From 66bf38a71426e6e5baaf0434c9c9791e70523b1c Mon Sep 17 00:00:00 2001
From: Simon Heybrock <simon.heybrock@ess.eu>
Date: Tue, 5 Apr 2022 11:10:48 +0200
Subject: [PATCH 01/25] Try to find datetime offset for all fields with time
 unit

---
 setup.cfg                      |  1 +
 src/scippnexus/_common.py      | 24 ++------------------
 src/scippnexus/nxevent_data.py |  8 ++-----
 src/scippnexus/nxlog.py        | 18 +++++++--------
 src/scippnexus/nxobject.py     | 40 ++++++++++++++++++++++++++++++++++
 5 files changed, 54 insertions(+), 37 deletions(-)

diff --git a/setup.cfg b/setup.cfg
index 4fd0e862..ba6da2ed 100644
--- a/setup.cfg
+++ b/setup.cfg
@@ -17,6 +17,7 @@ package_dir =
     = src
 packages = find:
 install_requires =
+    dateutil
     scipp>=0.12
     h5py
 python_requires = >=3.8
diff --git a/src/scippnexus/_common.py b/src/scippnexus/_common.py
index b0209b61..aaf4e57b 100644
--- a/src/scippnexus/_common.py
+++ b/src/scippnexus/_common.py
@@ -9,7 +9,6 @@
 
 def convert_time_to_datetime64(
         raw_times: sc.Variable,
-        group_path: str,
         start: str = None,
         scaling_factor: Union[float, np.float_] = None) -> sc.Variable:
     """
@@ -25,8 +24,6 @@ def convert_time_to_datetime64(
 
     Args:
         raw_times: The raw time data from a nexus file.
-        group_path: The path within the nexus file to the log being read.
-            Used to generate warnings if loading the log fails.
         start: Optional, the start time of the log in an ISO8601
             string. If not provided, defaults to the beginning of the
             unix epoch (1970-01-01T00:00:00).
@@ -34,31 +31,14 @@ def convert_time_to_datetime64(
             time series data and the unit of the raw_times Variable. If
             not provided, defaults to 1 (a no-op scaling factor).
     """
-    try:
-        raw_times_ns = sc.to_unit(raw_times, sc.units.ns, copy=False)
-    except sc.UnitError:
-        raise sc.UnitError(
-            f"The units of time in the entry at "
-            f"'{group_path}/time{{units}}' must be convertible to seconds, "
-            f"but this cannot be done for '{raw_times.unit}'. Skipping "
-            f"loading group at '{group_path}'.")
-
-    try:
-        _start_ts = sc.scalar(value=np.datetime64(start or "1970-01-01T00:00:00"),
-                              unit=sc.units.ns,
-                              dtype=sc.DType.datetime64)
-    except ValueError:
-        raise ValueError(
-            f"The date string '{start}' in the entry at "
-            f"'{group_path}/time@start' failed to parse as an ISO8601 date. "
-            f"Skipping loading group at '{group_path}'")
+    raw_times_ns = sc.to_unit(raw_times, sc.units.ns, copy=False)
 
     if scaling_factor is None:
         times = raw_times_ns.astype(sc.DType.int64, copy=False)
     else:
         _scale = sc.scalar(value=scaling_factor)
         times = (raw_times_ns * _scale).astype(sc.DType.int64, copy=False)
-    return _start_ts + times
+    return start + times
 
 
 def _to_canonical_select(dims: List[str], select: ScippIndex) -> ScippIndex:
diff --git a/src/scippnexus/nxevent_data.py b/src/scippnexus/nxevent_data.py
index c08dd8e7..b30b7d75 100644
--- a/src/scippnexus/nxevent_data.py
+++ b/src/scippnexus/nxevent_data.py
@@ -5,7 +5,7 @@
 import numpy as np
 import scipp as sc
 
-from ._common import to_plain_index, convert_time_to_datetime64
+from ._common import to_plain_index
 from .nxobject import NXobject, ScippIndex, NexusStructureError
 
 _event_dimension = "event"
@@ -57,11 +57,7 @@ def _getitem(self, select: ScippIndex) -> sc.DataArray:
             index = slice(start, stop, stride)
 
         event_index = self['event_index'][index].values
-        event_time_zero = self['event_time_zero']
-        event_time_zero = convert_time_to_datetime64(
-            event_time_zero[index],
-            start=event_time_zero.attrs.get('offset'),
-            group_path=self.name)
+        event_time_zero = self['event_time_zero'][index]
 
         num_event = self["event_time_offset"].shape[0]
         # Some files contain uint64 "max" indices, which turn into negatives during
diff --git a/src/scippnexus/nxlog.py b/src/scippnexus/nxlog.py
index 35f48060..289a842f 100644
--- a/src/scippnexus/nxlog.py
+++ b/src/scippnexus/nxlog.py
@@ -39,15 +39,15 @@ def _getitem(self, select: ScippIndex) -> sc.DataArray:
         # The 'time' field in NXlog contains extra properties 'start' and
         # 'scaling_factor' that are not handled by NXdata. These are used
         # to transform to a datetime-coord.
-        if 'time' in self:
-            if 'time' not in data.coords:
-                raise sc.DimensionError(
-                    "NXlog is time-dependent, but failed to load `time` dataset")
-            data.coords['time'] = convert_time_to_datetime64(
-                raw_times=data.coords.pop('time'),
-                start=self['time'].attrs.get('start'),
-                scaling_factor=self['time'].attrs.get('scaling_factor'),
-                group_path=self['time'].name)
+        #if 'time' in self:
+        #    if 'time' not in data.coords:
+        #        raise sc.DimensionError(
+        #            "NXlog is time-dependent, but failed to load `time` dataset")
+        #    data.coords['time'] = convert_time_to_datetime64(
+        #        raw_times=data.coords.pop('time'),
+        #        start=self['time'].attrs.get('start'),
+        #        scaling_factor=self['time'].attrs.get('scaling_factor'),
+        #        group_path=self['time'].name)
         return data
 
     def _get_field_dims(self, name: str) -> Union[None, List[str]]:
diff --git a/src/scippnexus/nxobject.py b/src/scippnexus/nxobject.py
index 9cbe7fec..f3c60002 100644
--- a/src/scippnexus/nxobject.py
+++ b/src/scippnexus/nxobject.py
@@ -6,6 +6,7 @@
 from enum import Enum, auto
 import functools
 from typing import List, Union, NoReturn, Any, Dict, Tuple, Protocol
+import dateutil
 import numpy as np
 import scipp as sc
 import h5py
@@ -14,6 +15,7 @@
 from ._hdf5_nexus import _ensure_supported_int_type, _warn_latin1_decode
 from .typing import H5Group, H5Dataset, ScippIndex
 from ._common import to_plain_index
+from ._common import convert_time_to_datetime64
 
 NXobjectIndex = Union[str, ScippIndex]
 
@@ -84,6 +86,9 @@ def __getitem__(self, name: str) -> Any:
     def __setitem__(self, name: str, val: Any):
         self._attrs[name] = val
 
+    def __iter__(self):
+        yield from self._attrs
+
     def get(self, name: str, default=None) -> Any:
         return self[name] if name in self else default
 
@@ -91,6 +96,31 @@ def keys(self):
         return self._attrs.keys()
 
 
+def _is_time(obj):
+    dummy = sc.empty(dims=[], shape=[], unit=obj.unit)
+    try:
+        dummy.to(unit='s')
+        return True
+    except sc.UnitError:
+        return False
+
+
+def _as_datetime(obj: Any):
+    if isinstance(obj, str):
+        try:
+            # datetime.fromisoformat cannot parse time zones and recommends dateutil
+            dt = dateutil.parser.isoparse(obj)
+            # NumPy and scipp cannot handle timezone information. We therefore strip it,
+            # i.e., interpret time as local time. If time is given in UTC this will lead
+            # to misleading results since we have no information about the actual time
+            # zone.
+            dt = dt.replace(tzinfo=None)
+            return sc.datetime(np.datetime64(dt), unit='ns')
+        except ValueError:
+            pass
+    return None
+
+
 class Field:
     """NeXus field.
 
