Big rewrite

conda/meta.yaml

@@ -11,7 +11,7 @@ requirements:
   run:
     - python>=3.8
     - python-dateutil
-    - scipp>=23.01.1
+    - scipp>=23.03.1
     - scipy
     - h5py
 

requirements/base.txt

@@ -1,4 +1,4 @@
-# SHA1:20e195123d87c5cf695f1f70a994a4a4b8d9178b
+# SHA1:c3c05567d668423130fa3b892cc2cae373423b1c
 #
 # This file is autogenerated by pip-compile-multi
 # To update, run:
@@ -20,7 +20,7 @@ python-dateutil==2.8.2
     # via -r base.in
 pyyaml==6.0
     # via confuse
-scipp==23.3.0
+scipp==23.3.1
     # via -r base.in
 scipy==1.10.0
     # via -r base.in

setup.cfg

@@ -24,7 +24,7 @@ package_dir =
 packages = find:
 install_requires =
     python-dateutil
-    scipp>=23.01.1
+    scipp>=23.03.1
     scipy  # we use scipp.interpolate which depends on this
     h5py
 python_requires = >=3.8

src/scippnexus/v2/__init__.py

@@ -0,0 +1,25 @@
+# SPDX-License-Identifier: BSD-3-Clause
+# Copyright (c) 2023 Scipp contributors (https://github.com/scipp)
+# @author Simon Heybrock
+
+# flake8: noqa
+import importlib.metadata
+
+try:
+    __version__ = importlib.metadata.version(__package__ or __name__)
+except importlib.metadata.PackageNotFoundError:
+    __version__ = "0.0.0"
+
+from .. import typing
+from .base import (
+    Group,
+    NexusStructureError,
+    NXobject,
+    base_definitions,
+    create_class,
+    create_field,
+)
+from .field import Field
+from .file import File
+from .nexus_classes import *
+from .nxdata import group_events_by_detector_number

src/scippnexus/v2/application_definitions/nxcansas/__init__.py

@@ -0,0 +1,9 @@
+# SPDX-License-Identifier: BSD-3-Clause
+# Copyright (c) 2023 Scipp contributors (https://github.com/scipp)
+# @author Simon Heybrock
+
+# flake8: noqa
+
+from .nxcansas import SASdata, SASentry, definitions
+
+__all__ = ['definitions', 'SASentry', 'SASdata']

