Handle more cases of empty string arrays

searcharray/phrase/middle_out.py

@@ -302,7 +302,8 @@ def __init__(self, encoded_term_posns: Union[ArrayDict, FilteredPosns], max_doc_
         self.termfreq_cache : Dict[int, Tuple[np.ndarray, np.ndarray]] = {}
 
     def memmap(self, data_dir):
-        self.encoded_term_posns = MemoryMappedArrays(data_dir, self.encoded_term_posns)
+        if self.encoded_term_posns:
+            self.encoded_term_posns = MemoryMappedArrays(data_dir, self.encoded_term_posns)
 
     def warm(self):
         """Warm tf / df cache of most common terms."""

test/test_search.py

@@ -1,12 +1,17 @@
 """Test postings array search functionality."""
 import numpy as np
 import pandas as pd
+import os
+import shutil
 import pytest
 from searcharray.postings import SearchArray
 from searcharray.similarity import bm25_similarity
 from test_utils import w_scenarios
 
 
+DATA_DIR = '/tmp/tmdb'
+
+
 @pytest.fixture
 def data():
     """Return a fixture of your data here that returns an instance of your ExtensionArray."""
@@ -33,6 +38,20 @@ def test_search_phrase_empty_str_batch_size():
     assert data.score(["foo", "bar"]).sum() == 0
 
 
+def test_search_phrase_empty_str_batch_size_memmap():
+    os.makedirs(DATA_DIR, exist_ok=True)
+    data = pd.DataFrame({"data": [""] * 10000})
+    data = SearchArray.index(data["data"],
+                             batch_size=1000,
+                             data_dir=DATA_DIR)
+    assert data.score(["foo", "bar"]).sum() == 0
+    pd.to_pickle(data, os.path.join(DATA_DIR, "data.pkl"))
+    reloaded = pd.read_pickle(os.path.join(DATA_DIR, "data.pkl"))
+    assert reloaded.score(["foo", "bar"]).sum() == 0
+
+    shutil.rmtree(DATA_DIR)
+
+
 def test_match(data):
     matches = data.termfreqs("foo") > 0
     assert (matches == [True, False, False, False] * 25).all()