@@ -137,6 +167,16 @@ def __getitem__(self, select) -> sc.Variable:
             self._dataset.read_direct(variable.values, source_sel=index)
         else:
             variable.values = self._dataset[index]
+        if _is_time(variable):
+            starts = []
+            for name in self.attrs:
+                if (dt := _as_datetime(self.attrs[name])) is not None:
+                    starts.append(dt)
+            if len(starts) == 1:
+                variable = convert_time_to_datetime64(
+                    variable,
+                    start=starts[0],
+                    scaling_factor=self.attrs.get('scaling_factor'))
         return variable
 
     def __repr__(self) -> str:

From 4ffef5614db3cce0af710f02090ce5f9ba6a429f Mon Sep 17 00:00:00 2001
From: Simon Heybrock <simon.heybrock@ess.eu>
Date: Tue, 5 Apr 2022 11:22:56 +0200
Subject: [PATCH 02/25] Add mechanism for forcing Field to datetime

---
 src/scippnexus/nxlog.py    | 19 +++++--------------
 src/scippnexus/nxobject.py | 10 +++++++---
 2 files changed, 12 insertions(+), 17 deletions(-)

diff --git a/src/scippnexus/nxlog.py b/src/scippnexus/nxlog.py
index 289a842f..7037401a 100644
--- a/src/scippnexus/nxlog.py
+++ b/src/scippnexus/nxlog.py
@@ -35,20 +35,11 @@ def _nxbase(self) -> NXdata:
         return NXdata(self._group, signal_name_default='value', axes=axes)
 
     def _getitem(self, select: ScippIndex) -> sc.DataArray:
-        data = self._nxbase[select]
-        # The 'time' field in NXlog contains extra properties 'start' and
-        # 'scaling_factor' that are not handled by NXdata. These are used
-        # to transform to a datetime-coord.
-        #if 'time' in self:
-        #    if 'time' not in data.coords:
-        #        raise sc.DimensionError(
-        #            "NXlog is time-dependent, but failed to load `time` dataset")
-        #    data.coords['time'] = convert_time_to_datetime64(
-        #        raw_times=data.coords.pop('time'),
-        #        start=self['time'].attrs.get('start'),
-        #        scaling_factor=self['time'].attrs.get('scaling_factor'),
-        #        group_path=self['time'].name)
-        return data
+        base = self._nxbase
+        # Field loads datetime offset attributes automatically, but for NXlog this
+        # may apparently be omitted and must then interpreted as relative to epoch.
+        base.child_params['time'] = {'is_time': True}
+        return base[select]
 
     def _get_field_dims(self, name: str) -> Union[None, List[str]]:
         return self._nxbase._get_field_dims(name)
diff --git a/src/scippnexus/nxobject.py b/src/scippnexus/nxobject.py
index f3c60002..93217a6b 100644
--- a/src/scippnexus/nxobject.py
+++ b/src/scippnexus/nxobject.py
@@ -126,9 +126,10 @@ class Field:
 
     In HDF5 fields are represented as dataset.
     """
-    def __init__(self, dataset: H5Dataset, dims=None):
+    def __init__(self, dataset: H5Dataset, dims=None, is_time=None):
         self._dataset = dataset
         self._shape = list(self._dataset.shape)
+        self._is_time = is_time
         # NeXus treats [] and [1] interchangeably. In general this is ill-defined, but
         # the best we can do appears to be squeezing unless the file provides names for
         # dimensions. The shape property of this class does thus not necessarily return
@@ -167,11 +168,13 @@ def __getitem__(self, select) -> sc.Variable:
             self._dataset.read_direct(variable.values, source_sel=index)
         else:
             variable.values = self._dataset[index]
-        if _is_time(variable):
+        if self._is_time or _is_time(variable):
             starts = []
             for name in self.attrs:
                 if (dt := _as_datetime(self.attrs[name])) is not None:
                     starts.append(dt)
+            if self._is_time and len(starts) == 0:
+                starts.append(sc.epoch(unit='ns'))
             if len(starts) == 1:
                 variable = convert_time_to_datetime64(
                     variable,
@@ -248,6 +251,7 @@ class NXobject:
     """
     def __init__(self, group: H5Group):
         self._group = group
+        self.child_params = {}
 
     def _get_child(
             self,
@@ -260,7 +264,7 @@ def _get_child(
             item = self._group[name]
             if hasattr(item, 'shape'):
                 dims = self._get_field_dims(name) if use_field_dims else None
-                return Field(item, dims=dims)
+                return Field(item, dims=dims, **self.child_params.get(name, {}))
             else:
                 return _make(item)
         da = self._getitem(name)

From 516ea4c238823e9a40ecf4d415836f67583ea4ac Mon Sep 17 00:00:00 2001
From: Simon Heybrock <simon.heybrock@ess.eu>
Date: Tue, 5 Apr 2022 11:34:02 +0200
Subject: [PATCH 03/25] Support writing datetimes

---
 src/scippnexus/nxobject.py |  5 +++++
 tests/nxdata_test.py       | 10 ++++++++++
 2 files changed, 15 insertions(+)

diff --git a/src/scippnexus/nxobject.py b/src/scippnexus/nxobject.py
index 93217a6b..874519d2 100644
--- a/src/scippnexus/nxobject.py
+++ b/src/scippnexus/nxobject.py
@@ -365,9 +365,14 @@ def create_field(self, name: str, data: DimensionedArray, **kwargs) -> Field:
         values = data.values
         if data.dtype == sc.DType.string:
             values = np.array(data.values, dtype=object)
+        elif data.dtype == sc.DType.datetime64:
+            start = sc.epoch(unit=data.unit)
+            values = (data - start).values
         dataset = self._group.create_dataset(name, data=values, **kwargs)
         if data.unit is not None:
             dataset.attrs['units'] = str(data.unit)
+        if data.dtype == sc.DType.datetime64:
+            dataset.attrs['start'] = str(start.value)
         return Field(dataset, data.dims)
 
     def create_class(self, name: str, nx_class: NX_class) -> NXobject:
diff --git a/tests/nxdata_test.py b/tests/nxdata_test.py
index 9da98752..a1dd6e09 100644
--- a/tests/nxdata_test.py
+++ b/tests/nxdata_test.py
@@ -1,4 +1,5 @@
 import h5py
+import numpy as np
 import scipp as sc
 from scippnexus import NXroot, NX_class
 import pytest
@@ -213,6 +214,15 @@ def test_create_field_from_variable(nxroot, unit):
     assert sc.identical(loaded, var.rename(xx=loaded.dim))
 
 
+def test_create_datetime_field_from_variable(nxroot):
+    var = sc.datetime(np.datetime64('now'), unit='ns') + sc.arange(
+        'time', 1, 4, dtype='int64', unit='ns')
+    nxroot.create_field('field', var)
+    loaded = nxroot['field'][...]
+    # Nexus does not support storing dim labels
+    assert sc.identical(loaded, var.rename(time=loaded.dim))
+
+
 @pytest.mark.parametrize("nx_class", [NX_class.NXdata, NX_class.NXlog])
 def test_create_class(nxroot, nx_class):
     group = nxroot.create_class('group', nx_class)

From 29987a37430f886b5eccda9b6b4445ea012864b7 Mon Sep 17 00:00:00 2001
From: Simon Heybrock <simon.heybrock@ess.eu>
Date: Tue, 5 Apr 2022 12:04:53 +0200
Subject: [PATCH 04/25] Test datetime loading and preserve precision

