#!/usr/bin/env bash
# Starts HTTP server on port 7860 immediately, runs conversion in background.
# The HTTP server is the main foreground process — container stays alive as
# long as it's running regardless of what the conversion does.

set -euo pipefail

mkdir -p /workspace/output /workspace/scratch
LOG_FILE="/workspace/output/conversion.log"

# Kick off the conversion pipeline in the background before anything else
(
  exec >> "$LOG_FILE" 2>&1

  echo ""
  echo "════════════════════════════════════════════════════"
  echo " DeepSeek-Math-V2 GGUF Converter"
  echo "════════════════════════════════════════════════════"
  echo "Date   : $(date)"
  echo "Disk   : $(df -h /workspace | tail -1)"
  echo "Memory : $(free -h | grep Mem)"
  echo ""

  # Build llama-quantize if not present
  if ! command -v llama-quantize &>/dev/null; then
    echo "Building llama-quantize (~2 min)..."
    cd /opt/llama.cpp
    cmake -B build \
      -DCMAKE_BUILD_TYPE=Release \
      -DLLAMA_NATIVE=OFF \
      -DBUILD_SHARED_LIBS=OFF \
      -DLLAMA_BUILD_TESTS=OFF \
      -DLLAMA_BUILD_EXAMPLES=OFF \
      -DLLAMA_BUILD_SERVER=OFF \
      2>&1 | tail -3
    cmake --build build --config Release --target llama-quantize -j$(nproc)
    install -m 755 build/bin/llama-quantize /usr/local/bin/llama-quantize
    echo "llama-quantize ready"
    cd /workspace
  fi

  bash /workspace/scripts/convert.sh

  if [[ $? -eq 0 ]]; then
    echo ""
    echo "════════════════════════════════════════════════════"
    echo " ✔ Conversion complete! Container staying alive."
    echo "════════════════════════════════════════════════════"
  else
    echo ""
    echo "════════════════════════════════════════════════════"
    echo " ✘ Conversion FAILED. Container staying alive."
    echo "════════════════════════════════════════════════════"
  fi
) &

# Start the HTTP status server in the FOREGROUND — this is PID 1's main job.
# Container lives exactly as long as this process lives.
exec python3 /workspace/scripts/status_server.py "$LOG_FILE"