Coverage for python/lsst/images/serialization/_io.py: 91%
88 statements
« prev ^ index » next coverage.py v7.15.2, created at 2026-07-16 15:24 -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 pathlib import Path
30from typing import IO, TYPE_CHECKING, Any, overload
32from lsst.resources import ResourcePathExpression
34from ._backends import backend_for_path
35from ._common import ArchiveReadError, ArchiveTree
37if TYPE_CHECKING:
38 from ._input_archive import ArchiveInfo
40_REGISTRY: dict[str, type[ArchiveTree]] = {}
41"""Map of ``SCHEMA_NAME`` to the registered ``ArchiveTree`` subclass.
43The registry is keyed by name only. Schema-version compatibility is
44enforced when the selected tree's ``model_validate*`` runs, via
45``min_read_version``.
46"""
48_SCHEMA_ENTRY_POINT_GROUP = "lsst.images.schemas"
49"""Entry point group for third-party serialization-model providers."""
51_BUILTIN_SCHEMA_PROVIDERS: dict[str, str] = {
52 "cell_aperture_correction_map": (
53 "lsst.images.cells._aperture_corrections:CellApertureCorrectionMapSerializationModel"
54 ),
55 "cell_coadd": "lsst.images.cells._coadd:CellCoaddSerializationModel",
56 "cell_psf": "lsst.images.cells._psf:CellPointSpreadFunctionSerializationModel",
57 "coadd_provenance": "lsst.images.cells._provenance:CoaddProvenanceSerializationModel",
58}
59"""Schema providers owned by this package but not imported by ``lsst.images``.
61These duplicate the package's own ``lsst.images.schemas`` entry points so
62source-tree use via ``PYTHONPATH=python`` has the same lazy-import behavior as
63an installed distribution with entry point metadata.
65Schemas whose model classes are imported unconditionally by ``lsst.images`` do
66not need built-in providers or entry points: their ``ArchiveTree`` subclass
67hooks register them before this lazy path is needed.
68"""
71def class_for_schema(schema_name: str) -> type[ArchiveTree] | None:
72 """Return the registered ``ArchiveTree`` subclass for ``schema_name``.
74 If no class is already registered, this attempts schema-specific lazy
75 imports from built-in providers and then from entry points in the
76 ``lsst.images.schemas`` group before returning `None`.
78 Parameters
79 ----------
80 schema_name
81 Schema name (e.g. ``"visit_image"``).
82 """
83 if (cls := _REGISTRY.get(schema_name)) is not None:
84 return cls
85 _load_builtin_schema_provider(schema_name)
86 if (cls := _REGISTRY.get(schema_name)) is not None:
87 return cls
88 _load_schema_entry_points(schema_name)
89 return _REGISTRY.get(schema_name)
92def register_schema_class(cls: type[ArchiveTree]) -> None:
93 """Register ``cls`` under ``cls.SCHEMA_NAME``.
95 No-op when the same class is registered again (re-import during
96 tests). Raises `RuntimeError` when a *different* class is
97 registered under an existing name.
99 Intended to be called from ``ArchiveTree.__pydantic_init_subclass__``;
100 not part of the public API.
101 """
102 key = cls.SCHEMA_NAME
103 existing = _REGISTRY.get(key)
104 if existing is cls:
105 return
106 if existing is not None:
107 raise RuntimeError(
108 f"Schema {cls.SCHEMA_NAME!r} is already registered to "
109 f"{existing.__qualname__}; refusing to replace it with "
110 f"{cls.__qualname__}."
111 )
112 _REGISTRY[key] = cls
115def _load_builtin_schema_provider(schema_name: str) -> None:
116 """Import a package-local provider for ``schema_name``, if one exists."""
117 provider = _BUILTIN_SCHEMA_PROVIDERS.get(schema_name)
118 if provider is None:
119 return
120 try:
121 obj = _load_provider_object(provider)
122 except Exception as err:
123 raise ArchiveReadError(
124 f"Could not load built-in schema provider {provider!r} for schema {schema_name!r}: {err}"
125 ) from err
126 _register_provider_object(obj)
127 if schema_name not in _REGISTRY: 127 ↛ 128line 127 didn't jump to line 128 because the condition on line 127 was never true
128 raise ArchiveReadError(
129 f"Built-in schema provider {provider!r} did not register schema {schema_name!r}."
