datachain 0.34.0__tar.gz → 0.34.2__tar.gz

{datachain-0.34.0 → datachain-0.34.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: datachain
-Version: 0.34.0
+Version: 0.34.2
 Summary: Wrangle unstructured AI data at scale
 Author-email: Dmitry Petrov <support@dvc.org>
 License-Expression: Apache-2.0

{datachain-0.34.0 → datachain-0.34.2}/src/datachain/catalog/catalog.py

@@ -144,26 +144,19 @@ def shutdown_process(
             return proc.wait()
 
 
-def process_output(stream: IO[bytes], callback: Callable[[str], None]) -> None:
+def _process_stream(stream: "IO[bytes]", callback: Callable[[str], None]) -> None:
     buffer = b""
+    while byt := stream.read(1):  # Read one byte at a time
+        buffer += byt
 
-    try:
-        while byt := stream.read(1):  # Read one byte at a time
-            buffer += byt
-
-            if byt in (b"\n", b"\r"):  # Check for newline or carriage return
-                line = buffer.decode("utf-8", errors="replace")
-                callback(line)
-                buffer = b""  # Clear buffer for the next line
-
-        if buffer:  # Handle any remaining data in the buffer
-            line = buffer.decode("utf-8", errors="replace")
+        if byt in (b"\n", b"\r"):  # Check for newline or carriage return
+            line = buffer.decode("utf-8")
             callback(line)
-    finally:
-        try:
-            stream.close()  # Ensure output is closed
-        except Exception:  # noqa: BLE001, S110
-            pass
+            buffer = b""  # Clear buffer for next line
+
+    if buffer:  # Handle any remaining data in the buffer
+        line = buffer.decode("utf-8")
+        callback(line)
 
 
 class DatasetRowsFetcher(NodesThreadPool):
@@ -1767,13 +1760,13 @@ class Catalog:
             recursive=recursive,
         )
 
-    @staticmethod
     def query(
+        self,
         query_script: str,
         env: Optional[Mapping[str, str]] = None,
         python_executable: str = sys.executable,
-        stdout_callback: Optional[Callable[[str], None]] = None,
-        stderr_callback: Optional[Callable[[str], None]] = None,
+        capture_output: bool = False,
+        output_hook: Callable[[str], None] = noop,
         params: Optional[dict[str, str]] = None,
         job_id: Optional[str] = None,
         interrupt_timeout: Optional[int] = None,
@@ -1788,18 +1781,13 @@ class Catalog:
             },
         )
         popen_kwargs: dict[str, Any] = {}
-
-        if stdout_callback is not None:
-            popen_kwargs = {"stdout": subprocess.PIPE}
-        if stderr_callback is not None:
-            popen_kwargs["stderr"] = subprocess.PIPE
+        if capture_output:
+            popen_kwargs = {"stdout": subprocess.PIPE, "stderr": subprocess.STDOUT}
 
         def raise_termination_signal(sig: int, _: Any) -> NoReturn:
             raise TerminationSignal(sig)
 
-        stdout_thread: Optional[Thread] = None
-        stderr_thread: Optional[Thread] = None
-
+        thread: Optional[Thread] = None
         with subprocess.Popen(cmd, env=env, **popen_kwargs) as proc:  # noqa: S603
             logger.info("Starting process %s", proc.pid)
 
@@ -1813,20 +1801,10 @@ class Catalog:
             orig_sigterm_handler = signal.getsignal(signal.SIGTERM)
             signal.signal(signal.SIGTERM, raise_termination_signal)
             try:
-                if stdout_callback is not None:
-                    stdout_thread = Thread(
-                        target=process_output,
-                        args=(proc.stdout, stdout_callback),
-                        daemon=True,
-                    )
-                    stdout_thread.start()
-                if stderr_callback is not None:
-                    stderr_thread = Thread(
-                        target=process_output,
-                        args=(proc.stderr, stderr_callback),
-                        daemon=True,
-                    )
-                    stderr_thread.start()
+                if capture_output:
+                    args = (proc.stdout, output_hook)
+                    thread = Thread(target=_process_stream, args=args, daemon=True)
+                    thread.start()
 
