datachain 0.36.2__tar.gz → 0.36.4__tar.gz

{datachain-0.36.2 → datachain-0.36.4}/.github/workflows/tests.yml

@@ -110,7 +110,7 @@ jobs:
 
       - name: Install FFmpeg on Windows
         if: runner.os == 'Windows'
-        run: choco install ffmpeg
+        run: choco install ffmpeg --version 7.0
 
       - name: Install FFmpeg on macOS
         if: runner.os == 'macOS'

{datachain-0.36.2 → datachain-0.36.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: datachain
-Version: 0.36.2
+Version: 0.36.4
 Summary: Wrangle unstructured AI data at scale
 Author-email: Dmitry Petrov <support@dvc.org>
 License-Expression: Apache-2.0
@@ -76,6 +76,7 @@ Requires-Dist: datasets[vision]>=4.0.0; extra == "hf"
 Requires-Dist: datasets[audio]>=4.0.0; (sys_platform == "linux" or sys_platform == "darwin") and extra == "hf"
 Requires-Dist: fsspec>=2024.12.0; extra == "hf"
 Requires-Dist: torch<2.9.0; extra == "hf"
+Requires-Dist: torchcodec<0.8.0; (sys_platform == "linux" or sys_platform == "darwin") and extra == "hf"
 Provides-Extra: video
 Requires-Dist: ffmpeg-python; extra == "video"
 Requires-Dist: imageio[ffmpeg,pyav]>=2.37.0; extra == "video"

{datachain-0.36.2 → datachain-0.36.4}/pyproject.toml

@@ -91,7 +91,8 @@ hf = [
   # Until datasets solve the issue, run test_hf_audio test to see if this can be removed
   # https://github.com/meta-pytorch/torchcodec/issues/912
   # https://github.com/huggingface/transformers/pull/41610
-  "torch<2.9.0"
+  "torch<2.9.0",
+  "torchcodec<0.8.0 ; (sys_platform == 'linux' or sys_platform == 'darwin')"
 ]
 video = [
   "ffmpeg-python",
@@ -139,6 +140,8 @@ examples = [
   "open_clip_torch",
   "openai",
   # Transformers still require it
+  # Remove this after HF tests and example work w/o these restrictions,
+  # make them >= probably
   "torchaudio<2.9.0"
 ]
 

{datachain-0.36.2 → datachain-0.36.4}/src/datachain/query/dataset.py

@@ -231,8 +231,9 @@ class DatasetDiffOperation(Step):
 
     def apply(self, query_generator, temp_tables: list[str]) -> "StepResult":
         source_query = query_generator.exclude(("sys__id",))
+        right_before = len(self.dq.temp_table_names)
         target_query = self.dq.apply_steps().select()
-        temp_tables.extend(self.dq.temp_table_names)
+        temp_tables.extend(self.dq.temp_table_names[right_before:])
 
         # creating temp table that will hold subtract results
         temp_table_name = self.catalog.warehouse.temp_table_name()
@@ -951,10 +952,12 @@ class SQLUnion(Step):
     def apply(
         self, query_generator: QueryGenerator, temp_tables: list[str]
     ) -> StepResult:
+        left_before = len(self.query1.temp_table_names)
         q1 = self.query1.apply_steps().select().subquery()
-        temp_tables.extend(self.query1.temp_table_names)
+        temp_tables.extend(self.query1.temp_table_names[left_before:])
+        right_before = len(self.query2.temp_table_names)
         q2 = self.query2.apply_steps().select().subquery()
-        temp_tables.extend(self.query2.temp_table_names)
+        temp_tables.extend(self.query2.temp_table_names[right_before:])
 
         columns1 = _drop_system_columns(q1.columns)
         columns2 = _drop_system_columns(q2.columns)
@@ -1004,8 +1007,9 @@ class SQLJoin(Step):
         return hashlib.sha256(b"".join(parts)).hexdigest()
 
     def get_query(self, dq: "DatasetQuery", temp_tables: list[str]) -> sa.Subquery:
+        temp_tables_before = len(dq.temp_table_names)
         query = dq.apply_steps().select()
-        temp_tables.extend(dq.temp_table_names)
+        temp_tables.extend(dq.temp_table_names[temp_tables_before:])
 
         if not any(isinstance(step, (SQLJoin, SQLUnion)) for step in dq.steps):
             return query.subquery(dq.table.name)

{datachain-0.36.2 → datachain-0.36.4}/src/datachain.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: datachain
-Version: 0.36.2
+Version: 0.36.4
 Summary: Wrangle unstructured AI data at scale
 Author-email: Dmitry Petrov <support@dvc.org>
 License-Expression: Apache-2.0
@@ -76,6 +76,7 @@ Requires-Dist: datasets[vision]>=4.0.0; extra == "hf"
 Requires-Dist: datasets[audio]>=4.0.0; (sys_platform == "linux" or sys_platform == "darwin") and extra == "hf"
 Requires-Dist: fsspec>=2024.12.0; extra == "hf"
 Requires-Dist: torch<2.9.0; extra == "hf"
+Requires-Dist: torchcodec<0.8.0; (sys_platform == "linux" or sys_platform == "darwin") and extra == "hf"
 Provides-Extra: video
 Requires-Dist: ffmpeg-python; extra == "video"
 Requires-Dist: imageio[ffmpeg,pyav]>=2.37.0; extra == "video"

