Coverage for python/lsst/images/serialization/_io.py: 91%
87 statements
« prev ^ index » next coverage.py v7.14.3, created at 2026-07-09 02:27 -0700
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_archive`` / ``write_archive`` dispatchers and the
12schema-name registry.
13"""
15from __future__ import annotations
17__all__ = (
18 "class_for_schema",
19 "parameterize_tree",
20 "public_type_for_schema",
21 "read_archive",
22 "register_schema_class",
23 "tree_class_for_info",
24 "write_archive",
25)
27import importlib
28import importlib.metadata
29from typing import IO, TYPE_CHECKING, Any, overload
31from lsst.resources import ResourcePathExpression
33from ._backends import backend_for_path
34from ._common import ArchiveReadError, ArchiveTree
36if TYPE_CHECKING:
37 from ._input_archive import ArchiveInfo
39_REGISTRY: dict[str, type[ArchiveTree]] = {}
40"""Map of ``SCHEMA_NAME`` to the registered ``ArchiveTree`` subclass.
42The registry is keyed by name only. Schema-version compatibility is
43enforced when the selected tree's ``model_validate*`` runs, via
44``min_read_version``.
45"""
47_SCHEMA_ENTRY_POINT_GROUP = "lsst.images.schemas"
48"""Entry point group for third-party serialization-model providers."""
50_BUILTIN_SCHEMA_PROVIDERS: dict[str, str] = {
51 "cell_aperture_correction_map": (
52 "lsst.images.cells._aperture_corrections:CellApertureCorrectionMapSerializationModel"
53 ),
54 "cell_coadd": "lsst.images.cells._coadd:CellCoaddSerializationModel",
55 "cell_psf": "lsst.images.cells._psf:CellPointSpreadFunctionSerializationModel",
56 "coadd_provenance": "lsst.images.cells._provenance:CoaddProvenanceSerializationModel",
57}
58"""Schema providers owned by this package but not imported by ``lsst.images``.
60These duplicate the package's own ``lsst.images.schemas`` entry points so
61source-tree use via ``PYTHONPATH=python`` has the same lazy-import behavior as
62an installed distribution with entry point metadata.
64Schemas whose model classes are imported unconditionally by ``lsst.images`` do
65not need built-in providers or entry points: their ``ArchiveTree`` subclass
66hooks register them before this lazy path is needed.
67"""
70def class_for_schema(schema_name: str) -> type[ArchiveTree] | None:
71 """Return the registered ``ArchiveTree`` subclass for ``schema_name``.
73 If no class is already registered, this attempts schema-specific lazy
74 imports from built-in providers and then from entry points in the
75 ``lsst.images.schemas`` group before returning `None`.
77 Parameters
78 ----------
79 schema_name
80 Schema name (e.g. ``"visit_image"``).
81 """
82 if (cls := _REGISTRY.get(schema_name)) is not None:
83 return cls
84 _load_builtin_schema_provider(schema_name)
85 if (cls := _REGISTRY.get(schema_name)) is not None:
86 return cls
87 _load_schema_entry_points(schema_name)
88 return _REGISTRY.get(schema_name)
91def register_schema_class(cls: type[ArchiveTree]) -> None:
92 """Register ``cls`` under ``cls.SCHEMA_NAME``.
94 No-op when the same class is registered again (re-import during
95 tests). Raises `RuntimeError` when a *different* class is
96 registered under an existing name.
98 Intended to be called from ``ArchiveTree.__pydantic_init_subclass__``;
99 not part of the public API.
100 """
101 key = cls.SCHEMA_NAME
102 existing = _REGISTRY.get(key)
103 if existing is cls:
104 return
105 if existing is not None:
106 raise RuntimeError(
107 f"Schema {cls.SCHEMA_NAME!r} is already registered to "
108 f"{existing.__qualname__}; refusing to replace it with "
109 f"{cls.__qualname__}."
110 )
111 _REGISTRY[key] = cls
114def _load_builtin_schema_provider(schema_name: str) -> None:
115 """Import a package-local provider for ``schema_name``, if one exists."""
