#!/usr/bin/env bash
set -euo pipefail

ROOT_INPUT="${1:-input}"   # ex: input
N="${2:-50}"               # nb dossiers à traiter

# Options
FORCE="${FORCE:-0}"        # FORCE=1 => retraiter même si output JSON existe
CLEAN="${CLEAN:-0}"        # CLEAN=1 => supprime outputs du dossier avant retraitement (recommandé avec FORCE)
RANDOM_PICK="${RANDOM_PICK:-0}"  # RANDOM_PICK=1 => choisir N dossiers aléatoires
MAX_PARALLEL="${OLLAMA_MAX_PARALLEL:-1}"

# Modèles
export OLLAMA_CODER_MODEL="${OLLAMA_CODER_MODEL:-gemma3:27b}"
export OLLAMA_VERIFIER_MODEL="${OLLAMA_VERIFIER_MODEL:-deepseek-v3.2:cloud}"
export OLLAMA_MAX_PARALLEL="$MAX_PARALLEL"

# Python du venv (fiable)
PY="./.venv/bin/python"
if [[ ! -x "$PY" ]]; then
  echo "❌ Venv introuvable: $PY"
  echo "   Active ton venv ou crée-le, puis relance."
  exit 1
fi

RUN_ID="$(date +%Y%m%d_%H%M%S)"
RUN_DIR="output/batch_runs/$RUN_ID"
LOG_DIR="output/batch_logs/$RUN_ID"
mkdir -p "$RUN_DIR" "$LOG_DIR"

IDS_FILE="$RUN_DIR/ids.txt"
FILES_FILE="$RUN_DIR/files.txt"

echo "=== Batch Run: $RUN_ID ===" | tee "$RUN_DIR/summary.txt"
echo "ROOT_INPUT=$ROOT_INPUT  N=$N  FORCE=$FORCE  CLEAN=$CLEAN  RANDOM_PICK=$RANDOM_PICK" | tee -a "$RUN_DIR/summary.txt"
echo "CODER=$OLLAMA_CODER_MODEL  VERIFIER=$OLLAMA_VERIFIER_MODEL  OLLAMA_MAX_PARALLEL=$OLLAMA_MAX_PARALLEL" | tee -a "$RUN_DIR/summary.txt"
echo | tee -a "$RUN_DIR/summary.txt"

# Liste des dossiers = sous-dossiers avec au moins 1 pdf
LIST_CMD=(find "$ROOT_INPUT" -mindepth 1 -maxdepth 1 -type d -print)
mapfile -t ALL_DIRS < <("${LIST_CMD[@]}" | while read -r d; do
  compgen -G "$d/*.pdf" >/dev/null && echo "$d"
done)

if [[ "${#ALL_DIRS[@]}" -eq 0 ]]; then
  echo "❌ Aucun dossier avec PDF trouvé dans: $ROOT_INPUT"
  exit 1
fi

# Sélection N dossiers
if [[ "$RANDOM_PICK" == "1" ]]; then
  mapfile -t DOSSIERS < <(printf "%s\n" "${ALL_DIRS[@]}" | shuf | head -n "$N")
else
  mapfile -t DOSSIERS < <(printf "%s\n" "${ALL_DIRS[@]}" | sort | head -n "$N")
fi

echo "→ Dossiers sélectionnés: ${#DOSSIERS[@]}" | tee -a "$RUN_DIR/summary.txt"

# Traitement
for d in "${DOSSIERS[@]}"; do
  id="$(basename "$d")"
  out_json="output/structured/$id/${id}_fusionne_cim10.json"
  log="$LOG_DIR/${id}.log"