                 proc.wait()
             except TerminationSignal as exc:
@@ -1844,22 +1822,8 @@ class Catalog:
             finally:
                 signal.signal(signal.SIGTERM, orig_sigterm_handler)
                 signal.signal(signal.SIGINT, orig_sigint_handler)
-                # wait for the reader thread
-                thread_join_timeout_seconds = 30
-                if stdout_thread is not None:
-                    stdout_thread.join(timeout=thread_join_timeout_seconds)
-                    if stdout_thread.is_alive():
-                        logger.warning(
-                            "stdout thread is still alive after %s seconds",
-                            thread_join_timeout_seconds,
-                        )
-                if stderr_thread is not None:
-                    stderr_thread.join(timeout=thread_join_timeout_seconds)
-                    if stderr_thread.is_alive():
-                        logger.warning(
-                            "stderr thread is still alive after %s seconds",
-                            thread_join_timeout_seconds,
-                        )
+                if thread:
+                    thread.join()  # wait for the reader thread
 
         logger.info("Process %s exited with return code %s", proc.pid, proc.returncode)
         if proc.returncode in (

{datachain-0.34.0 → datachain-0.34.2}/src/datachain/catalog/loader.py

@@ -3,6 +3,7 @@ import sys
 from importlib import import_module
 from typing import TYPE_CHECKING, Any, Optional
 
+from datachain.plugins import ensure_plugins_loaded
 from datachain.utils import get_envs_by_prefix
 
 if TYPE_CHECKING:
@@ -24,6 +25,8 @@ IN_MEMORY_ERROR_MESSAGE = "In-memory is only supported on SQLite"
 
 
 def get_metastore(in_memory: bool = False) -> "AbstractMetastore":
+    ensure_plugins_loaded()
+
     from datachain.data_storage import AbstractMetastore
     from datachain.data_storage.serializer import deserialize
 
@@ -64,6 +67,8 @@ def get_metastore(in_memory: bool = False) -> "AbstractMetastore":
 
 
 def get_warehouse(in_memory: bool = False) -> "AbstractWarehouse":
+    ensure_plugins_loaded()
+
     from datachain.data_storage import AbstractWarehouse
     from datachain.data_storage.serializer import deserialize
 