---
 src/scippnexus/_common.py  | 15 ++++++++++++---
 src/scippnexus/nxobject.py |  6 +-----
 tests/nxdata_test.py       | 17 +++++++++++++++++
 3 files changed, 30 insertions(+), 8 deletions(-)

diff --git a/src/scippnexus/_common.py b/src/scippnexus/_common.py
index aaf4e57b..ea691d56 100644
--- a/src/scippnexus/_common.py
+++ b/src/scippnexus/_common.py
@@ -31,13 +31,22 @@ def convert_time_to_datetime64(
             time series data and the unit of the raw_times Variable. If
             not provided, defaults to 1 (a no-op scaling factor).
     """
-    raw_times_ns = sc.to_unit(raw_times, sc.units.ns, copy=False)
+    if (raw_times.dtype
+            in (sc.DType.float64, sc.DType.float32)) or scaling_factor is not None:
+        unit = sc.units.ns
+    else:
+        # determine more precise unit
+        ratio = sc.scalar(1.0, unit=start.unit) / sc.scalar(
+            1.0, unit=raw_times.unit).to(unit=start.unit)
+        unit = start.unit if ratio.value < 1.0 else raw_times.unit
+
+    raw_times = raw_times.to(unit=unit, copy=False)
 
     if scaling_factor is None:
-        times = raw_times_ns.astype(sc.DType.int64, copy=False)
+        times = raw_times.astype(sc.DType.int64, copy=False)
     else:
         _scale = sc.scalar(value=scaling_factor)
-        times = (raw_times_ns * _scale).astype(sc.DType.int64, copy=False)
+        times = (raw_times * _scale).astype(sc.DType.int64, copy=False)
     return start + times
 
 
diff --git a/src/scippnexus/nxobject.py b/src/scippnexus/nxobject.py
index 874519d2..ef891329 100644
--- a/src/scippnexus/nxobject.py
+++ b/src/scippnexus/nxobject.py
@@ -6,7 +6,6 @@
 from enum import Enum, auto
 import functools
 from typing import List, Union, NoReturn, Any, Dict, Tuple, Protocol
-import dateutil
 import numpy as np
 import scipp as sc
 import h5py
@@ -108,14 +107,11 @@ def _is_time(obj):
 def _as_datetime(obj: Any):
     if isinstance(obj, str):
         try:
-            # datetime.fromisoformat cannot parse time zones and recommends dateutil
-            dt = dateutil.parser.isoparse(obj)
             # NumPy and scipp cannot handle timezone information. We therefore strip it,
             # i.e., interpret time as local time. If time is given in UTC this will lead
             # to misleading results since we have no information about the actual time
             # zone.
-            dt = dt.replace(tzinfo=None)
-            return sc.datetime(np.datetime64(dt), unit='ns')
+            return sc.datetime(np.datetime64(obj))
         except ValueError:
             pass
     return None
diff --git a/tests/nxdata_test.py b/tests/nxdata_test.py
index a1dd6e09..d3060c77 100644
--- a/tests/nxdata_test.py
+++ b/tests/nxdata_test.py
@@ -301,3 +301,20 @@ def test_unnamed_extra_dims_of_multidim_coords_are_squeezed(nxroot):
     assert data['xx'].ndim == 1
     assert data['xx'].shape == [2]
     assert sc.identical(data['xx'][...], xx['ignored', 0])
+
+def test_fields_with_datetime_attribute_are_loaded_as_datetime(nxroot):
+    da = sc.DataArray(
+        sc.epoch(unit='s') +
+        sc.array(dims=['xx', 'yy'], unit='s', values=[[1, 2, 3], [4, 5, 6]]))
+    da.coords['xx'] = da.data['yy', 0]
+    da.coords['xx2'] = da.data['yy', 1]
+    da.coords['yy'] = da.data['xx', 0]
+    data = nxroot.create_class('data1', NX_class.NXdata)
+    data.attrs['axes'] = da.dims
+    data.attrs['signal'] = 'signal'
+    data.create_field('signal', da.data)
+    data.create_field('xx', da.coords['xx'])
+    data.create_field('xx2', da.coords['xx2'])
+    data.create_field('yy', da.coords['yy'])
+    print(data[...], da)
+    assert sc.identical(data[...], da)

From 3033339798ce5a92ed32f1d525a60a68734b3a2c Mon Sep 17 00:00:00 2001
From: Simon Heybrock <simon.heybrock@ess.eu>
Date: Tue, 5 Apr 2022 12:43:38 +0200
Subject: [PATCH 05/25] Test datetime handlign and strip timezone

---
 setup.cfg                  |  1 -
 src/scippnexus/_common.py  |  2 +-
 src/scippnexus/nxobject.py |  7 ++++++
 tests/nexus_test.py        | 51 ++++++++++++++++++++++++++++++++++++++
 tests/nxdata_test.py       |  1 -
 5 files changed, 59 insertions(+), 3 deletions(-)

diff --git a/setup.cfg b/setup.cfg
index ba6da2ed..4fd0e862 100644
--- a/setup.cfg
+++ b/setup.cfg
@@ -17,7 +17,6 @@ package_dir =
     = src
 packages = find:
 install_requires =
-    dateutil
     scipp>=0.12
     h5py
 python_requires = >=3.8
diff --git a/src/scippnexus/_common.py b/src/scippnexus/_common.py
index ea691d56..5efa811e 100644
--- a/src/scippnexus/_common.py
+++ b/src/scippnexus/_common.py
@@ -47,7 +47,7 @@ def convert_time_to_datetime64(
     else:
         _scale = sc.scalar(value=scaling_factor)
         times = (raw_times * _scale).astype(sc.DType.int64, copy=False)
-    return start + times
+    return start.to(unit=unit, copy=False) + times
 
 
 def _to_canonical_select(dims: List[str], select: ScippIndex) -> ScippIndex:
diff --git a/src/scippnexus/nxobject.py b/src/scippnexus/nxobject.py
index ef891329..f60a7b94 100644
--- a/src/scippnexus/nxobject.py
+++ b/src/scippnexus/nxobject.py
@@ -2,6 +2,7 @@
 # Copyright (c) 2022 Scipp contributors (https://github.com/scipp)
 # @author Simon Heybrock
 from __future__ import annotations
+import re
 import warnings
 from enum import Enum, auto
 import functools
@@ -111,6 +112,12 @@ def _as_datetime(obj: Any):
             # i.e., interpret time as local time. If time is given in UTC this will lead
             # to misleading results since we have no information about the actual time
             # zone.
+            # Would like to use dateutil, but with Python's datetime we do not get
+            # nanosecond precision.
+            if 'T' in obj:
+                date, time = obj.split('T')
+                time = re.split('Z|\+|-', time)[0]
+                obj = f'{date}T{time}'
             return sc.datetime(np.datetime64(obj))
         except ValueError:
             pass
diff --git a/tests/nexus_test.py b/tests/nexus_test.py
index 97082219..4e03e88c 100644
--- a/tests/nexus_test.py
+++ b/tests/nexus_test.py
@@ -212,6 +212,57 @@ def test_field_of_extended_ascii_in_ascii_encoded_dataset_is_loaded_correctly():
             sc.array(dims=['dim_0'], values=["run at rot=90°", "run at rot=90°x"]))
 
