Lazy load

HISTORY.md

@@ -2,6 +2,7 @@
 
 ## UNRELEASED
 - Added support for Pydantic serialization (Issue [#537](https://github.com/drivendataorg/cloudpathlib/issues/537), PR [#538](https://github.com/drivendataorg/cloudpathlib/pull/538))
+- Improved import time by lazy-loading cloud provider modules (Issue [#544](https://github.com/drivendataorg/cloudpathlib/issues/544), PR [#TBD](https://github.com/drivendataorg/cloudpathlib/pull/TBD))
 
 ## v0.23.0 (2025-10-07)
 

cloudpathlib/__init__.py

@@ -1,17 +1,30 @@
 import os
 import sys
+from typing import TYPE_CHECKING
 
-from .anypath import AnyPath
-from .azure.azblobclient import AzureBlobClient
-from .azure.azblobpath import AzureBlobPath
-from .cloudpath import CloudPath, implementation_registry
-from .patches import patch_open, patch_os_functions, patch_glob, patch_all_builtins
-from .gs.gsclient import GSClient
-from .gs.gspath import GSPath
-from .http.httpclient import HttpClient, HttpsClient
-from .http.httppath import HttpPath, HttpsPath
-from .s3.s3client import S3Client
-from .s3.s3path import S3Path
+# Lazy imports for cloud providers to avoid loading heavy SDKs at import time
+# Google Cloud SDK alone adds ~200ms to import time
+
+if TYPE_CHECKING:
+    from .anypath import AnyPath as AnyPath
+    from .azure.azblobclient import AzureBlobClient as AzureBlobClient
+    from .azure.azblobpath import AzureBlobPath as AzureBlobPath
+    from .cloudpath import (
+        CloudPath as CloudPath,
+        implementation_registry as implementation_registry,
+    )
+    from .patches import (
+        patch_open as patch_open,
+        patch_os_functions as patch_os_functions,
+        patch_glob as patch_glob,
+        patch_all_builtins as patch_all_builtins,
+    )
+    from .gs.gsclient import GSClient as GSClient
+    from .gs.gspath import GSPath as GSPath
+    from .http.httpclient import HttpClient as HttpClient, HttpsClient as HttpsClient
+    from .http.httppath import HttpPath as HttpPath, HttpsPath as HttpsPath
+    from .s3.s3client import S3Client as S3Client
+    from .s3.s3path import S3Path as S3Path
 
 if sys.version_info[:2] >= (3, 8):
     import importlib.metadata as importlib_metadata
@@ -43,14 +56,66 @@
 ]
 
 
+# Lazy loading implementation
+_LAZY_IMPORTS = {
+    # Core
+    "AnyPath": ".anypath",
+    "CloudPath": ".cloudpath",
+    "implementation_registry": ".cloudpath",
+    # Patches
+    "patch_open": ".patches",
+    "patch_os_functions": ".patches",
+    "patch_glob": ".patches",
+    "patch_all_builtins": ".patches",
+    # S3
+    "S3Client": ".s3.s3client",
+    "S3Path": ".s3.s3path",
+    # GCS
+    "GSClient": ".gs.gsclient",
+    "GSPath": ".gs.gspath",
+    # Azure
+    "AzureBlobClient": ".azure.azblobclient",
+    "AzureBlobPath": ".azure.azblobpath",
+    # HTTP
+    "HttpClient": ".http.httpclient",
+    "HttpsClient": ".http.httpclient",
+    "HttpPath": ".http.httppath",
+    "HttpsPath": ".http.httppath",
+}
+
+
+def __getattr__(name: str):
+    if name in _LAZY_IMPORTS:
+        import importlib
+
+        module_path = _LAZY_IMPORTS[name]
+        module = importlib.import_module(module_path, __name__)
+        return getattr(module, name)
+    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
+
+
+def __dir__():
+    return __all__
+
+
+# Handle environment-variable-based patching
+# These need to be checked at import time, so we do them lazily only if env vars are set
 if bool(os.environ.get("CLOUDPATHLIB_PATCH_OPEN", "")):
-    patch_open()
+    from .patches import patch_open as _patch_open
+
+    _patch_open()
 
 if bool(os.environ.get("CLOUDPATHLIB_PATCH_OS", "")):
-    patch_os_functions()
+    from .patches import patch_os_functions as _patch_os_functions
+
+    _patch_os_functions()
 