116 provider = _BUILTIN_SCHEMA_PROVIDERS.get(schema_name)
117 if provider is None:
118 return
119 try:
120 obj = _load_provider_object(provider)
121 except Exception as err:
122 raise ArchiveReadError(
123 f"Could not load built-in schema provider {provider!r} for schema {schema_name!r}: {err}"
124 ) from err
125 _register_provider_object(obj)
126 if schema_name not in _REGISTRY: 126 ↛ 127line 126 didn't jump to line 127 because the condition on line 126 was never true
127 raise ArchiveReadError(
128 f"Built-in schema provider {provider!r} did not register schema {schema_name!r}."
129 )
132def _load_schema_entry_points(schema_name: str) -> None:
133 """Load entry points named ``schema_name`` from ``lsst.images.schemas``."""
134 loaded: list[str] = []
135 for entry_point in importlib.metadata.entry_points(
136 group=_SCHEMA_ENTRY_POINT_GROUP,
137 name=schema_name,
138 ):
139 loaded.append(entry_point.value)
140 try:
141 obj = entry_point.load()
142 except Exception as err:
143 raise ArchiveReadError(
144 f"Could not load schema provider entry point {entry_point.value!r} "
145 f"for schema {schema_name!r}: {err}"
146 ) from err
147 _register_provider_object(obj)
148 if loaded and schema_name not in _REGISTRY: 148 ↛ 149line 148 didn't jump to line 149 because the condition on line 148 was never true
149 raise ArchiveReadError(
150 f"Schema provider entry point(s) for {schema_name!r} did not register that schema: {loaded}."
151 )
154def _load_provider_object(provider: str) -> object:
155 """Load ``module[:attribute[.nested]]`` provider specifications."""
156 module_name, _, attr_path = provider.partition(":")
157 obj: object = importlib.import_module(module_name)
158 if attr_path: 158 ↛ 161line 158 didn't jump to line 161 because the condition on line 158 was always true
159 for attr in attr_path.split("."):
160 obj = getattr(obj, attr)
161 return obj
164def _register_provider_object(obj: object) -> None:
165 """Register ``obj`` if a provider returned an ``ArchiveTree`` subclass."""
166 if isinstance(obj, type) and issubclass(obj, ArchiveTree): 166 ↛ exitline 166 didn't return from function '_register_provider_object' because the condition on line 166 was always true
167 register_schema_class(obj)
170def tree_class_for_info(info: ArchiveInfo, path: ResourcePathExpression | IO[bytes]) -> type[ArchiveTree]:
171 """Return the registered `ArchiveTree` subclass for ``info``'s schema.
173 Parameters
174 ----------
175 info
176 Basic archive info whose ``schema_name`` selects the tree class.
177 path
178 Path or stream being opened, used only for the error message.
180 Raises
181 ------
182 ArchiveReadError
183 If no class is registered for the schema.
184 """
185 tree_cls = class_for_schema(info.schema_name)
186 if tree_cls is None:
187 raise ArchiveReadError(f"No registered schema {info.schema_name!r}; cannot open {path!r}.")
188 return tree_cls
191def parameterize_tree(
192 tree_cls: type[ArchiveTree],
193 pointer_type: type[Any],
194) -> type[ArchiveTree]:
195 """Parameterise ``tree_cls`` over ``pointer_type`` if it is generic.
197 Some `ArchiveTree` subclasses (e.g. ``SumFieldSerializationModel``)
198 take no type parameters; their ``_get_archive_tree_type`` returns
199 the class itself. Match that behaviour here so per-backend
200 ``open_tree`` implementations can call this uniformly.
202 Parameters
203 ----------
204 tree_cls
205 Archive tree class to parameterise.
206 pointer_type
207 Pointer type to parameterise ``tree_cls`` over when it is generic.
208 """
209 if not getattr(tree_cls, "__parameters__", ()):
210 return tree_cls
211 return tree_cls[pointer_type] # type: ignore[index]
214def public_type_for_schema(schema_name: str) -> type | None:
215 """Return the in-memory Python class produced when reading an archive
216 whose top-level tree has schema name ``schema_name``.
218 Looks the schema name up in the registry and returns the registered
219 tree's ``PUBLIC_TYPE`` ClassVar (the type its ``deserialize`` produces).
