#!/usr/bin/env bash
# Download every paper in manifest.tsv into reference_papers/<venue>/<year>_<slug>.pdf
# Skips files already present and valid. Polite to arXiv (3 s between requests).
cd "$(dirname "$0")"
UA="Mozilla/5.0 (X11; Linux x86_64) reference-fetch/1.0"
ok=0; bad=0
while IFS=$'\t' read -r venue year slug source title; do
  [[ "$venue" == \#* || -z "$venue" ]] && continue
  dir="$venue"; out="$dir/${year}_${slug}.pdf"; mkdir -p "$dir"
  if [ -s "$out" ] && head -c4 "$out" | grep -q '%PDF'; then ok=$((ok+1)); continue; fi
  case "$source" in
    acl:*)   url="https://aclanthology.org/${source#acl:}.pdf" ;;
    arxiv:*) url="https://arxiv.org/pdf/${source#arxiv:}"; sleep 3 ;;
    url:*)   url="${source#url:}" ;;
  esac
  curl -sL -m 120 -A "$UA" -o "$out" "$url"
  if head -c4 "$out" 2>/dev/null | grep -q '%PDF'; then ok=$((ok+1)); echo "ok   $out"
  else bad=$((bad+1)); echo "FAIL $out  <- $url ($(file -b "$out" | cut -c1-30))"; rm -f "$out"; fi
done < manifest.tsv
echo "downloaded/valid: $ok  failed: $bad"