#!/bin/bash
# Check everything we can before changing anything, so problems show up in the first minute, not the last.
. "$(dirname "$0")/lib.sh"

problems=0
need() { have "$1" || { warn "missing command: $1 ($2)"; problems=$((problems+1)); }; }

say "operating system"
[ -f /etc/os-release ] && . /etc/os-release && say "  ${PRETTY_NAME:-unknown} on $(uname -m)"
if ! have systemctl || [ ! -d /run/systemd/system ]; then
  die "this installer needs systemd (it runs the agent and the runtimes as services). On a system without systemd, run the agent by hand: see docs/en/INSTALL.md (Linux without systemd)."
fi
need curl "$(pkg_hint curl)"
need python3 "$(pkg_hint python3)"
need tar "$(pkg_hint tar)"
need sudo "$(pkg_hint sudo)"
need sha256sum "coreutils"
need getent "libc-bin"
[ "$problems" = 0 ] || die "install the missing commands above and run again"

say "agent user: $OWNER (uid $OWNER_UID)"

if step_enabled agent; then
  if [ -z "${WXCH_TOKEN:-}" ] && [ ! -s /etc/wxch-agent.env ]; then
    die "no agent token. On the server run: wxch-server -config <config.json> -add-agent <name>   and use the command it prints."
  fi
fi

say "network"
curl -fsS -o /dev/null --max-time 15 "$BASE/get" || die "cannot reach $BASE/get"
if step_enabled ollama;   then curl -fsS -o /dev/null --max-time 15 -I https://ollama.com        || warn "cannot reach ollama.com (Ollama install will fail)"; fi
if step_enabled lmstudio; then curl -fsS -o /dev/null --max-time 15 -I https://lmstudio.ai        || warn "cannot reach lmstudio.ai (LM Studio install will fail)"; fi
if step_enabled vllm;     then curl -fsS -o /dev/null --max-time 15 -I https://huggingface.co     || warn "cannot reach huggingface.co (vLLM model download will fail)"
                               curl -sS  -o /dev/null --max-time 15 -I https://nvcr.io/v2/        || warn "cannot reach nvcr.io (vLLM image pull will fail)"; fi

# vLLM needs Docker and an NVIDIA GPU. Where they are missing it is skipped with a note instead of failing the
# whole install (name it in WXCH_ONLY or set WXCH_VLLM=required to make it an error instead).
rm -f /run/wxch-auto-skip-*
if step_enabled vllm; then
  say "docker and GPU (vLLM)"
  why=""
  have nvidia-smi && nvidia-smi -L >/dev/null 2>&1 || why="no NVIDIA GPU found"
  [ -z "$why" ] && { have docker || why="docker is not installed"; }
  [ -z "$why" ] && { docker info >/dev/null 2>&1 || why="docker is installed but not running (systemctl start docker)"; }
  if [ -n "$why" ]; then
    if [ -n "${WXCH_ONLY:-}" ] || [ "${WXCH_VLLM:-}" = required ]; then die "vLLM cannot be installed: $why"; fi
    warn "skipping vLLM: $why. Ollama and LM Studio are still installed."
    touch /run/wxch-auto-skip-vllm
  else
    have nvidia-ctk || warn "nvidia-ctk not found: GPU containers may not start (install the NVIDIA container toolkit)"
  fi
fi
have nvidia-smi && say "  GPU: $(nvidia-smi --query-gpu=name --format=csv,noheader | head -1)" || warn "nvidia-smi not found: models would run on the CPU"

if step_enabled lmstudio; then
  ldconfig -p 2>/dev/null | grep -q libatomic || { say "installing $(pkg_name libatomic) (needed by LM Studio)"; pkg_install "$(pkg_name libatomic)" || warn "could not install $(pkg_name libatomic)"; }
fi

# disk space: Docker image ~11 GB, starter models ~3 GB, runtimes ~3 GB
target="$MODELS_DIR"; [ -d "$target" ] || target=/
free_gb=$(df -BG --output=avail "$target" 2>/dev/null | tail -1 | tr -dc '0-9')
root_gb=$(df -BG --output=avail / 2>/dev/null | tail -1 | tr -dc '0-9')
say "free space: ${free_gb:-?} GB for models, ${root_gb:-?} GB on the system disk"
[ "${root_gb:-100}" -ge 20 ] || warn "less than 20 GB free on the system disk (the vLLM image needs about 11 GB)"

say "preflight ok"