datachain-0.34.2/src/datachain/data_storage/serializer.py

@@ -0,0 +1,119 @@
+import base64
+import json
+from abc import abstractmethod
+from collections.abc import Callable
+from typing import Any, ClassVar
+
+from datachain.plugins import ensure_plugins_loaded
+
+
+class CallableRegistry:
+    _registry: ClassVar[dict[str, Callable]] = {}
+
+    @classmethod
+    def register(cls, callable_obj: Callable, name: str) -> str:
+        cls._registry[name] = callable_obj
+        return name
+
+    @classmethod
+    def get(cls, name: str) -> Callable:
+        return cls._registry[name]
+
+
+class Serializable:
+    @classmethod
+    @abstractmethod
+    def serialize_callable_name(cls) -> str:
+        """Return the registered name used for this class' factory callable."""
+
+    @abstractmethod
+    def clone_params(self) -> tuple[Callable[..., Any], list[Any], dict[str, Any]]:
+        """Return (callable, args, kwargs) necessary to recreate this object."""
+
+    def _prepare(self, params: tuple) -> dict:
+        callable, args, kwargs = params
+        callable_name = callable.__self__.serialize_callable_name()
+        return {
+            "callable": callable_name,
+            "args": args,
+            "kwargs": {
+                k: self._prepare(v) if isinstance(v, tuple) else v
+                for k, v in kwargs.items()
+            },
+        }
+
+    def serialize(self) -> str:
+        """Return a base64-encoded JSON string with registered callable + params."""
+        _ensure_default_callables_registered()
+        data = self.clone_params()
+        return base64.b64encode(json.dumps(self._prepare(data)).encode()).decode()
+
+
+def deserialize(s: str) -> Serializable:
+    """Deserialize from base64-encoded JSON using only registered callables.
+
+    Nested serialized objects are instantiated automatically except for those
+    passed via clone parameter tuples (keys ending with ``_clone_params``),
+    which must remain as (callable, args, kwargs) for later factory usage.
+    """
+    ensure_plugins_loaded()
+    _ensure_default_callables_registered()
+    decoded = base64.b64decode(s.encode())
+    data = json.loads(decoded.decode())
+
+    def _is_serialized(obj: Any) -> bool:
+        return isinstance(obj, dict) and {"callable", "args", "kwargs"}.issubset(
+            obj.keys()
+        )
+
+    def _reconstruct(obj: Any, nested: bool = False) -> Any:
+        if not _is_serialized(obj):
+            return obj
+        callable_name: str = obj["callable"]
+        args: list[Any] = obj["args"]
+        kwargs: dict[str, Any] = obj["kwargs"]
+        # Recurse only inside kwargs because serialize() only nests through kwargs
+        for k, v in list(kwargs.items()):
+            if _is_serialized(v):
+                kwargs[k] = _reconstruct(v, True)
+        callable_obj = CallableRegistry.get(callable_name)
+        if nested:
+            return (callable_obj, args, kwargs)
+        # Otherwise instantiate
+        return callable_obj(*args, **kwargs)
+
+    if not _is_serialized(data):
+        raise ValueError("Invalid serialized data format")
+    return _reconstruct(data, False)
+
+
+class _DefaultsState:
+    registered = False
+
+
+def _ensure_default_callables_registered() -> None:
+    if _DefaultsState.registered:
+        return
+
+    from datachain.data_storage.sqlite import (
+        SQLiteDatabaseEngine,
+        SQLiteMetastore,
+        SQLiteWarehouse,
+    )
+
+    # Register (idempotent by name overwrite is fine) using class-level
+    # serialization names to avoid hard-coded literals here.
+    CallableRegistry.register(
+        SQLiteDatabaseEngine.from_db_file,
+        SQLiteDatabaseEngine.serialize_callable_name(),
+    )
+    CallableRegistry.register(
+        SQLiteMetastore.init_after_clone,
+        SQLiteMetastore.serialize_callable_name(),
+    )
+    CallableRegistry.register(
+        SQLiteWarehouse.init_after_clone,
+        SQLiteWarehouse.serialize_callable_name(),
+    )
+
+    _DefaultsState.registered = True

{datachain-0.34.0 → datachain-0.34.2}/src/datachain/data_storage/sqlite.py

@@ -201,10 +201,14 @@ class SQLiteDatabaseEngine(DatabaseEngine):
         """
         return (
             SQLiteDatabaseEngine.from_db_file,
-            [self.db_file],
+            [str(self.db_file)],
             {},
         )
 
+    @classmethod
+    def serialize_callable_name(cls) -> str:
+        return "sqlite.from_db_file"
+
     def _reconnect(self) -> None:
         if not self.is_closed:
             raise RuntimeError("Cannot reconnect on still-open DB!")
@@ -403,6 +407,10 @@ class SQLiteMetastore(AbstractDBMetastore):
             },
         )
 
+    @classmethod
+    def serialize_callable_name(cls) -> str:
+        return "sqlite.metastore.init_after_clone"
+
     @classmethod
     def init_after_clone(
         cls,
@@ -610,6 +618,10 @@ class SQLiteWarehouse(AbstractWarehouse):
             {"db_clone_params": self.db.clone_params()},
         )
 
+    @classmethod
+    def serialize_callable_name(cls) -> str:
+        return "sqlite.warehouse.init_after_clone"
+
     @classmethod
     def init_after_clone(
         cls,

{datachain-0.34.0 → datachain-0.34.2}/src/datachain/lib/dc/datachain.py

@@ -1963,12 +1963,15 @@ class DataChain:
         self,
         flatten: bool = False,
         include_hidden: bool = True,
+        as_object: bool = False,
     ) -> "pd.DataFrame":
         """Return a pandas DataFrame from the chain.
 