{datachain-0.36.2 → datachain-0.36.4}/src/datachain.egg-info/SOURCES.txt

@@ -331,6 +331,7 @@ tests/func/test_retry.py
 tests/func/test_session.py
 tests/func/test_storage_pattern.py
 tests/func/test_studio_datetime_parsing.py
+tests/func/test_temp_table_tracking.py
 tests/func/test_to_database.py
 tests/func/test_toolkit.py
 tests/func/test_udf.py

{datachain-0.36.2 → datachain-0.36.4}/src/datachain.egg-info/requires.txt

@@ -76,6 +76,7 @@ torch<2.9.0
 
 [hf:sys_platform == "linux" or sys_platform == "darwin"]
 datasets[audio]>=4.0.0
+torchcodec<0.8.0
 
 [postgres]
 psycopg2-binary>=2.9.0

{datachain-0.36.2 → datachain-0.36.4}/tests/func/test_hf.py

@@ -27,7 +27,7 @@ def require_torchcodec(test_case):
     """
     if not importlib.util.find_spec("torchcodec"):
         test_case = pytest.mark.skip(
-            "test requires torchcoded, not available on Windows yet"
+            "test requires torchcodec, not available on Windows yet"
         )(test_case)
     return test_case
 

datachain-0.36.4/tests/func/test_temp_table_tracking.py

@@ -0,0 +1,68 @@
+import datachain as dc
+from datachain.query.dataset import DatasetQuery
+
+
+def _capture_temp_tables(mocker):
+    captured: list[list[str]] = []
+    original_cleanup = DatasetQuery.cleanup
+
+    def capture(self):
+        captured.append(list(self.temp_table_names))
+        return original_cleanup(self)
+
+    mocker.patch("datachain.query.dataset.DatasetQuery.cleanup", capture)
+    return captured
+
+
+def _assert_no_duplicate_temp_tables(captured: list[list[str]]):
+    for tables in captured:
+        assert len(tables) == len(set(tables))
+
+
+def test_nested_merge_has_no_duplicate_temp_tables(test_session, mocker):
+    captured = _capture_temp_tables(mocker)
+
+    base = dc.read_values(num=[1, 2], session=test_session)
+    generated = base.map(num_plus=lambda num: str(num + 10))
+    inner = generated.merge(base, on="num", inner=True)
+    chain = base.merge(inner, on="num", inner=True)
+
+    expected = chain.select("num").to_pandas()["num"].tolist()
+    assert expected == [1, 2]
+
+    rerun = chain.select("num").to_pandas()["num"].tolist()
+    assert rerun == expected
+
+    _assert_no_duplicate_temp_tables(captured)
+
+
+def test_union_has_no_duplicate_temp_tables(test_session, mocker):
+    captured = _capture_temp_tables(mocker)
+
+    left = dc.read_values(num=[1, 2], session=test_session)
+    right = dc.read_values(num=[3], session=test_session)
+    union_chain = left.union(right)
+
+    expected = sorted(union_chain.select("num").to_pandas()["num"].tolist())
+    assert expected == [1, 2, 3]
+
+    rerun = sorted(union_chain.select("num").to_pandas()["num"].tolist())
+    assert rerun == expected
+
+    _assert_no_duplicate_temp_tables(captured)
+
+
+def test_subtract_has_no_duplicate_temp_tables(test_session, mocker):
+    captured = _capture_temp_tables(mocker)
+
+    source = dc.read_values(num=[1, 2], session=test_session)
+    target = dc.read_values(num=[2], session=test_session)
+    subtract_chain = source.subtract(target, on="num")
+
+    expected = subtract_chain.select("num").to_pandas()["num"].tolist()
+    assert expected == [1]
+
+    rerun = subtract_chain.select("num").to_pandas()["num"].tolist()
+    assert rerun == expected
+
+    _assert_no_duplicate_temp_tables(captured)

{datachain-0.36.2 → datachain-0.36.4}/tests/unit/lib/test_datachain.py

@@ -1072,6 +1072,19 @@ def test_select_except(test_session):
     assert n == len(features_nested)
 
 
+def test_select_except_after_gen(test_session):
+    # https://github.com/iterative/datachain/issues/1359
+    # fixed by https://github.com/iterative/datachain/pull/1400
+    chain = dc.read_values(id=range(10), session=test_session)
+
+    chain = chain.gen(lambda id: [(id, 0)], output={"id": int, "x": int})
+    chain = chain.select_except("x")
+    chain = chain.merge(chain, on="id")
+    chain = chain.select_except("right_id")
+
+    assert set(chain.to_values("id")) == set(range(10))
+
+
 def test_select_wrong_type(test_session):
     chain = dc.read_values(fr1=features_nested, fr2=features, session=test_session)