Coverage for python/lsst/images/serialization/_io.py: 91%
87 statements
« prev ^ index » next coverage.py v7.14.3, created at 2026-06-29 22:40 +0000
1# This file is part of lsst-images.
2#
3# Developed for the LSST Data Management System.
4# This product includes software developed by the LSST Project
5# (https://www.lsst.org).
6# See the COPYRIGHT file at the top-level directory of this distribution
7# for details of code ownership.
8#
9# Use of this source code is governed by a 3-clause BSD-style
10# license that can be found in the LICENSE file.
11"""Generic ``read`` / ``write`` dispatchers and the schema-name registry."""
13from __future__ import annotations
15__all__ = (
16 "class_for_schema",
17 "parameterize_tree",
18 "public_type_for_schema",
19 "read",
20 "register_schema_class",
21 "tree_class_for_info",
22 "write",
23)
25import importlib
26import importlib.metadata
27from typing import TYPE_CHECKING, Any, overload
29from lsst.resources import ResourcePathExpression
31from ._backends import backend_for_path
32from ._common import ArchiveReadError, ArchiveTree
34if TYPE_CHECKING:
35 from ._input_archive import ArchiveInfo
37_REGISTRY: dict[str, type[ArchiveTree]] = {}
38"""Map of ``SCHEMA_NAME`` to the registered ``ArchiveTree`` subclass.
40The registry is keyed by name only. Schema-version compatibility is
41enforced when the selected tree's ``model_validate*`` runs, via
42``min_read_version``.
43"""
45_SCHEMA_ENTRY_POINT_GROUP = "lsst.images.schemas"
46"""Entry point group for third-party serialization-model providers."""
48_BUILTIN_SCHEMA_PROVIDERS: dict[str, str] = {
49 "cell_coadd": "lsst.images.cells._coadd:CellCoaddSerializationModel",
50 "cell_psf": "lsst.images.cells._psf:CellPointSpreadFunctionSerializationModel",
51 "coadd_provenance": "lsst.images.cells._provenance:CoaddProvenanceSerializationModel",
52}
53"""Schema providers owned by this package but not imported by ``lsst.images``.
55These duplicate the package's own ``lsst.images.schemas`` entry points so
56source-tree use via ``PYTHONPATH=python`` has the same lazy-import behavior as
57an installed distribution with entry point metadata.
59Schemas whose model classes are imported unconditionally by ``lsst.images`` do
60not need built-in providers or entry points: their ``ArchiveTree`` subclass
61hooks register them before this lazy path is needed.
62"""
65def class_for_schema(schema_name: str) -> type[ArchiveTree] | None:
66 """Return the registered ``ArchiveTree`` subclass for ``schema_name``.
68 If no class is already registered, this attempts schema-specific lazy
69 imports from built-in providers and then from entry points in the
70 ``lsst.images.schemas`` group before returning `None`.
72 Parameters
73 ----------
74 schema_name
75 Schema name (e.g. ``"visit_image"``).
76 """
77 if (cls := _REGISTRY.get(schema_name)) is not None:
78 return cls
79 _load_builtin_schema_provider(schema_name)
80 if (cls := _REGISTRY.get(schema_name)) is not None:
81 return cls
82 _load_schema_entry_points(schema_name)
83 return _REGISTRY.get(schema_name)
86def register_schema_class(cls: type[ArchiveTree]) -> None:
87 """Register ``cls`` under ``cls.SCHEMA_NAME``.
89 No-op when the same class is registered again (re-import during
90 tests). Raises `RuntimeError` when a *different* class is
91 registered under an existing name.
93 Intended to be called from ``ArchiveTree.__pydantic_init_subclass__``;
94 not part of the public API.
95 """
96 key = cls.SCHEMA_NAME
97 existing = _REGISTRY.get(key)
98 if existing is cls:
99 return
100 if existing is not None:
101 raise RuntimeError(
102 f"Schema {cls.SCHEMA_NAME!r} is already registered to "
103 f"{existing.__qualname__}; refusing to replace it with "
104 f"{cls.__qualname__}."