         Parameters:
             flatten: Whether to use a multiindex or flatten column names.
             include_hidden: Whether to include hidden columns.
+            as_object: Whether to emit a dataframe backed by Python objects
+                rather than pandas-inferred dtypes.
 
         Returns:
             pd.DataFrame: A pandas DataFrame representation of the chain.
@@ -1984,6 +1987,9 @@ class DataChain:
             columns = pd.MultiIndex.from_tuples(map(tuple, headers))
 
         results = self.results(include_hidden=include_hidden)
+        if as_object:
+            df = pd.DataFrame(results, columns=columns, dtype=object)
+            return df.where(pd.notna(df), None)
         return pd.DataFrame.from_records(results, columns=columns)
 
     def show(
@@ -2006,7 +2012,11 @@ class DataChain:
         import pandas as pd
 
         dc = self.limit(limit) if limit > 0 else self  # type: ignore[misc]
-        df = dc.to_pandas(flatten, include_hidden=include_hidden)
+        df = dc.to_pandas(
+            flatten,
+            include_hidden=include_hidden,
+            as_object=True,
+        )
 
         if df.empty:
             print("Empty result")

datachain-0.34.2/src/datachain/plugins.py

@@ -0,0 +1,30 @@
+"""Plugin loader for DataChain callables.
+
+Discovers and invokes entry points in the group "datachain.callables" once
+per process. This enables external packages (e.g., Studio) to register
+their callables with the serializer registry without explicit imports.
+"""
+
+from importlib import metadata as importlib_metadata
+
+_plugins_loaded = False
+
+
+def ensure_plugins_loaded() -> None:
+    global _plugins_loaded  # noqa: PLW0603
+    if _plugins_loaded:
+        return
+
+    # Compatible across importlib.metadata versions
+    eps_obj = importlib_metadata.entry_points()
+    if hasattr(eps_obj, "select"):
+        eps_list = eps_obj.select(group="datachain.callables")
+    else:
+        # Compatibility for older versions of importlib_metadata, Python 3.9
+        eps_list = eps_obj.get("datachain.callables", [])  # type: ignore[attr-defined]
+
+    for ep in eps_list:
+        func = ep.load()
+        func()
+
+    _plugins_loaded = True

{datachain-0.34.0 → datachain-0.34.2}/src/datachain.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: datachain
-Version: 0.34.0
+Version: 0.34.2
 Summary: Wrangle unstructured AI data at scale
 Author-email: Dmitry Petrov <support@dvc.org>
 License-Expression: Apache-2.0

{datachain-0.34.0 → datachain-0.34.2}/src/datachain.egg-info/SOURCES.txt

@@ -118,6 +118,7 @@ src/datachain/namespace.py
 src/datachain/node.py
 src/datachain/nodes_fetcher.py
 src/datachain/nodes_thread_pool.py
+src/datachain/plugins.py
 src/datachain/progress.py
 src/datachain/project.py
 src/datachain/py.typed

{datachain-0.34.0 → datachain-0.34.2}/tests/conftest.py

@@ -126,6 +126,7 @@ def clean_environment(
     working_dir = str(tmp_path_factory.mktemp("default_working_dir"))
     monkeypatch_session.chdir(working_dir)
     monkeypatch_session.delenv(DataChainDir.ENV_VAR, raising=False)
+    monkeypatch_session.delenv(DataChainDir.ENV_VAR_DATACHAIN_ROOT, raising=False)
 
 
 @pytest.fixture

{datachain-0.34.0 → datachain-0.34.2}/tests/func/test_datachain.py

@@ -35,6 +35,7 @@ from tests.utils import (
     TARRED_TREE,
     df_equal,
     images_equal,
+    skip_if_not_sqlite,
     sorted_dicts,
     text_embedding,
 )
@@ -559,6 +560,25 @@ def test_show(capsys, test_session):
         assert f"{i} {first_name[i]}" in normalized_output
 