130 )
133def _load_schema_entry_points(schema_name: str) -> None:
134 """Load entry points named ``schema_name`` from ``lsst.images.schemas``."""
135 loaded: list[str] = []
136 for entry_point in importlib.metadata.entry_points(
137 group=_SCHEMA_ENTRY_POINT_GROUP,
138 name=schema_name,
139 ):
140 loaded.append(entry_point.value)
141 try:
142 obj = entry_point.load()
143 except Exception as err:
144 raise ArchiveReadError(
145 f"Could not load schema provider entry point {entry_point.value!r} "
146 f"for schema {schema_name!r}: {err}"
147 ) from err
148 _register_provider_object(obj)
149 if loaded and schema_name not in _REGISTRY: 149 ↛ 150line 149 didn't jump to line 150 because the condition on line 149 was never true
150 raise ArchiveReadError(
151 f"Schema provider entry point(s) for {schema_name!r} did not register that schema: {loaded}."
152 )
155def _load_provider_object(provider: str) -> object:
156 """Load ``module[:attribute[.nested]]`` provider specifications."""
157 module_name, _, attr_path = provider.partition(":")
158 obj: object = importlib.import_module(module_name)
159 if attr_path: 159 ↛ 162line 159 didn't jump to line 162 because the condition on line 159 was always true
160 for attr in attr_path.split("."):
161 obj = getattr(obj, attr)
162 return obj
165def _register_provider_object(obj: object) -> None:
166 """Register ``obj`` if a provider returned an ``ArchiveTree`` subclass."""
167 if isinstance(obj, type) and issubclass(obj, ArchiveTree): 167 ↛ exitline 167 didn't return from function '_register_provider_object' because the condition on line 167 was always true
168 register_schema_class(obj)
171def tree_class_for_info(info: ArchiveInfo, path: ResourcePathExpression | IO[bytes]) -> type[ArchiveTree]:
172 """Return the registered `ArchiveTree` subclass for ``info``'s schema.
174 Parameters
175 ----------
176 info
177 Basic archive info whose ``schema_name`` selects the tree class.
178 path
179 Path or stream being opened, used only for the error message.
181 Raises
182 ------
183 ArchiveReadError
184 If no class is registered for the schema.
185 """
186 tree_cls = class_for_schema(info.schema_name)
187 if tree_cls is None:
188 raise ArchiveReadError(f"No registered schema {info.schema_name!r}; cannot open {path!r}.")
189 return tree_cls
192def parameterize_tree(
193 tree_cls: type[ArchiveTree],
194 pointer_type: type[Any],
195) -> type[ArchiveTree]:
196 """Parameterise ``tree_cls`` over ``pointer_type`` if it is generic.
198 Some `ArchiveTree` subclasses (e.g. ``SumFieldSerializationModel``)
199 take no type parameters; their ``_get_archive_tree_type`` returns
200 the class itself. Match that behaviour here so per-backend
201 ``open_tree`` implementations can call this uniformly.
203 Parameters
204 ----------
205 tree_cls
206 Archive tree class to parameterise.
207 pointer_type
208 Pointer type to parameterise ``tree_cls`` over when it is generic.
209 """
210 if not getattr(tree_cls, "__parameters__", ()):
211 return tree_cls
212 return tree_cls[pointer_type] # type: ignore[index]
215def public_type_for_schema(schema_name: str) -> type | None:
216 """Return the in-memory Python class produced when reading an archive
217 whose top-level tree has schema name ``schema_name``.
219 Looks the schema name up in the registry and returns the registered
220 tree's ``PUBLIC_TYPE`` ClassVar (the type its ``deserialize`` produces).
221 Returns `None` when nothing is registered for ``schema_name``.
223 Parameters
224 ----------
225 schema_name
226 Schema name (e.g. ``"visit_image"``).
227 """
228 tree_cls = class_for_schema(schema_name)
229 if tree_cls is None:
230 return None
231 return getattr(tree_cls, "PUBLIC_TYPE", None)
234@overload
235def read_archive[T](
236 path: ResourcePathExpression | IO[bytes], cls: type[T], *, format: str | None = ..., **kwargs: Any
237) -> T: ...
