Create a new merge() function, and use that for concatenate

It appears as if `union` wasn't really the right underlying function to use for `concatenate`, e.g. it doesn't deal as expected with sites above samples, so I had to roll my own `merge` utility.

Author: hyanwong

docs/python-api.md

@@ -261,6 +261,7 @@ which perform the same actions but modify the {class}`TableCollection` in place.
 .. autosummary::
   TreeSequence.simplify
   TreeSequence.subset
+  TreeSequence.merge
   TreeSequence.union
   TreeSequence.concatenate
   TreeSequence.keep_intervals
@@ -753,6 +754,7 @@ a functional way, returning a new tree sequence while leaving the original uncha
   TableCollection.delete_sites
   TableCollection.trim
   TableCollection.shift
+  TableCollection.merge
   TableCollection.union
   TableCollection.delete_older
 ```

python/CHANGELOG.rst

@@ -7,12 +7,19 @@
 - ``TreeSequence.map_to_vcf_model`` now also returns the transformed positions and
   contig length. (:user:`benjeffery`, :pr:`XXXX`, :issue:`3173`)
 
+- New ``merge`` functions for tree sequences and table collections, to merge another
+  into the current one (:user:`hyanwong`, :pr:`3183`, :issue:`3181`)
+
 **Bugfixes**
 
 - Fix bug in ``TreeSequence.pair_coalescence_counts`` when ``span_normalise=True``
   and a window breakpoint falls within an internal missing interval.
   (:user:`nspope`, :pr:`3176`, :issue:`3175`)
 
+- Change ``TreeSequence.concatenate`` to use ``merge``, as ``union`` does not
+  port edges, sites, or mutations from the added tree sequences if they are associated
+  with shared nodes (:user:`hyanwong`, :pr:`3183`, :issue:`3168`, :issue:`3182`)
+
 --------------------
 [0.6.4] - 2025-05-21
 --------------------

python/tests/test_topology.py

@@ -27,6 +27,7 @@
 import io
 import itertools
 import json
+import platform
 import random
 import sys
 import unittest
@@ -43,6 +44,9 @@
 import tskit.provenance as provenance
 
 
+IS_WINDOWS = platform.system() == "Windows"
+
+
 def simple_keep_intervals(tables, intervals, simplify=True, record_provenance=True):
     """
     Simple Python implementation of keep_intervals.
@@ -7141,18 +7145,223 @@ def test_bad_seq_len(self):
             ts.shift(1, sequence_length=1)
 
 
+class TestMerge:
+    def test_empty(self):
+        ts = tskit.TableCollection(2).tree_sequence()
+        merged_ts = ts.merge(ts, node_mapping=[])
+        assert merged_ts.num_nodes == 0
+        assert merged_ts.num_edges == 0
+        assert merged_ts.sequence_length == 2
+
+    def test_overlay(self):
+        ts1 = tskit.Tree.generate_balanced(4, span=2).tree_sequence
+        tables = tskit.Tree.generate_comb(4, span=2).tree_sequence.dump_tables()
+        tables.populations.add_row()
+        tables.nodes[5] = tables.nodes[5].replace(
+            flags=tskit.NODE_IS_SAMPLE, population=0
+        )
+        ts2 = tables.tree_sequence()
+        ts_merge = ts1.merge(ts2, node_mapping=np.full(ts1.num_nodes, tskit.NULL))
+        assert ts_merge.sequence_length == ts1.sequence_length
+        assert ts_merge.num_samples == ts1.num_samples + ts2.num_samples
+        assert ts_merge.num_nodes == ts1.num_nodes + ts2.num_nodes
+        assert ts_merge.num_edges == ts1.num_edges + ts2.num_edges
+        assert ts_merge.num_trees == 1
+        assert ts_merge.num_populations == 1
+        assert ts_merge.first().num_roots == 2
+
+    def test_split_and_merge(self):
+        # Cut up a single tree into alternating edges and mutations, then merge