 if bool(os.environ.get("CLOUDPATHLIB_PATCH_GLOB", "")):
-    patch_glob()
+    from .patches import patch_glob as _patch_glob
+
+    _patch_glob()
 
 if bool(os.environ.get("CLOUDPATHLIB_PATCH_ALL", "")):
-    patch_all_builtins()
+    from .patches import patch_all_builtins as _patch_all_builtins
+
+    _patch_all_builtins()

cloudpathlib/azure/__init__.py

@@ -1,7 +1,22 @@
-from .azblobclient import AzureBlobClient
-from .azblobpath import AzureBlobPath
+from typing import TYPE_CHECKING
+
+if TYPE_CHECKING:
+    from .azblobclient import AzureBlobClient as AzureBlobClient
+    from .azblobpath import AzureBlobPath as AzureBlobPath
 
 __all__ = [
     "AzureBlobClient",
     "AzureBlobPath",
 ]
+
+
+def __getattr__(name: str):
+    if name == "AzureBlobClient":
+        from .azblobclient import AzureBlobClient
+
+        return AzureBlobClient
+    if name == "AzureBlobPath":
+        from .azblobpath import AzureBlobPath
+
+        return AzureBlobPath
+    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")

cloudpathlib/cloudpath.py

@@ -80,9 +80,8 @@ def _make_selector(pattern_parts, _flavour, case_sensitive=True):  # noqa: F811
     from .legacy.glob import _make_selector  # noqa: F811
 
 
-from cloudpathlib.enums import FileCacheMode
+from .enums import FileCacheMode
 
-from . import anypath
 from .exceptions import (
     ClientMismatchError,
     CloudPathFileExistsError,
@@ -1143,6 +1142,8 @@ def _copy(
         force_overwrite_to_cloud: Optional[bool] = None,
         remove_src: bool = False,
     ) -> Union[Path, Self]:
+        from . import anypath
+
         if not self.exists():
             raise ValueError(f"Path {self} must exist to copy.")
 
@@ -1275,6 +1276,8 @@ def copy_into(
         force_overwrite_to_cloud: Optional[bool] = None,
     ) -> Union[Path, Self]:
         """Copy self into target directory, preserving the filename."""
+        from . import anypath
+
         target_path = anypath.to_anypath(target_dir) / self.name
 
         result = self._copy(
@@ -1312,6 +1315,8 @@ def copytree(
 
     def copytree(self, destination, force_overwrite_to_cloud=None, ignore=None):
         """Copy self to a directory, if self is a directory."""
+        from . import anypath
+
         if not self.is_dir():
             raise CloudPathNotADirectoryError(
                 f"Origin path {self} must be a directory. To copy a single file use the method copy."
@@ -1427,6 +1432,8 @@ def move_into(
         force_overwrite_to_cloud: Optional[bool] = None,
     ) -> Union[Path, Self]:
         """Move self into target directory, preserving the filename and removing the source."""
+        from . import anypath
+
         target_path = anypath.to_anypath(target_dir) / self.name
 
         result = self._copy(

cloudpathlib/gs/__init__.py

@@ -1,7 +1,22 @@
-from .gsclient import GSClient
-from .gspath import GSPath
+from typing import TYPE_CHECKING
+
+if TYPE_CHECKING:
+    from .gsclient import GSClient as GSClient
+    from .gspath import GSPath as GSPath
 
 __all__ = [
     "GSClient",
     "GSPath",
 ]
+
+
+def __getattr__(name: str):
+    if name == "GSClient":
+        from .gsclient import GSClient
+
+        return GSClient
+    if name == "GSPath":
+        from .gspath import GSPath
+
+        return GSPath
+    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")

cloudpathlib/s3/__init__.py

@@ -1,7 +1,22 @@
-from .s3client import S3Client
-from .s3path import S3Path
+from typing import TYPE_CHECKING
+
+if TYPE_CHECKING:
+    from .s3client import S3Client as S3Client
+    from .s3path import S3Path as S3Path
 
 __all__ = [
     "S3Client",
     "S3Path",
 ]
+
+
+def __getattr__(name: str):
+    if name == "S3Client":
+        from .s3client import S3Client
+
+        return S3Client
+    if name == "S3Path":
+        from .s3path import S3Path
+
+        return S3Path
+    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")