 
+def test_ms_field_with_second_datetime_attribute_loaded_as_ms_datetime(nxroot):
+    nxroot['mytime'] = sc.arange('ignored', 2, unit='ms')
+    nxroot['mytime'].attrs['start_time'] = '2022-12-12T12:13:14'
+    assert sc.identical(
+        nxroot['mytime'][...],
+        sc.datetimes(dims=['dim_0'],
+                     unit='ms',
+                     values=['2022-12-12T12:13:14.000', '2022-12-12T12:13:14.001']))
+
+
+
+def test_ns_field_with_second_datetime_attribute_loaded_as_ns_datetime(nxroot):
+    nxroot['mytime'] = sc.arange('ignored', 2, unit='ns')
+    nxroot['mytime'].attrs['start_time'] = '1970-01-01T00:00:00'
+    assert sc.identical(
+        nxroot['mytime'][...],
+        sc.datetimes(
+            dims=['dim_0'],
+            unit='ns',
+            values=['1970-01-01T00:00:00.000000000', '1970-01-01T00:00:00.000000001']))
+
+
+
+def test_second_field_with_ns_datetime_attribute_loaded_as_ns_datetime(nxroot):
+    nxroot['mytime'] = sc.arange('ignored', 2, unit='s')
+    nxroot['mytime'].attrs['start_time'] = '1970-01-01T00:00:00.000000000'
+    assert sc.identical(
+        nxroot['mytime'][...],
+        sc.datetimes(dims=['dim_0'],
+                     unit='ns',
+                     values=['1970-01-01T00:00:00', '1970-01-01T00:00:01']))
+
+
+
+@pytest.mark.parametrize('timezone', ['Z', '+04', '+00', '-02', '+1130', '-0930', '+11:30', '-09:30'])
+def test_timezone_information_in_datetime_attribute_is_dropped(nxroot, timezone):
+    nxroot['mytime'] = sc.arange('ignored', 2, unit='s')
+    nxroot['mytime'].attrs['start_time'] = f'1970-01-01T00:00:00{timezone}'
+    print(
+        nxroot['mytime'][...],
+        sc.datetimes(dims=['dim_0'],
+                     unit='s',
+                     values=['1970-01-01T00:00:00', '1970-01-01T00:00:01']))
+    assert sc.identical(
+        nxroot['mytime'][...],
+        sc.datetimes(dims=['dim_0'],
+                     unit='s',
+                     values=['1970-01-01T00:00:00', '1970-01-01T00:00:01']))
+
+
+
 def create_event_data_ids_1234(group):
     group['event_id'] = sc.array(dims=[''], unit=None, values=[1, 2, 4, 1, 2, 2])
     group['event_time_offset'] = sc.array(dims=[''],
diff --git a/tests/nxdata_test.py b/tests/nxdata_test.py
index d3060c77..66e94a57 100644
--- a/tests/nxdata_test.py
+++ b/tests/nxdata_test.py
@@ -316,5 +316,4 @@ def test_fields_with_datetime_attribute_are_loaded_as_datetime(nxroot):
     data.create_field('xx', da.coords['xx'])
     data.create_field('xx2', da.coords['xx2'])
     data.create_field('yy', da.coords['yy'])
-    print(data[...], da)
     assert sc.identical(data[...], da)

From 78e2913455752f78e973638ca8ce9d1f74573186 Mon Sep 17 00:00:00 2001
From: Simon Heybrock <simon.heybrock@ess.eu>
Date: Tue, 5 Apr 2022 12:58:53 +0200
Subject: [PATCH 06/25] Update Nexus format interpretation docs

---
 .../docs/our-interpretation-of-the-nexus-format.md     | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/src/scippnexus/docs/our-interpretation-of-the-nexus-format.md b/src/scippnexus/docs/our-interpretation-of-the-nexus-format.md
index ab3931be..af8576e9 100644
--- a/src/scippnexus/docs/our-interpretation-of-the-nexus-format.md
+++ b/src/scippnexus/docs/our-interpretation-of-the-nexus-format.md
@@ -47,6 +47,16 @@ More concretely this means that, e.g., for loading an `NXdetector` from a NH, th
 
 If the above yields no more than one item, the group can be loaded.
 
+## Datetime fields
+
+HDF5 does not support storing date and time information such as `np.datetime64`.
+`NXlog` and `NXevent_data` specify specific attributes for fields that have to be interpreted as date and time, in particular [NXlog/time@start](https://manual.nexusformat.org/classes/base_classes/NXlog.html#nxlog-time-start-attribute) and [NXevent_data/event_time_offset@offset](https://manual.nexusformat.org/classes/base_classes/NXevent_data.html#nxevent-data-event-time-offset-field).
+No *general* definition or intention is documented in the NF, but according to TR this is nevertheless standard.
+Due to the attribute naming mismatch in the two cases where it *is* specified we need to assume that naming is arbitrary.
+Therefore, we search *all* attributes of a field for a date and time offset, provided that the field's unit is a time unit.
+It is unclear what should be done in the case of multiple matches.
+As of April 2022 we ignore the offset in this case, since guessing which one to use based on the attribute name does not seem desirable.
+
 ## Bin edges
 
 For [NXdetector](https://manual.nexusformat.org/classes/base_classes/NXdetector.html) the NF defines a [time_of_flight](https://manual.nexusformat.org/classes/base_classes/NXdetector.html#nxdetector-time-of-flight-field) field, exceeding the data shape by one, i.e., it is meant as bin-edges.

From c7efa15377a37b896ec87fc937c4923f8bf5237a Mon Sep 17 00:00:00 2001
From: Simon Heybrock <simon.heybrock@ess.eu>
Date: Tue, 5 Apr 2022 13:01:47 +0200
Subject: [PATCH 07/25] Use re.escape

---
 src/scippnexus/nxobject.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/scippnexus/nxobject.py b/src/scippnexus/nxobject.py
index f60a7b94..efd04ee6 100644
--- a/src/scippnexus/nxobject.py
+++ b/src/scippnexus/nxobject.py
@@ -116,7 +116,7 @@ def _as_datetime(obj: Any):
             # nanosecond precision.
             if 'T' in obj:
                 date, time = obj.split('T')
-                time = re.split('Z|\+|-', time)[0]
+                time = re.split(f'Z|{re.escape("+")}|-', time)[0]
                 obj = f'{date}T{time}'
             return sc.datetime(np.datetime64(obj))
         except ValueError:

From 5c70f3fdd3faed53c72c4f528a57c02f3aee33f9 Mon Sep 17 00:00:00 2001
From: Simon Heybrock <simon.heybrock@ess.eu>
Date: Tue, 5 Apr 2022 13:06:23 +0200
Subject: [PATCH 08/25] Remove unused

---
 src/scippnexus/nxlog.py | 1 -
 1 file changed, 1 deletion(-)

diff --git a/src/scippnexus/nxlog.py b/src/scippnexus/nxlog.py
index 7037401a..e227723b 100644
--- a/src/scippnexus/nxlog.py
+++ b/src/scippnexus/nxlog.py
@@ -4,7 +4,6 @@
 
 from typing import List, Union
 import scipp as sc
-from ._common import convert_time_to_datetime64
 from .nxobject import NXobject, ScippIndex
 from .nxdata import NXdata
 

From 87a372e03be388aed5ba0fabd78bb65660e1eb5d Mon Sep 17 00:00:00 2001
From: SimonHeybrock <SimonHeybrock@users.noreply.github.com>
Date: Tue, 5 Apr 2022 11:06:51 +0000
Subject: [PATCH 09/25] Apply automatic formatting

---
 tests/nexus_test.py  | 7 ++-----
 tests/nxdata_test.py | 1 +
 2 files changed, 3 insertions(+), 5 deletions(-)

diff --git a/tests/nexus_test.py b/tests/nexus_test.py
index 4e03e88c..3938c344 100644
--- a/tests/nexus_test.py
+++ b/tests/nexus_test.py
@@ -222,7 +222,6 @@ def test_ms_field_with_second_datetime_attribute_loaded_as_ms_datetime(nxroot):
                      values=['2022-12-12T12:13:14.000', '2022-12-12T12:13:14.001']))
 