+        ts = tskit.Tree.generate_comb(4, span=10).tree_sequence
+        ts = msprime.sim_mutations(ts, rate=0.1, random_seed=1)
+        mut_counts = np.bincount(ts.mutations_site, minlength=ts.num_sites)
+        assert min(mut_counts) == 1
+        assert max(mut_counts) > 1
+        tables1 = ts.dump_tables()
+        tables1.mutations.clear()
+        tables2 = tables1.copy()
+        i = 0
+        for s in ts.sites():
+            for m in s.mutations:
+                i += 1
+                if i % 2:
+                    tables1.mutations.append(m.replace(parent=tskit.NULL))
+                else:
+                    tables2.mutations.append(m.replace(parent=tskit.NULL))
+        tables1.simplify()
+        tables2.simplify()
+        assert tables1.sites.num_rows != ts.num_sites
+        tables1.edges.clear()
+        tables2.edges.clear()
+        for e in ts.edges():
+            if e.id % 2:
+                tables1.edges.append(e)
+            else:
+                tables2.edges.append(e)
+        ts1 = tables1.tree_sequence()
+        ts2 = tables2.tree_sequence()
+        new_ts = ts1.merge(ts2, node_mapping=np.arange(ts.num_nodes)).simplify()
+        assert new_ts.equals(ts, ignore_provenance=True)
+
+    def test_multi_tree(self):
+        ts = msprime.sim_ancestry(
+            2, sequence_length=4, recombination_rate=1, random_seed=1
+        )
+        ts = msprime.sim_mutations(ts, rate=1, random_seed=1)
+        assert ts.num_trees > 3
+        assert ts.num_mutations > 4
+        ts1 = ts.keep_intervals([[0, 1.5]], simplify=False)
+        ts2 = ts.keep_intervals([[1.5, 4]], simplify=False)
+        new_ts = ts1.merge(
+            ts2, node_mapping=np.arange(ts.num_nodes), add_populations=False
+        )
+        assert new_ts.num_trees == ts.num_trees + 1
+        new_ts = new_ts.simplify()
+        new_ts.equals(ts, ignore_provenance=True)
+
+    def test_new_individuals(self):
+        ts1 = msprime.sim_ancestry(2, sequence_length=1, random_seed=1)
+        ts2 = msprime.sim_ancestry(2, sequence_length=1, random_seed=2)
+        tables = ts2.dump_tables()
+        tables.edges.clear()
+        ts2 = tables.tree_sequence()
+        node_map = np.full(ts2.num_nodes, tskit.NULL)
+        node_map[0:2] = [0, 1]  # map first two nodes to themselves
+        ts_merged = ts1.merge(ts2, node_mapping=node_map)
+        assert ts_merged.num_nodes == ts1.num_nodes + ts2.num_nodes - 2
+        assert ts1.num_individuals == 2
+        assert ts_merged.num_individuals == 3
+
+    def test_popcheck(self):
+        tables = tskit.TableCollection(1)
+        p1 = tables.populations.add_row(b"foo")
+        p2 = tables.populations.add_row(b"bar")
+        tables.nodes.add_row(time=0, flags=tskit.NODE_IS_SAMPLE, population=p1)
+        tables.nodes.add_row(time=0, flags=tskit.NODE_IS_SAMPLE, population=p2)
+        ts1 = tables.tree_sequence()
+        tables.populations[0] = tables.populations[0].replace(metadata=b"baz")
+        ts2 = tables.tree_sequence()
+        with pytest.raises(ValueError, match="Non-matching populations"):
+            ts1.merge(ts2, node_mapping=[0, 1])
+        ts1.merge(ts2, node_mapping=[0, 1], check_populations=False)
+        # Check with add_populations=False
+        ts1.merge(ts2, node_mapping=[-1, 1])  # only merge the last one
+        with pytest.raises(ValueError, match="Non-matching populations"):
+            ts1.merge(ts2, node_mapping=[-1, 1], add_populations=False)
+
+        with pytest.raises(ValueError, match="Non-matching populations"):