105 )
106 _REGISTRY[key] = cls
109def _load_builtin_schema_provider(schema_name: str) -> None:
110 """Import a package-local provider for ``schema_name``, if one exists."""
111 provider = _BUILTIN_SCHEMA_PROVIDERS.get(schema_name)
112 if provider is None:
113 return
114 try:
115 obj = _load_provider_object(provider)
116 except Exception as err:
117 raise ArchiveReadError(
118 f"Could not load built-in schema provider {provider!r} for schema {schema_name!r}: {err}"
119 ) from err
120 _register_provider_object(obj)
121 if schema_name not in _REGISTRY: 121 ↛ 122line 121 didn't jump to line 122 because the condition on line 121 was never true
122 raise ArchiveReadError(
123 f"Built-in schema provider {provider!r} did not register schema {schema_name!r}."
124 )
127def _load_schema_entry_points(schema_name: str) -> None:
128 """Load entry points named ``schema_name`` from ``lsst.images.schemas``."""
129 loaded: list[str] = []
130 for entry_point in importlib.metadata.entry_points(
131 group=_SCHEMA_ENTRY_POINT_GROUP,
132 name=schema_name,
133 ):
134 loaded.append(entry_point.value)
135 try:
136 obj = entry_point.load()
137 except Exception as err:
138 raise ArchiveReadError(
139 f"Could not load schema provider entry point {entry_point.value!r} "
140 f"for schema {schema_name!r}: {err}"
141 ) from err
142 _register_provider_object(obj)
143 if loaded and schema_name not in _REGISTRY: 143 ↛ 144line 143 didn't jump to line 144 because the condition on line 143 was never true
144 raise ArchiveReadError(
145 f"Schema provider entry point(s) for {schema_name!r} did not register that schema: {loaded}."
146 )
149def _load_provider_object(provider: str) -> object:
150 """Load ``module[:attribute[.nested]]`` provider specifications."""
151 module_name, _, attr_path = provider.partition(":")
152 obj: object = importlib.import_module(module_name)
153 if attr_path: 153 ↛ 156line 153 didn't jump to line 156 because the condition on line 153 was always true
154 for attr in attr_path.split("."):
155 obj = getattr(obj, attr)
156 return obj
159def _register_provider_object(obj: object) -> None:
160 """Register ``obj`` if a provider returned an ``ArchiveTree`` subclass."""
161 if isinstance(obj, type) and issubclass(obj, ArchiveTree): 161 ↛ exitline 161 didn't return from function '_register_provider_object' because the condition on line 161 was always true
162 register_schema_class(obj)
165def tree_class_for_info(info: ArchiveInfo, path: ResourcePathExpression) -> type[ArchiveTree]:
166 """Return the registered `ArchiveTree` subclass for ``info``'s schema.
168 Parameters
169 ----------
170 info
171 Basic archive info whose ``schema_name`` selects the tree class.
172 path
173 Path being opened, used only for the error message.
175 Raises
176 ------
177 ArchiveReadError
178 If no class is registered for the schema.
179 """
180 tree_cls = class_for_schema(info.schema_name)
181 if tree_cls is None:
182 raise ArchiveReadError(f"No registered schema {info.schema_name!r}; cannot open {path!r}.")
183 return tree_cls
186def parameterize_tree(
187 tree_cls: type[ArchiveTree],
188 pointer_type: type[Any],
189) -> type[ArchiveTree]:
190 """Parameterise ``tree_cls`` over ``pointer_type`` if it is generic.
192 Some `ArchiveTree` subclasses (e.g. ``SumFieldSerializationModel``)
193 take no type parameters; their ``_get_archive_tree_type`` returns
194 the class itself. Match that behaviour here so per-backend
195 ``open_tree`` implementations can call this uniformly.
197 Parameters
198 ----------
199 tree_cls
200 Archive tree class to parameterise.
201 pointer_type
202 Pointer type to parameterise ``tree_cls`` over when it is generic.
203 """
204 if not getattr(tree_cls, "__parameters__", ()):
205 return tree_cls
206 return tree_cls[pointer_type] # type: ignore[index]
209def public_type_for_schema(schema_name: str) -> type | None:
210 """Return the in-memory Python class produced when reading an archive
211 whose top-level tree has schema name ``schema_name``.