 
+@skip_if_not_sqlite
+def test_show_preserves_none(capsys, test_session):
+    chain = dc.read_values(
+        score=[1, None],
+        ts=[
+            datetime(2020, 1, 1, tzinfo=timezone.utc),
+            None,
+        ],
+        session=test_session,
+    )
+
+    chain.show()
+
+    captured = capsys.readouterr().out
+    assert "NaN" not in captured
+    assert "NaT" not in captured
+    assert captured.count("None") >= 2
+
+
 def test_show_without_temp_datasets(capsys, test_session):
     dc.read_values(
         key=[1, 2, 3, 4], session=test_session

{datachain-0.34.0 → datachain-0.34.2}/tests/unit/lib/test_datachain.py

@@ -3,6 +3,7 @@ import json
 import math
 import os
 import re
+from collections import Counter
 from collections.abc import Generator, Iterator
 from unittest.mock import ANY, patch
 
@@ -564,6 +565,32 @@ def test_from_features_simple_types_in_memory():
     assert df["odds"].tolist() == values
 
 
+@skip_if_not_sqlite
+def test_to_pandas_as_object_preserves_none(test_session):
+    timestamp = datetime.datetime(2020, 1, 1, tzinfo=datetime.timezone.utc)
+    chain = dc.read_values(
+        id=[1, None],
+        value=[3.14, None],
+        ts=[timestamp, None],
+        session=test_session,
+    )
+
+    df_default = chain.to_pandas()
+    assert df_default["id"].dtype != object
+    assert df_default["value"].dtype != object
+    assert df_default["id"].isna().sum() == 1
+    assert df_default["value"].isna().sum() == 1
+    assert pd.isna(df_default.loc[df_default["id"].isna(), "ts"]).all()
+
+    df_object = chain.to_pandas(as_object=True)
+    assert df_object["id"].dtype == object
+    assert df_object["value"].dtype == object
+    assert df_object["ts"].dtype == object
+    assert Counter(df_object["id"].tolist()) == Counter([1, None])
+    assert Counter(df_object["value"].tolist()) == Counter([3.14, None])
+    assert Counter(df_object["ts"].tolist()) == Counter([timestamp, None])
+
+
 def test_from_features_more_simple_types(test_session):
     ds = dc.read_values(
         t1=features,

{datachain-0.34.0 → datachain-0.34.2}/tests/unit/test_database_engine.py

@@ -1,12 +1,15 @@
 import base64
+import json
 import os
-import pickle
 
 import pytest
 from sqlalchemy import Column, Integer, Table
 
 from datachain.data_storage.serializer import deserialize
-from datachain.data_storage.sqlite import SQLiteDatabaseEngine, get_db_file_in_memory
+from datachain.data_storage.sqlite import (
+    SQLiteDatabaseEngine,
+    get_db_file_in_memory,
+)
 from tests.utils import skip_if_not_sqlite
 
 
@@ -24,6 +27,7 @@ def test_init_clone(tmp_dir, db_file, expected_db_file):
         expected_db_file = os.fspath(tmp_dir / expected_db_file)
 
     with SQLiteDatabaseEngine.from_db_file(db_file) as db:
+        assert isinstance(db, SQLiteDatabaseEngine)
         assert db.db_file == expected_db_file
 
         # Test clone
@@ -53,17 +57,15 @@ def test_get_db_file_in_memory(db_file, in_memory, expected):
 
 
 def test_serialize(sqlite_db):
-    # Test serialization
+    # JSON serialization format
     serialized = sqlite_db.serialize()
     assert serialized
-    serialized_pickled = base64.b64decode(serialized.encode())
-    assert serialized_pickled
-    (f, args, kwargs) = pickle.loads(serialized_pickled)  # noqa: S301
-    assert str(f) == str(SQLiteDatabaseEngine.from_db_file)
-    assert args == [":memory:"]
-    assert kwargs == {}
-
-    # Test deserialization
+    raw = base64.b64decode(serialized.encode())
+    data = json.loads(raw.decode())
+    assert data["callable"] == "sqlite.from_db_file"
+    assert data["args"] == [":memory:"]
+    assert data["kwargs"] == {}
+
     obj3 = deserialize(serialized)
     assert isinstance(obj3, SQLiteDatabaseEngine)
     assert obj3.db_file == ":memory:"