  # Enregistre l'ID (pour stats de fin)
  echo "$id" >> "$IDS_FILE"

  if [[ -f "$out_json" && "$FORCE" != "1" ]]; then
    echo "⏭️  SKIP $id (déjà traité)" | tee -a "$RUN_DIR/summary.txt"
    continue
  fi

  if [[ "$CLEAN" == "1" ]]; then
    rm -rf "output/structured/$id" "output/reports/$id" "output/anonymized/$id" 2>/dev/null || true
  fi

  echo "▶️  START $id" | tee -a "$RUN_DIR/summary.txt"
  ("$PY" -m src.main "$d") 2>&1 | tee "$log"
  echo "✅ DONE  $id" | tee -a "$RUN_DIR/summary.txt"
done

# Construit la liste des fichiers JSON réellement présents pour ce run
: > "$FILES_FILE"
while read -r id; do
  f="output/structured/$id/${id}_fusionne_cim10.json"
  [[ -f "$f" ]] && echo "$f" >> "$FILES_FILE"
done < "$IDS_FILE"

COUNT_FILES=$(wc -l < "$FILES_FILE" | tr -d ' ')
echo | tee -a "$RUN_DIR/summary.txt"
echo "→ JSON trouvés pour stats: $COUNT_FILES" | tee -a "$RUN_DIR/summary.txt"

if [[ "$COUNT_FILES" -eq 0 ]]; then
  echo "⚠️ Aucun JSON pour stats. Fin." | tee -a "$RUN_DIR/summary.txt"
  exit 0
fi

echo | tee -a "$RUN_DIR/summary.txt"
echo "=== STATS (sur ce run uniquement) ===" | tee -a "$RUN_DIR/summary.txt"

# 1) Verdicts
echo "--- Verdicts ---" | tee -a "$RUN_DIR/summary.txt"
xargs -a "$FILES_FILE" jq -r '(.veto_report.verdict // "NO_REPORT")' \
| sort | uniq -c | sort -nr | tee -a "$RUN_DIR/summary.txt"

# 2) Top VETOs
echo | tee -a "$RUN_DIR/summary.txt"
echo "--- Top VETOs ---" | tee -a "$RUN_DIR/summary.txt"
xargs -a "$FILES_FILE" jq -r '.veto_report.issues[]?.veto' \
| sort | uniq -c | sort -nr | head -n 20 | tee -a "$RUN_DIR/summary.txt"

# 3) HARD count
echo | tee -a "$RUN_DIR/summary.txt"
echo "--- Dossiers avec HARD ---" | tee -a "$RUN_DIR/summary.txt"
while read -r f; do
  id="$(basename "$f" _fusionne_cim10.json)"
  hard=$(jq '[.veto_report.issues[]? | select(.severity=="HARD")] | length' "$f")
  [[ "$hard" -gt 0 ]] && printf "%s\tHARD=%s\n" "$id" "$hard"
done < "$FILES_FILE" | sort -k2,2nr | tee -a "$RUN_DIR/summary.txt"

# 4) Downgrades (cim10_final != cim10_suggestion)
echo | tee -a "$RUN_DIR/summary.txt"
echo "--- Downgrades (TOP 30) ---" | tee -a "$RUN_DIR/summary.txt"
while read -r f; do
  id="$(basename "$f" _fusionne_cim10.json)"
  dw=$(jq '
    ([
      (.diagnostic_principal? | select(.cim10_final? and .cim10_suggestion? and .cim10_final != .cim10_suggestion) | 1),
      (.diagnostics_associes[]? | select(.cim10_final? and .cim10_suggestion? and .cim10_final != .cim10_suggestion) | 1)
    ] | add) // 0
  ' "$f")
  [[ "$dw" -gt 0 ]] && printf "%s\tDOWN=%s\n" "$id" "$dw"
done < "$FILES_FILE" | sort -k2,2nr | head -n 30 | tee -a "$RUN_DIR/summary.txt"

echo | tee -a "$RUN_DIR/summary.txt"
echo "✅ Stats écrites dans: $RUN_DIR/summary.txt"
echo "📁 Logs dossier par dossier: $LOG_DIR/"