220 Returns `None` when nothing is registered for ``schema_name``.
222 Parameters
223 ----------
224 schema_name
225 Schema name (e.g. ``"visit_image"``).
226 """
227 tree_cls = class_for_schema(schema_name)
228 if tree_cls is None:
229 return None
230 return getattr(tree_cls, "PUBLIC_TYPE", None)
233@overload
234def read_archive[T](
235 path: ResourcePathExpression | IO[bytes], cls: type[T], *, format: str | None = ..., **kwargs: Any
236) -> T: ...
237@overload
238def read_archive( 238 ↛ exitline 238 didn't return from function 'read_archive' because
239 path: ResourcePathExpression | IO[bytes], cls: None = ..., *, format: str | None = ..., **kwargs: Any
240) -> Any: ...
241def read_archive(
242 path: ResourcePathExpression | IO[bytes],
243 cls: type[Any] | None = None,
244 *,
245 format: str | None = None,
246 **kwargs: Any,
247) -> Any:
248 """Read an archive whose in-memory type is inferred from its schema.
250 Dispatches to the appropriate backend based on ``path``'s extension (or,
251 for stream input, its leading bytes), resolves the registered in-memory
252 type from the file's schema, and returns the fully deserialized object.
253 Schema-version compatibility is enforced when the model validates the
254 on-disk tree, via ``min_read_version``.
255 A path with a ``.gz``/``.zst`` compression suffix is decompressed
256 transparently; stream input must already be decompressed.
258 This is the convenient way to read a whole object. To read individual
259 components, or to reach the metadata and butler info stored alongside the
260 object, use `open_archive` instead.
262 Parameters
263 ----------
264 path
265 File to read; convertible to `lsst.resources.ResourcePath`, or a
266 seekable binary stream containing the file's content (e.g.
267 ``io.BytesIO(data)`` for in-memory bytes).
268 cls
269 Optional expected in-memory type.
270 When given, the file's schema is checked against ``cls`` and the
271 deserialized object is validated with ``isinstance`` (raising
272 `TypeError` otherwise), and the static return type is ``T``.
273 format
274 Optional backend name (``"fits"``, ``"ndf"``, or ``"json"``)
275 forcing the backend, instead of dispatching on the path's extension
276 or the stream's leading bytes.
277 **kwargs
278 Type-specific keyword arguments forwarded to the object's
279 ``deserialize`` (e.g. ``bbox`` for an image subset read).
280 Mis-targeted arguments surface as ``TypeError``.
281 Backend-specific open options (e.g. ``page_size``) are not accepted
282 here; use `open_archive` for those.
284 Returns
285 -------
286 object
287 The deserialized object.
289 Raises
290 ------
291 ValueError
292 Raised by `backend_for_path` if the file extension is not
293 recognized, by `backend_for_stream` if a stream's leading bytes are
294 not recognized, or by `backend_for_name` if ``format`` is not a
295 known backend name.
296 ArchiveReadError
297 Raised when the file's ``schema_name`` is not registered, or
298 propagated from the model's ``min_read_version`` check on
299 ``model_validate*``.
300 TypeError
301 Raised when ``cls`` is given and the file's schema or the
302 deserialized object is not compatible with it.
303 """
304 # Imported here to break the _io <-> _reader import cycle: _reader imports
305 # class_for_schema / public_type_for_schema from this module at load time.
306 from ._reader import open_archive
308 # A subset read (any deserialize kwarg with a value) reads incrementally;
309 # a plain whole-object read may slurp the file up front. This mirrors the
310 # ``partial`` default the per-backend readers used.
311 partial = any(value is not None for value in kwargs.values())
312 with open_archive(path, cls, format=format, partial=partial) as reader:
313 return reader.read(**kwargs)
316def write_archive(obj: Any, path: str, **kwargs: Any) -> Any:
317 """Write ``obj`` to ``path``, dispatching by file extension.
319 Forwards ``**kwargs`` to the per-backend ``write`` (e.g.
320 ``compression_options`` for FITS). No registry lookup is performed:
321 the per-backend ``write`` already accepts any object with a
322 ``serialize`` method.
324 Parameters
325 ----------
326 obj
327 Object to write; must implement ``serialize`` like the per-backend
328 write functions expect.
329 path
330 Destination path. The extension selects the backend.
331 **kwargs
332 Forwarded verbatim to the backend's ``write``.
334 Returns
335 -------
336 Any
337 Whatever the per-backend ``write`` returns (the serialised
338 archive tree).
339 """
340 return backend_for_path(path).write(obj, path, **kwargs)