 
-
 def test_ns_field_with_second_datetime_attribute_loaded_as_ns_datetime(nxroot):
     nxroot['mytime'] = sc.arange('ignored', 2, unit='ns')
     nxroot['mytime'].attrs['start_time'] = '1970-01-01T00:00:00'
@@ -234,7 +233,6 @@ def test_ns_field_with_second_datetime_attribute_loaded_as_ns_datetime(nxroot):
             values=['1970-01-01T00:00:00.000000000', '1970-01-01T00:00:00.000000001']))
 
 
-
 def test_second_field_with_ns_datetime_attribute_loaded_as_ns_datetime(nxroot):
     nxroot['mytime'] = sc.arange('ignored', 2, unit='s')
     nxroot['mytime'].attrs['start_time'] = '1970-01-01T00:00:00.000000000'
@@ -245,8 +243,8 @@ def test_second_field_with_ns_datetime_attribute_loaded_as_ns_datetime(nxroot):
                      values=['1970-01-01T00:00:00', '1970-01-01T00:00:01']))
 
 
-
-@pytest.mark.parametrize('timezone', ['Z', '+04', '+00', '-02', '+1130', '-0930', '+11:30', '-09:30'])
+@pytest.mark.parametrize(
+    'timezone', ['Z', '+04', '+00', '-02', '+1130', '-0930', '+11:30', '-09:30'])
 def test_timezone_information_in_datetime_attribute_is_dropped(nxroot, timezone):
     nxroot['mytime'] = sc.arange('ignored', 2, unit='s')
     nxroot['mytime'].attrs['start_time'] = f'1970-01-01T00:00:00{timezone}'
@@ -262,7 +260,6 @@ def test_timezone_information_in_datetime_attribute_is_dropped(nxroot, timezone)
                      values=['1970-01-01T00:00:00', '1970-01-01T00:00:01']))
 
 
-
 def create_event_data_ids_1234(group):
     group['event_id'] = sc.array(dims=[''], unit=None, values=[1, 2, 4, 1, 2, 2])
     group['event_time_offset'] = sc.array(dims=[''],
diff --git a/tests/nxdata_test.py b/tests/nxdata_test.py
index 66e94a57..077bcf10 100644
--- a/tests/nxdata_test.py
+++ b/tests/nxdata_test.py
@@ -302,6 +302,7 @@ def test_unnamed_extra_dims_of_multidim_coords_are_squeezed(nxroot):
     assert data['xx'].shape == [2]
     assert sc.identical(data['xx'][...], xx['ignored', 0])
 
+
 def test_fields_with_datetime_attribute_are_loaded_as_datetime(nxroot):
     da = sc.DataArray(
         sc.epoch(unit='s') +

From ade5af88dcccc745e51e54308d87d6718c483241 Mon Sep 17 00:00:00 2001
From: Simon Heybrock <simon.heybrock@ess.eu>
Date: Tue, 5 Apr 2022 15:24:06 +0200
Subject: [PATCH 10/25] Remove print

---
 tests/nexus_test.py | 5 -----
 1 file changed, 5 deletions(-)

diff --git a/tests/nexus_test.py b/tests/nexus_test.py
index 3938c344..ea4d1853 100644
--- a/tests/nexus_test.py
+++ b/tests/nexus_test.py
@@ -248,11 +248,6 @@ def test_second_field_with_ns_datetime_attribute_loaded_as_ns_datetime(nxroot):
 def test_timezone_information_in_datetime_attribute_is_dropped(nxroot, timezone):
     nxroot['mytime'] = sc.arange('ignored', 2, unit='s')
     nxroot['mytime'].attrs['start_time'] = f'1970-01-01T00:00:00{timezone}'
-    print(
-        nxroot['mytime'][...],
-        sc.datetimes(dims=['dim_0'],
-                     unit='s',
-                     values=['1970-01-01T00:00:00', '1970-01-01T00:00:01']))
     assert sc.identical(
         nxroot['mytime'][...],
         sc.datetimes(dims=['dim_0'],

From 15c74b590274cbb64a00e0407619d8a67bc41e92 Mon Sep 17 00:00:00 2001
From: Simon Heybrock <simon.heybrock@ess.eu>
Date: Tue, 5 Apr 2022 15:27:25 +0200
Subject: [PATCH 11/25] Test that multiple candidate offsets are simply ignored

---
 tests/nexus_test.py | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/tests/nexus_test.py b/tests/nexus_test.py
index ea4d1853..630b6cf0 100644
--- a/tests/nexus_test.py
+++ b/tests/nexus_test.py
@@ -255,6 +255,14 @@ def test_timezone_information_in_datetime_attribute_is_dropped(nxroot, timezone)
                      values=['1970-01-01T00:00:00', '1970-01-01T00:00:01']))
 
 
+def test_loads_bare_timestamps_if_multiple_candidate_datetime_offsets_found(nxroot):
+    offsets = sc.arange('ignored', 2, unit='ms')
+    nxroot['mytime'] = offsets
+    nxroot['mytime'].attrs['offset'] = '2022-12-12T12:13:14'
+    nxroot['mytime'].attrs['start_time'] = '2022-12-12T12:13:15'
+    assert sc.identical(nxroot['mytime'][...], offsets.rename(ignored='dim_0'))
+
+
 def create_event_data_ids_1234(group):
     group['event_id'] = sc.array(dims=[''], unit=None, values=[1, 2, 4, 1, 2, 2])
     group['event_time_offset'] = sc.array(dims=[''],

From b807db854fab69ddfc32fab1766e8cde51ad79ca Mon Sep 17 00:00:00 2001
From: Simon Heybrock <simon.heybrock@ess.eu>
Date: Wed, 6 Apr 2022 12:48:43 +0200
Subject: [PATCH 12/25] Improve readability

---
 src/scippnexus/_common.py | 10 ++++------
 1 file changed, 4 insertions(+), 6 deletions(-)

diff --git a/src/scippnexus/_common.py b/src/scippnexus/_common.py
index 5efa811e..fad7edc7 100644
--- a/src/scippnexus/_common.py
+++ b/src/scippnexus/_common.py
@@ -40,14 +40,12 @@ def convert_time_to_datetime64(
             1.0, unit=raw_times.unit).to(unit=start.unit)
         unit = start.unit if ratio.value < 1.0 else raw_times.unit
 
-    raw_times = raw_times.to(unit=unit, copy=False)
-
     if scaling_factor is None:
-        times = raw_times.astype(sc.DType.int64, copy=False)
+        times = raw_times
     else:
-        _scale = sc.scalar(value=scaling_factor)
-        times = (raw_times * _scale).astype(sc.DType.int64, copy=False)
-    return start.to(unit=unit, copy=False) + times
+        times = raw_times * sc.scalar(value=scaling_factor)
+    return start.to(unit=unit, copy=False) + times.to(
+        dtype=sc.DType.int64, unit=unit, copy=False)
 
 
 def _to_canonical_select(dims: List[str], select: ScippIndex) -> ScippIndex:

From 4c7a89806bde3323759dc9df1ba11e98f2dc9da7 Mon Sep 17 00:00:00 2001
From: Simon Heybrock <simon.heybrock@ess.eu>
Date: Wed, 6 Apr 2022 12:50:27 +0200
Subject: [PATCH 13/25] Clarify docs

---
 src/scippnexus/docs/our-interpretation-of-the-nexus-format.md | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/scippnexus/docs/our-interpretation-of-the-nexus-format.md b/src/scippnexus/docs/our-interpretation-of-the-nexus-format.md
index af8576e9..022d8a0f 100644
--- a/src/scippnexus/docs/our-interpretation-of-the-nexus-format.md
+++ b/src/scippnexus/docs/our-interpretation-of-the-nexus-format.md
@@ -55,7 +55,7 @@ No *general* definition or intention is documented in the NF, but according to T
 Due to the attribute naming mismatch in the two cases where it *is* specified we need to assume that naming is arbitrary.
 Therefore, we search *all* attributes of a field for a date and time offset, provided that the field's unit is a time unit.
 It is unclear what should be done in the case of multiple matches.
-As of April 2022 we ignore the offset in this case, since guessing which one to use based on the attribute name does not seem desirable.
+As of April 2022 we ignore the date and time offsets in this case, since guessing which one to use based on the attribute name does not seem desirable.
 