src/scippnexus/v2/application_definitions/nxcansas/nxcansas.py

@@ -0,0 +1,128 @@
+# SPDX-License-Identifier: BSD-3-Clause
+# Copyright (c) 2023 Scipp contributors (https://github.com/scipp)
+# @author Simon Heybrock
+from typing import Any, Callable, Dict, Literal, Optional, Union
+
+import scipp as sc
+
+from ....typing import H5Group
+from ...base import Group, NXobject, base_definitions, create_field
+from ...field import Field
+from ...nxdata import NXdata
+
+
+class SASentry:
+    nx_class = 'NXentry'
+
+    def __init__(self, *, title: str, run: Union[str, int]):
+        self.title = title
+        self.run = run
+
+    def __write_to_nexus_group__(self, group: H5Group):
+        group.attrs['canSAS_class'] = 'SASentry'
+        group.attrs['version'] = '1.0'
+        group.attrs['definition'] = 'NXcanSAS'
+        create_field(group, 'title', self.title)
+        create_field(group, 'run', self.run)
+
+
+class SASdata:
+    nx_class = 'NXdata'
+
+    def __init__(self,
+                 data: sc.DataArray,
+                 Q_variances: Optional[Literal['uncertainties', 'resolutions']] = None):
+        self.data = data
+        valid = ('uncertainties', 'resolutions')
+        if Q_variances not in (None, ) + valid:
+            raise ValueError(f"Q_variances must be in {valid}")
+        self._variances = Q_variances
+
+    def __write_to_nexus_group__(self, group: H5Group):
+        da = self.data
+        group.attrs['canSAS_class'] = 'SASdata'
+        group.attrs['signal'] = 'I'
+        group.attrs['axes'] = da.dims  # for NeXus compliance, same as I_axes
+        group.attrs['I_axes'] = da.dims
+        group.attrs['Q_indices'] = tuple(da.dims.index(d) for d in da.coords['Q'].dims)
+        # TODO writing Field should deal with variances
+        signal = create_field(group, 'I', sc.values(da.data))
+        # We use the _errors suffix for NeXus compliance, unlike the examples given in
+        # NXcanSAS.
+        if da.variances is not None:
+            signal.attrs['uncertainties'] = 'I_errors'
+            create_field(group, 'I_errors', sc.stddevs(da.data))
+        if da.coords.is_edges('Q'):
+            raise ValueError(
+                "Q is given as bin-edges, but NXcanSAS requires Q points (such as "
+                "bin centers).")
+        coord = create_field(group, 'Q', da.coords['Q'])
+        if da.coords['Q'].variances is not None:
+            if self._variances is None:
+                raise ValueError(
+                    "Q has variances, must specify whether these represent "
+                    "'uncertainties' or 'resolutions' using the 'Q_variances' option'")
+
+            coord.attrs[self._variances] = 'Q_errors'
+            create_field(group, 'Q_errors', sc.stddevs(da.coords['Q']))
+
+
+class _SASdata(NXdata):
+
+    def __init__(self, attrs: Dict[str, Any], children: Dict[str, Union[Field, Group]]):
+        fallback_dims = attrs.get('I_axes')
+        if fallback_dims is not None:
+            fallback_dims = (fallback_dims, )
+        super().__init__(attrs=attrs,
+                         children=children,
+                         fallback_dims=fallback_dims,
+                         fallback_signal_name='I')
+
+    # TODO Mechanism for custom error names
+    @staticmethod
+    def signal_errors(group: NXobject) -> Optional[str]:
+        signal_name = group.attrs.get('signal', 'I')
+        signal = group._group[signal_name]
+        return signal.attrs.get('uncertainties')
+
+    def coord_errors(group: NXobject, name: str) -> Optional[str]:
+        if name != 'Q':
+            return None
+        # TODO This naively stores this as Scipp errors, which are just Gaussian.
+        # This is probably not correct in all cases.
+        uncertainties = group[name].attrs.get('uncertainties')
+        resolutions = group[name].attrs.get('resolutions')
+        if uncertainties is None:
+            return resolutions
+        elif resolutions is None:
+            return uncertainties
+        raise RuntimeError("Cannot handle both uncertainties and resolutions for Q")
+
+
+class _SAStransmission_spectrum(NXdata):
+
+    def __init__(self, attrs: Dict[str, Any], children: Dict[str, Union[Field, Group]]):
+        # TODO A valid file should have T_axes, do we need to fallback?
+        super().__init__(attrs=attrs,
+                         children=children,
+                         fallback_dims=(attrs.get('T_axes', 'lambda'), ),
+                         fallback_signal_name='T')
+
+
+class NXcanSAS:
+
+    def get(self, key: type, default: Callable) -> Callable:
+
+        def _definition_factory(attrs: Dict[str, Any],
+                                children: Dict[str, Union[Field, Group]]) -> NXobject:
+            if (cls := attrs.get('canSAS_class')) is not None:
+                if cls == 'SASdata':
+                    return _SASdata(attrs, children)
+                if cls == 'SAStransmission_spectrum':
+                    return _SAStransmission_spectrum(attrs, children)
+            return base_definitions.get(key, default)(attrs, children)
+
+        return _definition_factory
+
+
+definitions = NXcanSAS()

src/scippnexus/v2/attrs.py

@@ -0,0 +1,28 @@
+# SPDX-License-Identifier: BSD-3-Clause
+# Copyright (c) 2023 Scipp contributors (https://github.com/scipp)
+# @author Simon Heybrock
+from collections.abc import Mapping
+from typing import Any, Iterator
+
+from .._hdf5_nexus import _cset_to_encoding, _ensure_str
+
+
+class Attrs(Mapping):
+
+    def __init__(self, attrs: Mapping):
+        self._base_attrs = attrs
+        self._attrs = dict(attrs) if attrs else dict()
+
+    def __getitem__(self, name: str) -> Any:
+        attr = self._attrs[name]
+        # Is this check for string attributes sufficient? Is there a better way?
+        if isinstance(attr, (str, bytes)):
+            cset = self._base_attrs.get_id(name.encode("utf-8")).get_type().get_cset()
+            return _ensure_str(attr, _cset_to_encoding(cset))
+        return attr
+
+    def __iter__(self) -> Iterator[str]:
+        return iter(self._attrs)
+
+    def __len__(self) -> int:
+        return len(self._attrs)