#!/usr/bin/env bash
# Build a uv-managed Python for Overcast's visual DB providers.
#
# Usage:
#   scripts/visual-db-uv.sh           # image matching deps: opencv + numpy
#   scripts/visual-db-uv.sh --face    # also install DeepFace stack
#   scripts/visual-db-uv.sh --clip    # also install OpenAI CLIP (open_clip + torch)
#   scripts/visual-db-uv.sh --detect  # also install the OWLv2 open-vocab DETECTOR (torch + transformers + scipy) for `see --detect`
#   scripts/visual-db-uv.sh --audio   # also install audio fingerprint deps (scipy)
#   scripts/visual-db-uv.sh --clap    # also install LAION CLAP audio embeddings (transformers + torch)
#   scripts/visual-db-uv.sh --voice   # also install pyannote.audio (enhance --ops separate + voice match)
#   scripts/visual-db-uv.sh --segment # also install transformers+SAM2/GroundingDINO (enhance --ops segment)
#   scripts/visual-db-uv.sh --enhance # both enhance stacks (--voice + --segment)
#   scripts/visual-db-uv.sh --all     # install everything (face + clip + detector + audio-fp + CLAP + enhance)
set -euo pipefail

ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
# Default venv location: repo-local .dev/ for a dev checkout, but the DURABLE
# overcast home for an installed package — a venv inside the npm package dir is
# wiped by any reinstall/upgrade. Override with OVERCAST_VISUAL_DB_VENV.
if [ -e "$ROOT/.git" ]; then
  DEFAULT_VENV="$ROOT/.dev/visual-db-py"
else
  DEFAULT_VENV="${OVERCAST_HOME:-$HOME/.overcast}/visual-db-py"
fi
VENV="${OVERCAST_VISUAL_DB_VENV:-$DEFAULT_VENV}"
PYVER="${OVERCAST_VISUAL_DB_PYTHON:-3.12}"
MODE="${1:-}"
if [ -z "$MODE" ]; then
  MODE="--image"
fi

if ! command -v uv >/dev/null 2>&1; then
  echo "uv is required. Install it from https://docs.astral.sh/uv/ or run: curl -LsSf https://astral.sh/uv/install.sh | sh" >&2
  exit 1
fi

mkdir -p "$(dirname "$VENV")"
uv venv --allow-existing --python "$PYVER" "$VENV"

"$VENV/bin/python" -m ensurepip --upgrade >/dev/null 2>&1 || true
uv pip install --python "$VENV/bin/python" --upgrade pip wheel setuptools
# Pin opencv-python < 5: the 5.0.x wheels ship an EMPTY cv2/data/ (no bundled
# haarcascade XMLs), which breaks DeepFace's default opencv face detector
# ("Confirm that opencv is installed…"). 4.x bundles them. Installed FIRST so the
# later deepface/clip/etc. installs keep this resolved version.
uv pip install --python "$VENV/bin/python" "opencv-python<5" numpy

install_voice() {   # pyannote: enhance --ops separate diarization + `voice` speaker match
  uv pip install --python "$VENV/bin/python" "pyannote.audio>=4.0" torch torchaudio
}
install_segment() { # GroundingDINO + SAM 2.1 for enhance --ops segment
  uv pip install --python "$VENV/bin/python" "transformers>=4.56" torch pillow timm
}

case "$MODE" in
  --face|face)
    uv pip install --python "$VENV/bin/python" deepface tf-keras
    ;;
  --clip|clip)
    uv pip install --python "$VENV/bin/python" open-clip-torch torch pillow
    ;;
  --detect|detect)
    # OWLv2 open-vocabulary detector (transformers zero-shot-object-detection);
    # scipy is used by the OWLv2 post-processor. timm is only needed for the
    # optional Grounding DINO model (DETECT_MODEL=IDEA-Research/grounding-dino-tiny).
    uv pip install --python "$VENV/bin/python" torch transformers scipy pillow
    ;;
  --audio|audio)
    uv pip install --python "$VENV/bin/python" scipy
    ;;
  --clap|clap)
    uv pip install --python "$VENV/bin/python" torch transformers
    ;;
  --voice|voice)
    install_voice
    ;;
  --segment|segment)
    install_segment
    ;;
  --enhance|enhance)
    install_voice
    install_segment
    ;;
  --all|all)
    # torch-owning packages first so uv resolves a single shared torch for the
    # CLIP + CLAP stacks (see docs/providers.md on the shared-venv trade).
    uv pip install --python "$VENV/bin/python" open-clip-torch torch transformers pillow
    uv pip install --python "$VENV/bin/python" deepface tf-keras
    # scipy also completes the OWLv2 detector stack (torch + transformers + pillow above)
    uv pip install --python "$VENV/bin/python" scipy
    install_voice
    install_segment
    ;;
  --image|image|"")
    ;;
  *)
    echo "unknown mode: $MODE (expected --image | --face | --clip | --detect | --audio | --clap | --voice | --segment | --enhance | --all)" >&2
    exit 2
    ;;
esac

cat <<EOF
visual DB Python ready:
  $VENV/bin/python

Put this in .env if it is not already set (overcast auto-discovers the repo
.dev/ venv and the \$OVERCAST_HOME/visual-db-py venv without it):
  OC_VISUAL_DB_PY=$VENV/bin/python
EOF

case "$MODE" in
  --detect|detect|--all|all)
    cat <<EOF
  DETECT_PY=$VENV/bin/python          # OWLv2 open-vocab detector for \`see --detect\`

Bind the detector as the see provider, then detect + crop:
  overcast setup provider see "exec:$VENV/bin/python $ROOT/providers/senses/detect/detect.py"
  overcast see ./scene.jpg --detect "person, helmet, truck" --json
EOF
    ;;
esac