 ## Bin edges
 

From bc2d8de24cd13cd546ae57ebf74544ea91ca3951 Mon Sep 17 00:00:00 2001
From: Simon Heybrock <simon.heybrock@ess.eu>
Date: Wed, 6 Apr 2022 12:57:35 +0200
Subject: [PATCH 14/25] Simpler regex

---
 src/scippnexus/nxobject.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/scippnexus/nxobject.py b/src/scippnexus/nxobject.py
index efd04ee6..1cb04d08 100644
--- a/src/scippnexus/nxobject.py
+++ b/src/scippnexus/nxobject.py
@@ -116,7 +116,7 @@ def _as_datetime(obj: Any):
             # nanosecond precision.
             if 'T' in obj:
                 date, time = obj.split('T')
-                time = re.split(f'Z|{re.escape("+")}|-', time)[0]
+                time = re.split(r'Z|\+|-', time)[0]
                 obj = f'{date}T{time}'
             return sc.datetime(np.datetime64(obj))
         except ValueError:

From 0422010f765380f9cd453402653cd9ef5b984849 Mon Sep 17 00:00:00 2001
From: Simon Heybrock <simon.heybrock@ess.eu>
Date: Thu, 7 Apr 2022 15:52:29 +0200
Subject: [PATCH 15/25] Apply instead of drop timezone offset

---
 src/scippnexus/nxobject.py | 16 +++++++++++++---
 1 file changed, 13 insertions(+), 3 deletions(-)

diff --git a/src/scippnexus/nxobject.py b/src/scippnexus/nxobject.py
index 1cb04d08..bc812649 100644
--- a/src/scippnexus/nxobject.py
+++ b/src/scippnexus/nxobject.py
@@ -4,6 +4,7 @@
 from __future__ import annotations
 import re
 import warnings
+import dateutil
 from enum import Enum, auto
 import functools
 from typing import List, Union, NoReturn, Any, Dict, Tuple, Protocol
@@ -116,9 +117,18 @@ def _as_datetime(obj: Any):
             # nanosecond precision.
             if 'T' in obj:
                 date, time = obj.split('T')
-                time = re.split(r'Z|\+|-', time)[0]
-                obj = f'{date}T{time}'
-            return sc.datetime(np.datetime64(obj))
+                time_and_timezone_offset = re.split(r'Z|\+|-', time)
+                time = time_and_timezone_offset[0]
+                if len(time_and_timezone_offset) == 1:
+                    dt = np.datetime64(f'{date}T{time}')
+                else:
+                    timezone_aware = dateutil.parser.isoparse(obj)
+                    offset = timezone_aware.replace(tzinfo=dateutil.tz.tzutc())
+                    delta = timezone_aware - offset
+                    dt = np.datetime64(f'{date}T{time}') + delta
+            else:
+                dt = np.datetime64(obj)
+            return sc.datetime(dt)
         except ValueError:
             pass
     return None

From b0bbe6ceb9e2a3762f3a1889756621bdcddd3c8b Mon Sep 17 00:00:00 2001
From: Simon Heybrock <simon.heybrock@ess.eu>
Date: Thu, 7 Apr 2022 16:24:22 +0200
Subject: [PATCH 16/25] Fixes and update tests

---
 src/scippnexus/nxobject.py | 17 +++++++++--------
 tests/nexus_test.py        | 19 +++++++++----------
 2 files changed, 18 insertions(+), 18 deletions(-)

diff --git a/src/scippnexus/nxobject.py b/src/scippnexus/nxobject.py
index bc812649..10acaec1 100644
--- a/src/scippnexus/nxobject.py
+++ b/src/scippnexus/nxobject.py
@@ -119,15 +119,16 @@ def _as_datetime(obj: Any):
                 date, time = obj.split('T')
                 time_and_timezone_offset = re.split(r'Z|\+|-', time)
                 time = time_and_timezone_offset[0]
-                if len(time_and_timezone_offset) == 1:
-                    dt = np.datetime64(f'{date}T{time}')
-                else:
-                    timezone_aware = dateutil.parser.isoparse(obj)
-                    offset = timezone_aware.replace(tzinfo=dateutil.tz.tzutc())
-                    delta = timezone_aware - offset
-                    dt = np.datetime64(f'{date}T{time}') + delta
+                dt = sc.datetime(np.datetime64(f'{date}T{time}'))
+                if len(time_and_timezone_offset) > 1:
+                    utcoffset = dateutil.parser.isoparse(obj).utcoffset()
+                    seconds = sc.scalar(value=utcoffset.total_seconds(),
+                                        unit='s',
+                                        dtype='int64')
+                    dt -= seconds.to(unit=dt.unit)
+                return dt
             else:
-                dt = np.datetime64(obj)
+                return sc.datetime(np.datetime64(obj))
             return sc.datetime(dt)
         except ValueError:
             pass
diff --git a/tests/nexus_test.py b/tests/nexus_test.py
index 630b6cf0..4f66c8d4 100644
--- a/tests/nexus_test.py
+++ b/tests/nexus_test.py
@@ -243,16 +243,15 @@ def test_second_field_with_ns_datetime_attribute_loaded_as_ns_datetime(nxroot):
                      values=['1970-01-01T00:00:00', '1970-01-01T00:00:01']))
 
 
-@pytest.mark.parametrize(
-    'timezone', ['Z', '+04', '+00', '-02', '+1130', '-0930', '+11:30', '-09:30'])
-def test_timezone_information_in_datetime_attribute_is_dropped(nxroot, timezone):
-    nxroot['mytime'] = sc.arange('ignored', 2, unit='s')
-    nxroot['mytime'].attrs['start_time'] = f'1970-01-01T00:00:00{timezone}'
-    assert sc.identical(
-        nxroot['mytime'][...],
-        sc.datetimes(dims=['dim_0'],
-                     unit='s',
-                     values=['1970-01-01T00:00:00', '1970-01-01T00:00:01']))
+@pytest.mark.parametrize('timezone,hhmm', [('Z', '12:00'), ('+04', '08:00'),
+                                           ('+00', '12:00'), ('-02', '14:00'),
+                                           ('+1130', '00:30'), ('-0930', '21:30'),
+                                           ('+11:30', '00:30'), ('-09:30', '21:30')])
+def test_timezone_information_in_datetime_attribute_is_applied(nxroot, timezone, hhmm):
+    nxroot['mytime'] = sc.scalar(value=3, unit='s')
+    nxroot['mytime'].attrs['start_time'] = f'1970-01-01T12:00:00{timezone}'
+    assert sc.identical(nxroot['mytime'][...],
+                        sc.datetime(unit='s', value=f'1970-01-01T{hhmm}:03'))
 
 
 def test_loads_bare_timestamps_if_multiple_candidate_datetime_offsets_found(nxroot):

From b59d5ee68c8a8cd0366795a6cb75904545a804e6 Mon Sep 17 00:00:00 2001
From: Simon Heybrock <simon.heybrock@ess.eu>
Date: Thu, 7 Apr 2022 16:26:58 +0200
Subject: [PATCH 17/25] Test timezone and ns precision