213 Looks the schema name up in the registry and returns the registered
214 tree's ``PUBLIC_TYPE`` ClassVar (the type its ``deserialize`` produces).
215 Returns `None` when nothing is registered for ``schema_name``.
217 Parameters
218 ----------
219 schema_name
220 Schema name (e.g. ``"visit_image"``).
221 """
222 tree_cls = class_for_schema(schema_name)
223 if tree_cls is None:
224 return None
225 return getattr(tree_cls, "PUBLIC_TYPE", None)
228@overload
229def read[T](path: ResourcePathExpression, cls: type[T], **kwargs: Any) -> T: ...
230@overload
231def read(path: ResourcePathExpression, cls: None = ..., **kwargs: Any) -> Any: ... 231 ↛ exitline 231 didn't return from function 'read' because
232def read(path: ResourcePathExpression, cls: type[Any] | None = None, **kwargs: Any) -> Any:
233 """Read an archive whose in-memory type is inferred from its schema.
235 Dispatches to the appropriate backend based on ``path``'s extension,
236 resolves the registered in-memory type from the file's schema, and
237 returns the fully deserialized object.
238 Schema-version compatibility is enforced when the model validates the
239 on-disk tree, via ``min_read_version``.
241 This is the convenient way to read a whole object. To read individual
242 components, or to reach the metadata and butler info stored alongside the
243 object, use `open` instead.
245 Parameters
246 ----------
247 path
248 File to read; convertible to `lsst.resources.ResourcePath`.
249 cls
250 Optional expected in-memory type.
251 When given, the file's schema is checked against ``cls`` and the
252 deserialized object is validated with ``isinstance`` (raising
253 `TypeError` otherwise), and the static return type is ``T``.
254 **kwargs
255 Type-specific keyword arguments forwarded to the object's
256 ``deserialize`` (e.g. ``bbox`` for an image subset read).
257 Mis-targeted arguments surface as ``TypeError``.
258 Backend-specific open options (e.g. ``page_size``) are not accepted
259 here; use `open` for those.
261 Returns
262 -------
263 object
264 The deserialized object.
266 Raises
267 ------
268 ValueError
269 Raised by `backend_for_path` if the file extension is not recognized.
270 ArchiveReadError
271 Raised when the file's ``schema_name`` is not registered, or
272 propagated from the model's ``min_read_version`` check on
273 ``model_validate*``.
274 TypeError
275 Raised when ``cls`` is given and the file's schema or the
276 deserialized object is not compatible with it.
277 """
278 # Imported here to break the _io <-> _reader import cycle: _reader imports
279 # class_for_schema / public_type_for_schema from this module at load time.
280 from ._reader import open as open_archive
282 # A subset read (any deserialize kwarg with a value) reads incrementally;
283 # a plain whole-object read may slurp the file up front. This mirrors the
284 # ``partial`` default the per-backend readers used.
285 partial = any(value is not None for value in kwargs.values())
286 with open_archive(path, cls, partial=partial) as reader:
287 return reader.read(**kwargs)
290def write(obj: Any, path: str, **kwargs: Any) -> Any:
291 """Write ``obj`` to ``path``, dispatching by file extension.
293 Forwards ``**kwargs`` to the per-backend ``write`` (e.g.
294 ``compression_options`` for FITS). No registry lookup is performed:
295 the per-backend ``write`` already accepts any object with a
296 ``serialize`` method.
298 Parameters
299 ----------
300 obj
301 Object to write; must implement ``serialize`` like the per-backend
302 write functions expect.
303 path
304 Destination path. The extension selects the backend.
305 **kwargs
306 Forwarded verbatim to the backend's ``write``.
308 Returns
309 -------
310 Any
311 Whatever the per-backend ``write`` returns (the serialised
312 archive tree).
313 """
314 return backend_for_path(path).write(obj, path, **kwargs)