Coverage for python/lsst/obs/base/gen2to3/rootRepoConverter.py : 18%

Hot-keys on this page
r m x p toggle line displays
j k next/prev highlighted chunk
0 (zero) top of page
1 (one) first highlighted chunk
1# This file is part of obs_base.
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# This program is free software: you can redistribute it and/or modify
10# it under the terms of the GNU General Public License as published by
11# the Free Software Foundation, either version 3 of the License, or
12# (at your option) any later version.
13#
14# This program is distributed in the hope that it will be useful,
15# but WITHOUT ANY WARRANTY; without even the implied warranty of
16# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17# GNU General Public License for more details.
18#
19# You should have received a copy of the GNU General Public License
20# along with this program. If not, see <http://www.gnu.org/licenses/>.
21from __future__ import annotations
23__all__ = ["RootRepoConverter"]
25import os
26import re
27import itertools
28from typing import TYPE_CHECKING, Iterator, Optional, Tuple, List, Set
30from lsst.skymap import BaseSkyMap
31from lsst.daf.butler import DatasetType, DatasetRef, DimensionGraph, FileDataset
32from .standardRepoConverter import StandardRepoConverter
34SKYMAP_DATASET_TYPES = {
35 coaddName: f"{coaddName}Coadd_skyMap" for coaddName in ("deep", "goodSeeing", "dcr")
36}
38if TYPE_CHECKING: 38 ↛ 39line 38 didn't jump to line 39, because the condition on line 38 was never true
39 from lsst.daf.butler import SkyPixDimension
42def getDataPaths(dataRefs):
43 """Strip HDU identifiers from paths and return a unique set of paths.
45 Parameters
46 ----------
47 dataRefs : `lsst.daf.persistence.ButlerDataRef`
48 The gen2 datarefs to strip "[HDU]" values from.
50 Returns
51 -------
52 paths : `set` [`str`]
53 The unique file paths without appended "[HDU]".
54 """
55 paths = set()
56 for dataRef in dataRefs:
57 path = dataRef.getUri()
58 # handle with FITS files with multiple HDUs (e.g. decam raw)
59 paths.add(path.split('[')[0])
60 return paths
63class RootRepoConverter(StandardRepoConverter):
64 """A specialization of `RepoConverter` for root data repositories.
66 `RootRepoConverter` adds support for raw images (mostly delegated to the
67 parent task's `RawIngestTask` subtask) and reference catalogs.
69 Parameters
70 ----------
71 kwds
72 Keyword arguments are forwarded to (and required by) `RepoConverter`.
73 """
75 def __init__(self, **kwds):
76 super().__init__(run=None, **kwds)
77 self._refCats: List[Tuple[str, SkyPixDimension]] = []
78 if self.task.config.rootSkyMapName is not None:
79 self._rootSkyMap = self.task.config.skyMaps[self.task.config.rootSkyMapName].skyMap.apply()
80 else:
81 self._rootSkyMap = None # All access to _rootSkyMap is guarded
82 self._chain = {}
83 self._rawRefs = []
85 def isDatasetTypeSpecial(self, datasetTypeName: str) -> bool:
86 # Docstring inherited from RepoConverter.
87 return (
88 super().isDatasetTypeSpecial(datasetTypeName)
89 or datasetTypeName in ("raw", "ref_cat", "ref_cat_config")
90 # in Gen2, some of these are in the root repo, not a calib repo
91 or datasetTypeName in self.task.config.curatedCalibrations
92 )
94 def getSpecialDirectories(self) -> List[str]:
95 # Docstring inherited from RepoConverter.
96 return super().getSpecialDirectories() + ["CALIB", "ref_cats", "rerun"]
98 def findMatchingSkyMap(self, datasetTypeName: str) -> Tuple[Optional[BaseSkyMap], Optional[str]]:
99 # Docstring inherited from StandardRepoConverter.findMatchingSkyMap.
100 skyMap, name = super().findMatchingSkyMap(datasetTypeName)
101 if skyMap is None and self.task.config.rootSkyMapName is not None:
102 self.task.log.debug(
103 ("Assuming configured root skymap with name '%s' for dataset %s."),
104 self.task.config.rootSkyMapName, datasetTypeName
105 )
106 skyMap = self._rootSkyMap
107 name = self.task.config.rootSkyMapName
108 return skyMap, name
110 def runRawIngest(self):
111 if self.task.raws is None:
112 return
113 self.task.log.info(f"Finding raws in root {self.root}.")