+            ts1.simplify([0]).merge(ts2, node_mapping=[-1, 1])
+
+    def test_isolated_mutations(self):
+        tables = tskit.TableCollection(1)
+        u = tables.nodes.add_row(time=0, flags=tskit.NODE_IS_SAMPLE)
+        s = tables.sites.add_row(0.5, "A")
+        tables.mutations.add_row(s, u, derived_state="T", time=1, metadata=b"xxx")
+        ts1 = tables.tree_sequence()
+        tables.mutations[0] = tables.mutations[0].replace(time=0.5, metadata=b"yyy")
+        ts2 = tables.tree_sequence()
+        ts_merge = ts1.merge(ts2, node_mapping=[0])
+        assert ts_merge.num_sites == 1
+        assert ts_merge.num_mutations == 2
+        assert ts_merge.mutation(0).time == 1
+        assert ts_merge.mutation(0).parent == tskit.NULL
+        assert ts_merge.mutation(0).metadata == b"xxx"
+        assert ts_merge.mutation(1).time == 0.5
+        assert ts_merge.mutation(1).parent == 0
+        assert ts_merge.mutation(1).metadata == b"yyy"
+
+    def test_identity(self):
+        tables = tskit.TableCollection(1)
+        tables.nodes.add_row(time=0, flags=tskit.NODE_IS_SAMPLE)
+        ts = tables.tree_sequence()
+        ts_merge = ts.merge(ts, node_mapping=[0])
+        assert ts.equals(ts_merge, ignore_provenance=True)
+
+    @pytest.mark.skipif(IS_WINDOWS, reason="Msprime gives different result on Windows")
+    def test_migrations(self):
+        pop_configs = [msprime.PopulationConfiguration(3) for _ in range(2)]
+        migration_matrix = [[0, 0.001], [0.001, 0]]
+        ts = msprime.simulate(
+            population_configurations=pop_configs,
+            migration_matrix=migration_matrix,
+            record_migrations=True,
+            recombination_rate=2,
+            random_seed=42,  # pick a seed that gives min(migrations.left) > 0
+            end_time=100,
+        )
+        # No migration_table.squash() function exists, so we just try to cut on the
+        # LHS of all the migrations
+        assert ts.num_migrations > 0
+        assert ts.migrations_left.min() > 0
+        cutpoint = ts.migrations_left.min()
+        ts1 = ts.keep_intervals([[0, cutpoint]], simplify=False)
+        ts2 = ts.keep_intervals([[cutpoint, ts.sequence_length]], simplify=False)
+        ts_new = ts1.merge(ts2, node_mapping=np.arange(ts.num_nodes))
+        tables = ts_new.dump_tables()
+        tables.edges.squash()
+        tables.sort()
+        ts_new = tables.tree_sequence()
+        ts.tables.assert_equals(ts_new.tables, ignore_provenance=True)
+
+    def test_provenance(self):
+        tables = tskit.TableCollection(1)
+        tables.nodes.add_row(time=0, flags=tskit.NODE_IS_SAMPLE)
+        ts = tables.tree_sequence()
+        ts_merge = ts.merge(ts, node_mapping=[0], record_provenance=False)
+        assert ts_merge.num_provenances == ts.num_provenances
+        ts_merge = ts.merge(ts, node_mapping=[0])
+        assert ts_merge.num_provenances == ts.num_provenances + 1
+        prov = json.loads(ts_merge.provenance(-1).record)
+        assert prov["parameters"]["command"] == "merge"
+        assert prov["parameters"]["node_mapping"] == [0]
+        assert prov["parameters"]["add_populations"] is True
+        assert prov["parameters"]["check_populations"] is True
+
+    def test_bad_sequence_length(self):
+        ts1 = tskit.TableCollection(1).tree_sequence()
+        ts2 = tskit.TableCollection(2).tree_sequence()
+        with pytest.raises(ValueError, match="sequence length"):
+            ts1.merge(ts2, node_mapping=[])
+
