Drop metadata schemas on tables.

Closes #2944

python/CHANGELOG.rst

@@ -8,6 +8,9 @@
   using recombination information, leading to unary nodes in many trees and
   fewer edges. (:user:`petrelharp`, :user:`hfr1tz3`, :user:`avabamf`, :pr:`2651`)
 
+- Add ``Table.drop_metadata`` to make clearing metadata from tables easy.
+  (:user:`jeromekelleher`, :pr:`2944`)
+
 **Bugfixes**
 
 - Fix to the folded, expected allele frequency spectrum (i.e.,

python/tests/test_tables.py

@@ -919,6 +919,25 @@ def test_random_metadata(self):
             )
             assert metadatas == unpacked_metadatas
 
+    def test_drop_metadata(self):
+        for num_rows in [1, 10, 100]:
+            input_data = self.make_input_data(num_rows)
+            table_no_meta = self.table_class()
+            table_with_meta = self.table_class()
+            table_with_meta.set_columns(**input_data)
+            if not getattr(self, "metadata_mandatory", False):
+                del input_data["metadata"]
+                del input_data["metadata_offset"]
+            else:
+                # Have to do this slightly circular way for the population
+                # table because it requires metadata.
+                input_data["metadata"] = []
+                input_data["metadata_offset"][:] = 0
+            table_no_meta.set_columns(**input_data)
+            assert not table_no_meta.equals(table_with_meta)
+            table_with_meta.drop_metadata()
+            table_no_meta.assert_equals(table_with_meta)
+
     def test_optional_metadata(self):
         if not getattr(self, "metadata_mandatory", False):
             for num_rows in [0, 10, 100]:
@@ -982,6 +1001,28 @@ def test_set_metadata_schema(self):
         ):
             table.metadata_schema = {}
 
+    def test_drop_metadata_with_schema(self):
+        table = self.table_class()
+        table.metadata_schema = metadata.MetadataSchema.permissive_json()
+        data = self.input_data_for_add_row()
+        data["metadata"] = {"a": "dict"}
+        table.add_row(**data)
+        assert table[0].metadata == {"a": "dict"}
+        table.drop_metadata()
+        assert table.metadata_schema == metadata.MetadataSchema.null()
+        assert table[0].metadata == b""
+
+    def test_drop_metadata_keep_schema(self):
+        table = self.table_class()
+        table.metadata_schema = metadata.MetadataSchema.permissive_json()
+        data = self.input_data_for_add_row()
+        data["metadata"] = {"a": "dict"}
+        table.add_row(**data)
+        assert table[0].metadata == {"a": "dict"}
+        table.drop_metadata(keep_schema=True)
+        assert table.metadata_schema == metadata.MetadataSchema.permissive_json()
+        assert table[0].metadata == {}
+
     def test_default_metadata_schema(self):
         # Default should allow bytes as in pre-exisiting code
         table = self.table_class()

python/tskit/metadata.py

@@ -722,6 +722,14 @@ def permissive_json():
         """
         return MetadataSchema({"codec": "json"})
 
+    @staticmethod
+    def null():
+        """
+        The null schema which defines no properties and results in raw bytes
+        being returned on accessing metadata column.
+        """
+        return MetadataSchema(None)
+
 
 # Often many replicate tree sequences are processed with identical schemas, so cache them
 @functools.lru_cache(maxsize=128)
@@ -734,7 +742,7 @@ def parse_metadata_schema(encoded_schema: str) -> MetadataSchema:
     :return: A subclass of AbstractMetadataSchema.
     """
     if encoded_schema == "":
-        return MetadataSchema(schema=None)
+        return MetadataSchema.null()
     else:
         try:
             decoded = json.loads(

python/tskit/tables.py

@@ -794,6 +794,20 @@ def getter(d, k):
             )
         return out
 
+    def drop_metadata(self, *, keep_schema=False):
+        """
+        Drops all metadata in this table. By default, the schema is also cleared,
+        except if ``keep_schema`` is True.
+
+        :param bool keep_schema: True if the current schema should be kept intact.
+        """
+        data = self.asdict()
+        data["metadata"] = []
+        data["metadata_offset"][:] = 0
+        self.set_columns(**data)
+        if not keep_schema:
+            self.metadata_schema = metadata.MetadataSchema.null()
+
 
 class IndividualTable(MetadataTable):
     """