238@overload
239def read_archive( 239 ↛ exitline 239 didn't return from function 'read_archive' because
240 path: ResourcePathExpression | IO[bytes], cls: None = ..., *, format: str | None = ..., **kwargs: Any
241) -> Any: ...
242def read_archive(
243 path: ResourcePathExpression | IO[bytes],
244 cls: type[Any] | None = None,
245 *,
246 format: str | None = None,
247 **kwargs: Any,
248) -> Any:
249 """Read an archive whose in-memory type is inferred from its schema.
251 Dispatches to the appropriate backend based on ``path``'s extension (or,
252 for stream input, its leading bytes), resolves the registered in-memory
253 type from the file's schema, and returns the fully deserialized object.
254 Schema-version compatibility is enforced when the model validates the
255 on-disk tree, via ``min_read_version``.
256 A path with a ``.gz``/``.zst`` compression suffix is decompressed
257 transparently; stream input must already be decompressed.
259 This is the convenient way to read a whole object. To read individual
260 components, or to reach the metadata and butler info stored alongside the
261 object, use `open_archive` instead.
263 Parameters
264 ----------
265 path
266 File to read; convertible to `lsst.resources.ResourcePath`, or a
267 seekable binary stream containing the file's content (e.g.
268 ``io.BytesIO(data)`` for in-memory bytes).
269 cls
270 Optional expected in-memory type.
271 When given, the file's schema is checked against ``cls`` and the
272 deserialized object is validated with ``isinstance`` (raising
273 `TypeError` otherwise), and the static return type is ``T``.
274 format
275 Optional backend name (``"fits"``, ``"ndf"``, or ``"json"``)
276 forcing the backend, instead of dispatching on the path's extension
277 or the stream's leading bytes.
278 **kwargs
279 Type-specific keyword arguments forwarded to the object's
280 ``deserialize`` (e.g. ``bbox`` for an image subset read).
281 Mis-targeted arguments surface as ``TypeError``.
282 Backend-specific open options (e.g. ``page_size``) are not accepted
283 here; use `open_archive` for those.
285 Returns
286 -------
287 object
288 The deserialized object.
290 Raises
291 ------
292 ValueError
293 Raised by `backend_for_path` if the file extension is not
294 recognized, by `backend_for_stream` if a stream's leading bytes are
295 not recognized, or by `backend_for_name` if ``format`` is not a
296 known backend name.
297 ArchiveReadError
298 Raised when the file's ``schema_name`` is not registered, or
299 propagated from the model's ``min_read_version`` check on
300 ``model_validate*``.
301 TypeError
302 Raised when ``cls`` is given and the file's schema or the
303 deserialized object is not compatible with it.
304 """
305 # Imported here to break the _io <-> _reader import cycle: _reader imports
306 # class_for_schema / public_type_for_schema from this module at load time.
307 from ._reader import open_archive
309 # A subset read (any deserialize kwarg with a value) reads incrementally;
310 # a plain whole-object read may slurp the file up front. This mirrors the
311 # ``partial`` default the per-backend readers used.
312 partial = any(value is not None for value in kwargs.values())
313 with open_archive(path, cls, format=format, partial=partial) as reader:
314 return reader.read(**kwargs)
317def write_archive(obj: Any, path: Path | str, **kwargs: Any) -> Any:
318 """Write ``obj`` to ``path``, dispatching by file extension.
320 Forwards ``**kwargs`` to the per-backend ``write`` (e.g.
321 ``compression_options`` for FITS). No registry lookup is performed:
322 the per-backend ``write`` already accepts any object with a
323 ``serialize`` method.
325 Parameters
326 ----------
327 obj
328 Object to write; must implement ``serialize`` like the per-backend
329 write functions expect.
330 path
331 Destination path. The extension selects the backend.
332 **kwargs
333 Forwarded verbatim to the backend's ``write``.
335 Returns
336 -------
337 Any
338 Whatever the per-backend ``write`` returns (the serialised
339 archive tree).
340 """
341 return backend_for_path(path).write(obj, path, **kwargs)