+    def test_bad_node_mapping(self):
+        ts = tskit.Tree.generate_comb(5).tree_sequence
+        with pytest.raises(ValueError, match="node_mapping"):
+            ts.merge(ts, node_mapping=[0, 1, 2])
+
+    def test_bad_populations(self):
+        tables = tskit.TableCollection(1)
+        tables = tskit.TableCollection(1)
+        p1 = tables.populations.add_row()
+        p2 = tables.populations.add_row()
+        tables.nodes.add_row(time=0, flags=tskit.NODE_IS_SAMPLE, population=p1)
+        tables.nodes.add_row(time=0, flags=tskit.NODE_IS_SAMPLE, population=p1)
+        tables.nodes.add_row(time=0, flags=tskit.NODE_IS_SAMPLE, population=p2)
+        ts2 = tables.tree_sequence()
+        ts1 = ts2.simplify([0, 1])
+        assert ts1.num_populations == 1
+        assert ts2.num_populations == 2
+        ts2.merge(ts1, [0, -1], check_populations=False, add_populations=False)
+        with pytest.raises(ValueError, match="population not present"):
+            ts1.merge(ts2, [0, -1, -1], check_populations=False, add_populations=False)
+
+
 class TestConcatenate:
     def test_simple(self):
         ts1 = tskit.Tree.generate_comb(5, span=2).tree_sequence
+        ts1 = msprime.sim_mutations(ts1, rate=1, random_seed=1)
         ts2 = tskit.Tree.generate_balanced(5, arity=3, span=3).tree_sequence
+        ts2 = msprime.sim_mutations(ts2, rate=1, random_seed=1)
         assert ts1.num_samples == ts2.num_samples
         assert ts1.num_nodes != ts2.num_nodes
         joint_ts = ts1.concatenate(ts2)
         assert joint_ts.num_nodes == ts1.num_nodes + ts2.num_nodes - 5
         assert joint_ts.sequence_length == ts1.sequence_length + ts2.sequence_length
         assert joint_ts.num_samples == ts1.num_samples
+        assert joint_ts.num_sites == ts1.num_sites + ts2.num_sites
+        assert joint_ts.num_mutations == ts1.num_mutations + ts2.num_mutations
         ts3 = joint_ts.delete_intervals([[2, 5]]).rtrim()
         # Have to simplify here, to remove the redundant nodes
+        ts3.tables.assert_equals(ts1.tables, ignore_provenance=True)
         assert ts3.equals(ts1.simplify(), ignore_provenance=True)
         ts4 = joint_ts.delete_intervals([[0, 2]]).ltrim()
         assert ts4.equals(ts2.simplify(), ignore_provenance=True)
@@ -7183,6 +7392,13 @@ def test_empty(self):
         assert ts.num_nodes == 0
         assert ts.sequence_length == 40
 
+    def test_check_populations(self):
+        ts = msprime.sim_ancestry(2)
+        ts1 = ts.concatenate(ts, ts, check_populations=True)
+        assert ts1.num_populations == 1
+        ts2 = ts.concatenate(ts, ts, add_populations=True, check_populations=True)
+        assert ts2.num_populations == 3
+
     def test_samples_at_end(self):
         ts1 = tskit.Tree.generate_comb(5, span=2).tree_sequence
         ts2 = tskit.Tree.generate_balanced(5, arity=3, span=3).tree_sequence
@@ -7200,22 +7416,58 @@ def test_internal_samples(self):
         nodes_flags[:] = tskit.NODE_IS_SAMPLE
         nodes_flags[-1] = 0  # Only root is not a sample
         tables.nodes.flags = nodes_flags
-        ts = tables.tree_sequence()
+        ts = msprime.sim_mutations(tables.tree_sequence(), rate=0.5, random_seed=1)
+        assert ts.num_mutations > 0
+        assert ts.num_mutations > ts.num_sites
         joint_ts = ts.concatenate(ts)
         assert joint_ts.num_samples == ts.num_samples
         assert joint_ts.num_nodes == ts.num_nodes + 1
+        assert joint_ts.num_mutations == ts.num_mutations * 2