---
 tests/nexus_test.py | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/tests/nexus_test.py b/tests/nexus_test.py
index 4f66c8d4..137ffbc0 100644
--- a/tests/nexus_test.py
+++ b/tests/nexus_test.py
@@ -254,6 +254,13 @@ def test_timezone_information_in_datetime_attribute_is_applied(nxroot, timezone,
                         sc.datetime(unit='s', value=f'1970-01-01T{hhmm}:03'))
 
 
+def test_timezone_information_in_datetime_attribute_preserves_ns_precision(nxroot):
+    nxroot['mytime'] = sc.scalar(value=3, unit='s')
+    nxroot['mytime'].attrs['start_time'] = f'1970-01-01T12:00:00.123456789+0200'
+    assert sc.identical(nxroot['mytime'][...],
+                        sc.datetime(unit='ns', value=f'1970-01-01T10:00:03.123456789'))
+
+
 def test_loads_bare_timestamps_if_multiple_candidate_datetime_offsets_found(nxroot):
     offsets = sc.arange('ignored', 2, unit='ms')
     nxroot['mytime'] = offsets

From 39c9a4efeba46f9566481747e7850e021dbe80ed Mon Sep 17 00:00:00 2001
From: Simon Heybrock <simon.heybrock@ess.eu>
Date: Thu, 7 Apr 2022 16:31:14 +0200
Subject: [PATCH 18/25] Cleanup

---
 setup.cfg                  |  1 +
 src/scippnexus/nxobject.py | 26 ++++++++++++--------------
 2 files changed, 13 insertions(+), 14 deletions(-)

diff --git a/setup.cfg b/setup.cfg
index 4fd0e862..18b45e56 100644
--- a/setup.cfg
+++ b/setup.cfg
@@ -17,6 +17,7 @@ package_dir =
     = src
 packages = find:
 install_requires =
+    python-dateutil
     scipp>=0.12
     h5py
 python_requires = >=3.8
diff --git a/src/scippnexus/nxobject.py b/src/scippnexus/nxobject.py
index 10acaec1..8fb62abe 100644
--- a/src/scippnexus/nxobject.py
+++ b/src/scippnexus/nxobject.py
@@ -109,27 +109,25 @@ def _is_time(obj):
 def _as_datetime(obj: Any):
     if isinstance(obj, str):
         try:
-            # NumPy and scipp cannot handle timezone information. We therefore strip it,
-            # i.e., interpret time as local time. If time is given in UTC this will lead
-            # to misleading results since we have no information about the actual time
-            # zone.
-            # Would like to use dateutil, but with Python's datetime we do not get
-            # nanosecond precision.
-            if 'T' in obj:
+            # NumPy and scipp cannot handle timezone information. We therefore apply it,
+            # i.e., convert to UTC.
+            # Would like to use dateutil directly, but with Python's datetime we do not
+            # get nanosecond precision. Therefore we combine numpy and dateutil parsing.
+            date_only = 'T' in obj
+            if date_only:
+                return sc.datetime(np.datetime64(obj))
+            else:
                 date, time = obj.split('T')
                 time_and_timezone_offset = re.split(r'Z|\+|-', time)
                 time = time_and_timezone_offset[0]
+                # Strip timezone and parse with numpy
                 dt = sc.datetime(np.datetime64(f'{date}T{time}'))
                 if len(time_and_timezone_offset) > 1:
+                    # There is timezone info. Parse with dateutil.
                     utcoffset = dateutil.parser.isoparse(obj).utcoffset()
-                    seconds = sc.scalar(value=utcoffset.total_seconds(),
-                                        unit='s',
-                                        dtype='int64')
-                    dt -= seconds.to(unit=dt.unit)
+                    seconds = sc.scalar(value=utcoffset.total_seconds(), unit='s')
+                    dt -= seconds.to(unit=dt.unit, dtype='int64')
                 return dt
-            else:
-                return sc.datetime(np.datetime64(obj))
-            return sc.datetime(dt)
         except ValueError:
             pass
     return None

From a5aa5d347803dfb018a44a3462dadf97e764c307 Mon Sep 17 00:00:00 2001
From: Simon Heybrock <simon.heybrock@ess.eu>
Date: Thu, 7 Apr 2022 16:35:20 +0200
Subject: [PATCH 19/25] syntax

---
 tests/nexus_test.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/tests/nexus_test.py b/tests/nexus_test.py
index 137ffbc0..706efc7c 100644
--- a/tests/nexus_test.py
+++ b/tests/nexus_test.py
@@ -256,9 +256,9 @@ def test_timezone_information_in_datetime_attribute_is_applied(nxroot, timezone,
 
 def test_timezone_information_in_datetime_attribute_preserves_ns_precision(nxroot):
     nxroot['mytime'] = sc.scalar(value=3, unit='s')
-    nxroot['mytime'].attrs['start_time'] = f'1970-01-01T12:00:00.123456789+0200'
+    nxroot['mytime'].attrs['start_time'] = '1970-01-01T12:00:00.123456789+0200'
     assert sc.identical(nxroot['mytime'][...],
-                        sc.datetime(unit='ns', value=f'1970-01-01T10:00:03.123456789'))
+                        sc.datetime(unit='ns', value='1970-01-01T10:00:03.123456789'))
 
 
 def test_loads_bare_timestamps_if_multiple_candidate_datetime_offsets_found(nxroot):

From 43bd97d4ac7c1144ef429bc9c3536173f8cde466 Mon Sep 17 00:00:00 2001
From: Simon Heybrock <simon.heybrock@ess.eu>
Date: Fri, 8 Apr 2022 07:12:36 +0200
Subject: [PATCH 20/25] Avoid using epoch where possible

---
 tests/nexus_test.py | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/tests/nexus_test.py b/tests/nexus_test.py
index 706efc7c..837c1b82 100644
--- a/tests/nexus_test.py
+++ b/tests/nexus_test.py
@@ -235,12 +235,12 @@ def test_ns_field_with_second_datetime_attribute_loaded_as_ns_datetime(nxroot):
 
 def test_second_field_with_ns_datetime_attribute_loaded_as_ns_datetime(nxroot):
     nxroot['mytime'] = sc.arange('ignored', 2, unit='s')
-    nxroot['mytime'].attrs['start_time'] = '1970-01-01T00:00:00.000000000'
+    nxroot['mytime'].attrs['start_time'] = '1984-01-01T00:00:00.000000000'
     assert sc.identical(
         nxroot['mytime'][...],
         sc.datetimes(dims=['dim_0'],
                      unit='ns',
-                     values=['1970-01-01T00:00:00', '1970-01-01T00:00:01']))
+                     values=['1984-01-01T00:00:00', '1984-01-01T00:00:01']))
 
 
 @pytest.mark.parametrize('timezone,hhmm', [('Z', '12:00'), ('+04', '08:00'),
@@ -249,16 +249,16 @@ def test_second_field_with_ns_datetime_attribute_loaded_as_ns_datetime(nxroot):
                                            ('+11:30', '00:30'), ('-09:30', '21:30')])
 def test_timezone_information_in_datetime_attribute_is_applied(nxroot, timezone, hhmm):
     nxroot['mytime'] = sc.scalar(value=3, unit='s')
-    nxroot['mytime'].attrs['start_time'] = f'1970-01-01T12:00:00{timezone}'
+    nxroot['mytime'].attrs['start_time'] = f'1984-01-01T12:00:00{timezone}'
     assert sc.identical(nxroot['mytime'][...],
-                        sc.datetime(unit='s', value=f'1970-01-01T{hhmm}:03'))
+                        sc.datetime(unit='s', value=f'1984-01-01T{hhmm}:03'))
 
 
 def test_timezone_information_in_datetime_attribute_preserves_ns_precision(nxroot):
     nxroot['mytime'] = sc.scalar(value=3, unit='s')
-    nxroot['mytime'].attrs['start_time'] = '1970-01-01T12:00:00.123456789+0200'
+    nxroot['mytime'].attrs['start_time'] = '1984-01-01T12:00:00.123456789+0200'
     assert sc.identical(nxroot['mytime'][...],
-                        sc.datetime(unit='ns', value='1970-01-01T10:00:03.123456789'))
+                        sc.datetime(unit='ns', value='1984-01-01T10:00:03.123456789'))
 
 
 def test_loads_bare_timestamps_if_multiple_candidate_datetime_offsets_found(nxroot):

