Read hdf returns unexpected values for categorical (#39420)

Author: nofarm3

Diff for: doc/source/whatsnew/v1.3.0.rst

@@ -375,6 +375,7 @@ I/O
 - Bug in :func:`read_csv` not switching ``true_values`` and ``false_values`` for nullable ``boolean`` dtype (:issue:`34655`)
 - Bug in :func:`read_json` when ``orient="split"`` does not maintain numeric string index (:issue:`28556`)
 - :meth:`read_sql` returned an empty generator if ``chunksize`` was no-zero and the query returned no results. Now returns a generator with a single empty dataframe (:issue:`34411`)
+- Bug in :func:`read_hdf` returning unexpected records when filtering on categorical string columns using ``where`` parameter (:issue:`39189`)
 
 Period
 ^^^^^^

Diff for: pandas/core/computation/pytables.py

@@ -210,12 +210,10 @@ def stringify(value):
             return TermValue(int(v), v, kind)
         elif meta == "category":
             metadata = extract_array(self.metadata, extract_numpy=True)
-            result = metadata.searchsorted(v, side="left")
-
-            # result returns 0 if v is first element or if v is not in metadata
-            # check that metadata contains v
-            if not result and v not in metadata:
+            if v not in metadata:
                 result = -1
+            else:
+                result = metadata.searchsorted(v, side="left")
             return TermValue(result, result, "integer")
         elif kind == "integer":
             v = int(float(v))

Diff for: pandas/tests/io/pytables/test_categorical.py

@@ -184,3 +184,25 @@ def test_categorical_nan_only_columns(setup_path):
         df.to_hdf(path, "df", format="table", data_columns=True)
         result = read_hdf(path, "df")
         tm.assert_frame_equal(result, expected)
+
+
+@pytest.mark.parametrize(
+    "where, df, expected",
+    [
+        ('col=="q"', DataFrame({"col": ["a", "b", "s"]}), DataFrame({"col": []})),
+        ('col=="a"', DataFrame({"col": ["a", "b", "s"]}), DataFrame({"col": ["a"]})),
+    ],
+)
+def test_convert_value(setup_path, where: str, df: DataFrame, expected: DataFrame):
+    # GH39420
+    # Check that read_hdf with categorical columns can filter by where condition.
+    df.col = df.col.astype("category")
+    max_widths = {"col": 1}
+    categorical_values = sorted(df.col.unique())
+    expected.col = expected.col.astype("category")
+    expected.col.cat.set_categories(categorical_values, inplace=True)
+
+    with ensure_clean_path(setup_path) as path:
+        df.to_hdf(path, "df", format="table", min_itemsize=max_widths)
+        result = read_hdf(path, where=where)
+        tm.assert_frame_equal(result, expected)