Fix Tabular handling of explicitly selected falsy ID columns (#1332)

src/python/txtai/pipeline/data/tabular.py

@@ -99,14 +99,14 @@ class Tabular(Pipeline):
         columns = self.textcolumns
         if not columns:
             columns = list(df.columns)
-            if self.idcolumn:
+            if self.idcolumn is not None:
                 columns.remove(self.idcolumn)
 
         # Transform into (id, text, tag) tuples
         # Tuple iteration preserves column types instead of coercing each row to one dtype.
         for index, values in zip(df.index, df.itertuples(index=False, name=None)):
             row = dict(zip(df.columns, values))
-            uid = row[self.idcolumn] if self.idcolumn else index
+            uid = row[self.idcolumn] if self.idcolumn is not None else index
             uid = uid if uid is not None else index
             text = self.concat(row, columns)
 

test/python/testpipeline/testdata/testtabular.py

@@ -157,6 +157,16 @@ class TestTabular(unittest.TestCase):
 
         self.assertIsNone(data["metadata"])
 
+    def testFalsyIDColumn(self):
+        """
+        Explicit column labels are used for IDs and excluded from default text.
+        """
+
+        for column in ("", 0, "id"):
+            with self.subTest(column=column):
+                rows = Tabular(idcolumn=column)([{column: "doc-7", "text": "hello"}])
+                self.assertEqual(rows, [("doc-7", "hello", None)])
+
     def testNoColumns(self):
         """
         Test creating text without specifying columns