114 if self.subset is not None:
115 dataRefs = itertools.chain.from_iterable(
116 self.butler2.subset(self.task.config.rawDatasetType,
117 visit=visit) for visit in self.subset.visits
118 )
119 else:
120 dataRefs = self.butler2.subset(self.task.config.rawDatasetType)
121 dataPaths = getDataPaths(dataRefs)
122 self.task.log.info("Ingesting raws from root %s into run %s.", self.root, self.task.raws.butler.run)
123 self._rawRefs.extend(self.task.raws.run(dataPaths))
124 self._chain = {self.task.raws.butler.run: {self.task.raws.datasetType.name}}
126 def runDefineVisits(self):
127 if self.task.defineVisits is None:
128 return
129 dimensions = DimensionGraph(self.task.universe, names=["exposure"])
130 exposureDataIds = set(ref.dataId.subset(dimensions) for ref in self._rawRefs)
131 self.task.log.info("Defining visits from exposures.")
132 self.task.defineVisits.run(exposureDataIds)
134 def prep(self):
135 # Docstring inherited from RepoConverter.
136 # Gather information about reference catalogs.
137 if self.task.isDatasetTypeIncluded("ref_cat") and len(self.task.config.refCats) != 0:
138 from lsst.meas.algorithms import DatasetConfig as RefCatDatasetConfig
139 for refCat in os.listdir(os.path.join(self.root, "ref_cats")):
140 path = os.path.join(self.root, "ref_cats", refCat)
141 configFile = os.path.join(path, "config.py")
142 if not os.path.exists(configFile):
143 continue
144 if refCat not in self.task.config.refCats:
145 continue
146 self.task.log.info(f"Preparing ref_cat {refCat} from root {self.root}.")
147 onDiskConfig = RefCatDatasetConfig()
148 onDiskConfig.load(configFile)
149 if onDiskConfig.indexer.name != "HTM":
150 raise ValueError(f"Reference catalog '{refCat}' uses unsupported "
151 f"pixelization '{onDiskConfig.indexer.name}'.")
152 level = onDiskConfig.indexer["HTM"].depth
153 try:
154 dimension = self.task.universe[f"htm{level}"]
155 except KeyError as err:
156 raise ValueError(f"Reference catalog {refCat} uses HTM level {level}, but no htm{level} "
157 f"skypix dimension is configured for this registry.") from err
158 self.task.useSkyPix(dimension)
159 self._refCats.append((refCat, dimension))
160 if self.task.isDatasetTypeIncluded("brightObjectMask") and self.task.config.rootSkyMapName:
161 self.task.useSkyMap(self._rootSkyMap, self.task.config.rootSkyMapName)
162 super().prep()
164 def iterDatasets(self) -> Iterator[FileDataset]:
165 # Docstring inherited from RepoConverter.
166 # Iterate over reference catalog files.
167 for refCat, dimension in self._refCats:
168 datasetType = DatasetType(refCat, dimensions=[dimension], universe=self.task.universe,
169 storageClass="SimpleCatalog")
170 if self.subset is None:
171 regex = re.compile(r"(\d+)\.fits")
172 for fileName in os.listdir(os.path.join(self.root, "ref_cats", refCat)):
173 m = regex.match(fileName)
174 if m is not None:
175 htmId = int(m.group(1))
176 dataId = self.task.registry.expandDataId({dimension: htmId})
177 yield FileDataset(path=os.path.join(self.root, "ref_cats", refCat, fileName),
178 refs=DatasetRef(datasetType, dataId))
179 else:
180 for begin, end in self.subset.skypix[dimension]:
181 for htmId in range(begin, end):
182 dataId = self.task.registry.expandDataId({dimension: htmId})
183 yield FileDataset(path=os.path.join(self.root, "ref_cats", refCat, f"{htmId}.fits"),
184 refs=DatasetRef(datasetType, dataId))
185 yield from super().iterDatasets()
187 def getRun(self, datasetTypeName: str) -> str:
188 # Docstring inherited from RepoConverter.
189 run = self.task.config.runs[datasetTypeName]
190 self._chain.setdefault(run, set()).add(datasetTypeName)
191 return run
193 def getCollectionChain(self) -> List[Tuple[str, Set[str]]]:
194 """Return tuples of run name and associated dataset type names that
195 can be used to construct a chained collection that refers to the
196 converted root repository (`list` [ `tuple` ]).
197 """
198 return list(self._chain.items())