{datachain-0.34.0 → datachain-0.34.2}/tests/unit/test_metastore.py

@@ -1,5 +1,5 @@
 import base64
-import pickle
+import json
 
 import pytest
 
@@ -24,18 +24,19 @@ def test_sqlite_metastore(sqlite_db):
     assert obj2.db.db_file == sqlite_db.db_file
     assert obj2.clone_params() == obj.clone_params()
 
-    # Test serialization
+    # Test serialization JSON format
     serialized = obj.serialize()
     assert serialized
-    serialized_pickled = base64.b64decode(serialized.encode())
-    assert serialized_pickled
-    (f, args, kwargs) = pickle.loads(serialized_pickled)  # noqa: S301
-    assert str(f) == str(SQLiteMetastore.init_after_clone)
-    assert args == []
-    assert kwargs["uri"] == uri
-    assert str(kwargs["db_clone_params"]) == str(sqlite_db.clone_params())
-
-    # Test deserialization
+    raw = base64.b64decode(serialized.encode())
+    data = json.loads(raw.decode())
+    assert data["callable"] == "sqlite.metastore.init_after_clone"
+    assert data["args"] == []
+    assert data["kwargs"]["uri"] == uri
+    nested = data["kwargs"]["db_clone_params"]
+    assert nested["callable"] == "sqlite.from_db_file"
+    assert nested["args"] == [":memory:"]
+    assert nested["kwargs"] == {}
+
     obj3 = deserialize(serialized)
     assert isinstance(obj3, SQLiteMetastore)
     assert obj3.uri == uri

{datachain-0.34.0 → datachain-0.34.2}/tests/unit/test_query.py

@@ -42,31 +42,12 @@ def test_args(catalog, mock_popen):
     mock_popen.assert_called_once_with(["mypython", "-c", "pass"], env=expected_env)
 
 
-def test_capture_stdout(catalog, mock_popen):
-    mock_popen.stdout = io.BytesIO(b"Hello, World!\rLorem Ipsum\nDolor Sit Amet\nconse")
-    stdout = []
-
-    catalog.query("pass", stdout_callback=stdout.append)
-    assert stdout == ["Hello, World!\r", "Lorem Ipsum\n", "Dolor Sit Amet\n", "conse"]
-
-
-def test_capture_stderr(catalog, mock_popen):
-    mock_popen.stderr = io.BytesIO(b"Hello, World!\rLorem Ipsum\nDolor Sit Amet\nconse")
-    stderr = []
-
-    catalog.query("pass", stderr_callback=stderr.append)
-    assert stderr == ["Hello, World!\r", "Lorem Ipsum\n", "Dolor Sit Amet\n", "conse"]
-
-
 def test_capture_output(catalog, mock_popen):
     mock_popen.stdout = io.BytesIO(b"Hello, World!\rLorem Ipsum\nDolor Sit Amet\nconse")
-    mock_popen.stderr = io.BytesIO(b"foo\nbar")
-    stdout = []
-    stderr = []
+    lines = []
 
-    catalog.query("pass", stdout_callback=stdout.append, stderr_callback=stderr.append)
-    assert stdout == ["Hello, World!\r", "Lorem Ipsum\n", "Dolor Sit Amet\n", "conse"]
-    assert stderr == ["foo\n", "bar"]
+    catalog.query("pass", capture_output=True, output_hook=lines.append)
+    assert lines == ["Hello, World!\r", "Lorem Ipsum\n", "Dolor Sit Amet\n", "conse"]
 
 
 def test_canceled_by_user(catalog, mock_popen):