+        assert joint_ts.num_sites == ts.num_sites * 2
         assert joint_ts.sequence_length == ts.sequence_length * 2
 
     def test_some_shared_samples(self):
-        ts1 = tskit.Tree.generate_comb(4, span=2).tree_sequence
-        ts2 = tskit.Tree.generate_balanced(8, arity=3, span=3).tree_sequence
-        shared = np.full(ts2.num_nodes, tskit.NULL)
-        shared[0] = 1
-        shared[1] = 0
-        joint_ts = ts1.concatenate(ts2, node_mappings=[shared])
-        assert joint_ts.sequence_length == ts1.sequence_length + ts2.sequence_length
-        assert joint_ts.num_samples == ts1.num_samples + ts2.num_samples - 2
-        assert joint_ts.num_nodes == ts1.num_nodes + ts2.num_nodes - 2
+        tables = tskit.Tree.generate_comb(5).tree_sequence.dump_tables()
+        tables.nodes[5] = tables.nodes[5].replace(flags=tskit.NODE_IS_SAMPLE)
+        ts1 = tables.tree_sequence()
+        tables = tskit.Tree.generate_balanced(5).tree_sequence.dump_tables()
+        tables.nodes[5] = tables.nodes[5].replace(flags=tskit.NODE_IS_SAMPLE)
+        ts2 = tables.tree_sequence()
+        assert ts1.num_samples == ts2.num_samples
+        joint_ts = ts1.concatenate(ts2)
+        assert joint_ts.num_samples == ts1.num_samples
+        assert joint_ts.num_edges == ts1.num_edges + ts2.num_edges
+        for tree in joint_ts.trees():
+            assert tree.num_roots == 1
+
+    @pytest.mark.parametrize("simplify", [True, False])
+    def test_wf_sim(self, simplify):
+        # Test that we can split & concat a wf_sim ts, which has internal samples
+        tables = wf.wf_sim(
+            6,
+            5,
+            seed=3,
+            deep_history=True,
+            initial_generation_samples=True,
+            num_loci=10,
+        )
+        tables.sort()
+        tables.simplify()
+        ts = msprime.mutate(tables.tree_sequence(), rate=0.05, random_seed=234)
+        assert ts.num_trees > 2
+        assert len(np.unique(ts.nodes_time[ts.samples()])) > 1
+        ts1 = ts.keep_intervals([[0, 4.5]], simplify=False).trim()
+        ts2 = ts.keep_intervals([[4.5, ts.sequence_length]], simplify=False).trim()
+        if simplify:
+            ts1 = ts1.simplify(filter_nodes=False)
+            ts2, node_map = ts2.simplify(map_nodes=True)
+            node_mapping = np.zeros_like(node_map, shape=ts2.num_nodes)
+            kept = node_map != tskit.NULL
+            node_mapping[node_map[kept]] = np.arange(len(node_map))[kept]
+        else:
+            node_mapping = np.arange(ts.num_nodes)
+        ts_new = ts1.concatenate(ts2, node_mappings=[node_mapping]).simplify()
+        ts_new.tables.assert_equals(ts.tables, ignore_provenance=True)
 
     def test_provenance(self):
         ts = tskit.Tree.generate_comb(2).tree_sequence
@@ -7233,9 +7485,12 @@ def test_unequal_samples(self):
         with pytest.raises(ValueError, match="must have the same number of samples"):
             ts1.concatenate(ts2)
 
-    @pytest.mark.skip(
-        reason="union bug: https://github.com/tskit-dev/tskit/issues/3168"
-    )
+    def test_different_sample_numbers(self):
+        ts1 = tskit.Tree.generate_comb(5, span=2).tree_sequence
+        ts2 = tskit.Tree.generate_balanced(4, arity=3, span=3).tree_sequence
+        with pytest.raises(ValueError, match="must have the same number of samples"):
+            ts1.concatenate(ts2)
+
     def test_duplicate_ts(self):
         ts1 = tskit.Tree.generate_comb(3, span=4).tree_sequence
         ts = ts1.keep_intervals([[0, 1]]).trim()  # a quarter of the original