From dfde14eff026f05bcf0fd5ff6ee655d4580b4fb0 Mon Sep 17 00:00:00 2001
From: Simon Heybrock <simon.heybrock@ess.eu>
Date: Fri, 8 Apr 2022 07:38:37 +0200
Subject: [PATCH 21/25] Fix inverted condition

---
 src/scippnexus/nxobject.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/scippnexus/nxobject.py b/src/scippnexus/nxobject.py
index 8fb62abe..3917be20 100644
--- a/src/scippnexus/nxobject.py
+++ b/src/scippnexus/nxobject.py
@@ -113,7 +113,7 @@ def _as_datetime(obj: Any):
             # i.e., convert to UTC.
             # Would like to use dateutil directly, but with Python's datetime we do not
             # get nanosecond precision. Therefore we combine numpy and dateutil parsing.
-            date_only = 'T' in obj
+            date_only = 'T' not in obj
             if date_only:
                 return sc.datetime(np.datetime64(obj))
             else:

From 48a9389f5370ed14eeb70b91f94632192ce061d5 Mon Sep 17 00:00:00 2001
From: Simon Heybrock <simon.heybrock@ess.eu>
Date: Fri, 8 Apr 2022 08:09:05 +0200
Subject: [PATCH 22/25] Rewrite timezone-aware parsing to avoid rounding errors

---
 src/scippnexus/nxobject.py | 31 +++++++++++++++++++------------
 1 file changed, 19 insertions(+), 12 deletions(-)

diff --git a/src/scippnexus/nxobject.py b/src/scippnexus/nxobject.py
index 3917be20..b6a0332d 100644
--- a/src/scippnexus/nxobject.py
+++ b/src/scippnexus/nxobject.py
@@ -4,6 +4,7 @@
 from __future__ import annotations
 import re
 import warnings
+import datetime
 import dateutil
 from enum import Enum, auto
 import functools
@@ -115,19 +116,25 @@ def _as_datetime(obj: Any):
             # get nanosecond precision. Therefore we combine numpy and dateutil parsing.
             date_only = 'T' not in obj
             if date_only:
-                return sc.datetime(np.datetime64(obj))
+                return sc.datetime(obj)
+            date, time = obj.split('T')
+            time_and_timezone_offset = re.split(r'Z|\+|-', time)
+            time = time_and_timezone_offset[0]
+            if len(time_and_timezone_offset) == 1:
+                # No timezone, parse directly (scipp based on numpy)
+                return sc.datetime(f'{date}T{time}')
             else:
-                date, time = obj.split('T')
-                time_and_timezone_offset = re.split(r'Z|\+|-', time)
-                time = time_and_timezone_offset[0]
-                # Strip timezone and parse with numpy
-                dt = sc.datetime(np.datetime64(f'{date}T{time}'))
-                if len(time_and_timezone_offset) > 1:
-                    # There is timezone info. Parse with dateutil.
-                    utcoffset = dateutil.parser.isoparse(obj).utcoffset()
-                    seconds = sc.scalar(value=utcoffset.total_seconds(), unit='s')
-                    dt -= seconds.to(unit=dt.unit, dtype='int64')
-                return dt
+                # There is timezone info. Parse with dateutil.
+                dt = dateutil.parser.isoparse(obj)
+                dt = dt.replace(microsecond=0)  # handled by numpy
+                dt = dt.astimezone(datetime.timezone.utc)
+                dt = dt.replace(tzinfo=None).isoformat()
+                # We operate with string operations here and thus end up parsing date
+                # and time twice. The reason is the the timezone-offset arithmetic must
+                # cannot be done, e.g., in nanoseconds without causing rounding errors.
+                if '.' in time:
+                    dt += f".{time.split('.')[1]}"
+                return sc.datetime(dt)
         except ValueError:
             pass
     return None

From 5ce78ff03d306b761ec27711027bcc65e8c10a75 Mon Sep 17 00:00:00 2001
From: Simon Heybrock <simon.heybrock@ess.eu>
Date: Fri, 8 Apr 2022 08:12:07 +0200
Subject: [PATCH 23/25] Treat warnings as errors when running pytest

---
 pyproject.toml             | 4 ++++
 src/scippnexus/nxobject.py | 2 +-
 2 files changed, 5 insertions(+), 1 deletion(-)

diff --git a/pyproject.toml b/pyproject.toml
index 38ed7627..e46a344d 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -10,3 +10,7 @@ build-backend = "setuptools.build_meta"
 [tool.pytest.ini_options]
 addopts = "-ra -v"
 testpaths = "tests"
+filterwarnings = [
+  "error",
+  "ignore::UserWarning",
+]
diff --git a/src/scippnexus/nxobject.py b/src/scippnexus/nxobject.py
index b6a0332d..120c5e81 100644
--- a/src/scippnexus/nxobject.py
+++ b/src/scippnexus/nxobject.py
@@ -5,7 +5,7 @@
 import re
 import warnings
 import datetime
-import dateutil
+import dateutil.parser
 from enum import Enum, auto
 import functools
 from typing import List, Union, NoReturn, Any, Dict, Tuple, Protocol

From 00c576fd5fe212e9b41ac5a434a2fcd161c44799 Mon Sep 17 00:00:00 2001
From: Simon Heybrock <simon.heybrock@ess.eu>
Date: Mon, 11 Apr 2022 08:10:55 +0200
Subject: [PATCH 24/25] Update src/scippnexus/nxobject.py

Co-authored-by: Jan-Lukas Wynen <j-l.wynen@hotmail.de>
---
 src/scippnexus/nxobject.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/scippnexus/nxobject.py b/src/scippnexus/nxobject.py
index 120c5e81..7210e94c 100644
--- a/src/scippnexus/nxobject.py
+++ b/src/scippnexus/nxobject.py
@@ -130,7 +130,7 @@ def _as_datetime(obj: Any):
                 dt = dt.astimezone(datetime.timezone.utc)
                 dt = dt.replace(tzinfo=None).isoformat()
                 # We operate with string operations here and thus end up parsing date
-                # and time twice. The reason is the the timezone-offset arithmetic must
+                # and time twice. The reason is that the timezone-offset arithmetic
                 # cannot be done, e.g., in nanoseconds without causing rounding errors.
                 if '.' in time:
                     dt += f".{time.split('.')[1]}"

From c1185a6ad84323bd09bea3d663ab6a3d1d47d9c4 Mon Sep 17 00:00:00 2001
From: Simon Heybrock <simon.heybrock@ess.eu>
Date: Tue, 12 Apr 2022 12:39:04 +0200
Subject: [PATCH 25/25] Missing python-dateutil for conda build

---
 conda/meta.yaml | 1 +
 1 file changed, 1 insertion(+)

diff --git a/conda/meta.yaml b/conda/meta.yaml
index b0aae100..36a7cef6 100644
--- a/conda/meta.yaml
+++ b/conda/meta.yaml
@@ -9,6 +9,7 @@ requirements:
     - setuptools
   run:
     - python>=3.8
+    - python-dateutil
     - scipp
     - h5py