.byte_size')
  python3 - "$OFFSET" "$PART_SIZE" "$CERUL_TMP/part.bin" <<'PY'
import sys
offset, size, target = sys.argv[1:]
remaining = int(size)
with open('talk.mp4', 'rb') as source, open(target, 'wb') as part:
    source.seek(int(offset))
    while remaining:
        chunk = source.read(min(1024 * 1024, remaining))
        if not chunk:
            raise RuntimeError('Source shorter than reserved upload')
        part.write(chunk)
        remaining -= len(chunk)
PY
  curl --fail-with-body --silent --show-error -X PUT "$PART_URL" \
    --connect-timeout 15 --max-time 1800 \
    -H "Content-Type: $UPLOAD_CONTENT_TYPE" -H "Content-Length: $PART_SIZE" \
    --upload-file "$CERUL_TMP/part.bin"
  OFFSET=$((OFFSET + PART_SIZE))
done < "$CERUL_TMP/parts.jsonl"

# Defaults to speech and visual indexing. OCR is an explicit paid option.
cerul "https://api.cerul.ai/videos/$CERUL_VIDEO_ID/complete" \
  -H "Idempotency-Key: complete-video-$CERUL_QUICKSTART_RUN_ID" \
  -d "$(jq -n --argjson duration "$DURATION_SECONDS" --arg hash "$CONTENT_SHA256" \
    --argjson ocr "$OCR_ENABLED" \
    '{duration_seconds:$duration,content_sha256:$hash,processing:{ocr:{enabled:$ocr,interval_seconds:10}}}')" \
  > "$CERUL_TMP/completed.json"

DEADLINE=$((SECONDS + 1800))
while :; do
  VIDEO=$(cerul "https://api.cerul.ai/videos/$CERUL_VIDEO_ID")
  STATUS=$(printf '%s' "$VIDEO" | jq -er '.data.status')
  case "$STATUS" in
    ready) printf '%s' "$VIDEO" | jq '.data.coverage'; break ;;
    failed) printf '%s' "$VIDEO" | jq '.data.error' >&2; exit 1 ;;
    uploading|uploaded|processing) ;;
    *) echo "Unexpected video status: $STATUS" >&2; exit 1 ;;
  esac
  if (( SECONDS >= DEADLINE )); then
    echo "Timed out; save $CERUL_VIDEO_ID and resume polling. The server job continues." >&2
    exit 1
  fi
  sleep 5
done

SEARCH=$(cerul https://api.cerul.ai/search \
  -H "Idempotency-Key: search-quickstart-$CERUL_QUICKSTART_RUN_ID" \
  -d "$(jq -n --arg video "$CERUL_VIDEO_ID" \
    '{query:"when did they discuss scaling laws",filter:{video_ids:[$video]},limit:5}')")
printf '%s' "$SEARCH" | jq '{request_id,usage,warnings,data}'

# A valid search can return no results. Export only a bounded evidence span.
if ! HIT=$(printf '%s' "$SEARCH" | jq -ce 'first(.data[] | select(
  (.evidence.start_seconds | type) == "number" and
  (.evidence.end_seconds | type) == "number" and
  .evidence.end_seconds > .evidence.start_seconds))'); then
  echo "No evidence span to export. Review the query or available coverage."
  exit 0
fi
CLIP_BODY=$(printf '%s' "$HIT" | jq --arg video "$CERUL_VIDEO_ID" \
  '{video_id:$video,start_s:.evidence.start_seconds,end_s:.evidence.end_seconds,evidence_ids:[.evidence.id]}')
CLIP=$(cerul https://api.cerul.ai/clips \
  -H "Idempotency-Key: clip-quickstart-$CERUL_QUICKSTART_RUN_ID" -d "$CLIP_BODY")
CLIP_JOB_ID=$(printf '%s' "$CLIP" | jq -er '.data.id')

DEADLINE=$((SECONDS + 1800))
while :; do
  JOB=$(cerul "https://api.cerul.ai/jobs/$CLIP_JOB_ID" -D "$CERUL_TMP/job-headers")
  CLIP_STATUS=$(printf '%s' "$JOB" | jq -er '.data.status')
  case "$CLIP_STATUS" in
    succeeded) break ;;
    failed|canceled) printf '%s' "$JOB" | jq '.data' >&2; exit 1 ;;
  esac
  if (( SECONDS >= DEADLINE )); then
    echo "Timed out; resume polling job $CLIP_JOB_ID. The server job continues." >&2
    exit 1
  fi
  RETRY_AFTER=$(awk 'tolower($1)=="retry-after:" {gsub("\r", "", $2); print $2}' "$CERUL_TMP/job-headers")
  case "$RETRY_AFTER" in ''|*[!0-9]*) RETRY_AFTER=3 ;; esac
  sleep "$RETRY_AFTER"
done

ARTIFACTS=$(cerul "https://api.cerul.ai/jobs/$CLIP_JOB_ID/artifacts")
ARTIFACT_ID=$(printf '%s' "$ARTIFACTS" | jq -er '.data[0].id')
cerul "https://api.cerul.ai/artifacts/$ARTIFACT_ID" | jq '.data'
cerul "https://api.cerul.ai/artifacts/$ARTIFACT_ID/content" --output quickstart-clip.mp4