chore(deps): let download_go_deps.py fetch Go DeepDoc .ort weights (#19376)

Go DeepDoc now uses .ort weights downloaded via download_go_deps.py — no MODEL_DIR needed from repo root. Script fetches 5 model files into rag/res/deepdoc/; supports mirrors; CI updated. .onnx path unchanged.

scripts/fetch_deepdoc_testdata.sh

@@ -16,9 +16,17 @@
 # Env:
 #   RAGFLOW_TESTDATA_REPO  repo "owner/name" (default infiniflow/ragflow-testdata)
 #   RAGFLOW_TESTDATA_REF   override the anchor tag/ref (else read testdata.ref)
+#   RAGFLOW_TESTDATA_DIR   pre-seeded asset repo root (first level: deepdoc/).
+#                          When it carries deepdoc/<pkg>/testdata, that copy is
+#                          used as-is: no ref pinning, no network fetch. This is
+#                          what a CI runner that mounts the asset repo should
+#                          set; a missing subtree falls back to cloning.
 #   XDG_CACHE_HOME         cache base (default ~/.cache)
 #
 # Behavior:
+#   - With RAGFLOW_TESTDATA_DIR set, the pre-seeded copy wins over everything
+#     below. The ref pin is bypassed — whoever seeds the directory owns its
+#     freshness.
 #   - If testdata is already present INLINE (a real dir, pre-migration), this
 #     script leaves it untouched and exits (nothing to fetch).
 #   - If a correct symlink already exists, it exits.
@@ -40,6 +48,49 @@ PKG="$1"
 SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 ROOT="$(cd "$SCRIPT_DIR/.." && pwd)"
 
+TARGET="$ROOT/internal/deepdoc/$PKG/testdata"
+
+# Determine whether we need a writable copy (regeneration) or a symlink.
+NEED_WRITE=0
+for v in "${!GEN_@}"; do
+  if [ -n "${!v:-}" ]; then NEED_WRITE=1; break; fi
+done
+
+# Pre-seeded testdata: CI runners mount the asset repo at a fixed path instead
+# of cloning it. Use that copy directly when it carries this package's subtree.
+# This runs before the ref is resolved: a pre-seeded copy carries no ref, so
+# packages without a testdata.ref can be served this way too.
+if [ -n "${RAGFLOW_TESTDATA_DIR:-}" ]; then
+  PRESET="$RAGFLOW_TESTDATA_DIR/deepdoc/$PKG/testdata"
+  if [ -d "$PRESET" ] && [ -n "$(ls -A "$PRESET" 2>/dev/null)" ]; then
+    if [ -L "$TARGET" ] && [ "$(readlink -f "$TARGET")" = "$(readlink -f "$PRESET")" ] && [ -n "$(ls -A "$PRESET" 2>/dev/null)" ]; then
+      echo "fetch_deepdoc_testdata: $PKG already linked to pre-seeded $PRESET"
+      exit 0
+    fi
+    if [ -d "$TARGET" ] && [ ! -L "$TARGET" ] && [ -n "$(ls -A "$TARGET" 2>/dev/null)" ]; then
+      echo "fetch_deepdoc_testdata: $PKG testdata already present inline at $TARGET"
+      exit 0
+    fi
+    # rm -rf, not rm -f: TARGET may be a leftover EMPTY directory (a failed
+    # GEN_* copy, a half-restored cache). rm -f cannot remove it, and ln -s
+    # would then nest the link inside it (<TARGET>/testdata) while this script
+    # still reports success. Both guards above have already returned for a
+    # non-empty real dir, so nothing real is deleted here.
+    rm -rf -- "$TARGET"
+    if [ "$NEED_WRITE" -eq 1 ]; then
+      echo "fetch_deepdoc_testdata: copying writable pre-seeded testdata for regeneration ($PKG)"
+      rm -rf "$TARGET"
+      cp -r "$PRESET" "$TARGET"
+    else
+      echo "fetch_deepdoc_testdata: linking $TARGET -> $PRESET (pre-seeded)"
+      ln -s "$PRESET" "$TARGET"
+    fi
+    echo "fetch_deepdoc_testdata: done ($PKG, pre-seeded)"
+    exit 0
+  fi
+  echo "fetch_deepdoc_testdata: RAGFLOW_TESTDATA_DIR set but deepdoc/$PKG/testdata is missing; falling back to clone" >&2
+fi
+
 REF_FILE="$ROOT/internal/deepdoc/$PKG/testdata.ref"
 REF="${RAGFLOW_TESTDATA_REF:-}"
 if [ -z "$REF" ] && [ -f "$REF_FILE" ]; then
@@ -54,13 +105,6 @@ REPO="${RAGFLOW_TESTDATA_REPO:-infiniflow/ragflow-testdata}"
 CACHE_BASE="${XDG_CACHE_HOME:-$HOME/.cache}/ragflow-testdata"
 CACHE="$CACHE_BASE/$REF"
 SRC="$CACHE/deepdoc/$PKG/testdata"
-TARGET="$ROOT/internal/deepdoc/$PKG/testdata"
-
-# Determine whether we need a writable copy (regeneration) or a symlink.
-NEED_WRITE=0
-for v in "${!GEN_@}"; do
-  if [ -n "${!v:-}" ]; then NEED_WRITE=1; break; fi
-done
 
 # Already a correct symlink -> done.
 if [ -L "$TARGET" ] && [ "$(readlink -f "$TARGET")" = "$(readlink -f "$SRC")" ] && [ -n "$(ls -A "$SRC" 2>/dev/null)" ]; then
@@ -75,7 +119,9 @@ if [ -d "$TARGET" ] && [ ! -L "$TARGET" ] && [ -n "$(ls -A "$TARGET" 2>/dev/null
 fi
 
 # Absent (or stale symlink) -> fetch.
-rm -f "$TARGET" 2>/dev/null || true
+# rm -rf for the same reason as the pre-seeded branch: a leftover empty
+# directory must be removed, not silently turned into a nested symlink.
+rm -rf -- "$TARGET"
 
 if [ ! -e "$SRC" ] || [ -z "$(ls -A "$SRC" 2>/dev/null)" ]; then
   echo "fetch_deepdoc_testdata: cloning $REPO @ $REF (subtree deepdoc/$PKG/testdata)"