Upgrade to release 0.26.2:
Major changes
- check(), +/-, comparisons and to_reduced_units() now use the
active context, e.g. adding kN and kg in a context relating force
and mass (#1951, #1748, #1002, #1067)
- <number> / <offset or logarithmic unit> (e.g. 1 / ureg.degC) now
raises OffsetUnitCalculusError; Unit * bool and Quantity * str
now raise TypeError
- Add a deprecation warning to ApplicationRegistry.__getitem__()
matching direct registry item access
- Improve and test / and * between Quantity and Unit, supporting
more edge cases; improve their type annotations as well
- Preserve the registry's Quantity type in wraps annotations for a
single return unit, allowing array results to be indexed
- Improve type annotations for Unit: / and ** keep the unit type,
and __copy__/__deepcopy__ keep the registry subclass
Units and parsing
- Update definition of US survey units to reflect the 2022 revision
by NIST/NGS
- Fix fathom to equal two international yards; the old value is now
survey_fathom
- Fix the angular mil so one turn is 6400 mil
- Parse $ as a unit symbol like £, and raise DefinitionSyntaxError
for unknown operators such as & instead of ignoring them
- Raise DefinitionSyntaxError instead of an internal TypeError for
addition or subtraction in unit expressions, e.g. kilo-watt
- Reject invalid unit aliases at definition time while preserving
Unicode and degree aliases
Formatting changes
- Add TypstFormatter and ZeroFormatter to allow formatting to plain
Typst and zero package
- Format temperature units as valid LaTeX (~L) and siunitx (Lx),
e.g. \degreeCelsius
- str() and format() of a Unit now use the registry's fmt_locale,
like Quantities
- Keep the row layout of multi-dimensional arrays in
Quantity._repr_html_, as repr does
- Fix babel not localizing unit names containing underscores, e.g.
kilowatt_hour
- Fix formatting a unit with only negative exponents, e.g. 1 /
second, raising IndexError when a locale is set
- Restore the previous numeric locale when localized formatting
raises an exception
NumPy
- Support the numpy.vecdot, numpy.matvec and numpy.vecmat ufuncs
and numpy.linalg.vecdot on Quantities
- Support numpy.repeat on Quantities
- numpy.sum/numpy.nansum now raise DimensionalityError for a bare
non-zero initial on a dimensioned Quantity, like numpy.max
- Fix numpy.sum/numpy.nansum not converting a Quantity initial
value, unlike numpy.max/numpy.min
- Fix numpy.nanmax/numpy.nanmin raising ValueError on a Quantity
initial value, unlike numpy.max/numpy.min
- Fix numpy.clip not converting the units of the numpy 2.1 min/max
keywords
- Fix recursion in numpy.average with dimensionless Quantity
weights and returned=False
- Fix assigning offset-unit quantities (e.g. degC into a kelvin
array) via __setitem__
- Fix numpy.linspace(..., retstep=True) on Quantities failing to
return separate samples and step; offset temperature steps use
delta units
- Fix numpy.block raising TypeError for nested lists of Quantities,
e.g. numpy.block([[a, b], [c, d]])
- Fix numpy.percentile, numpy.nanpercentile, numpy.quantile, and
numpy.nanquantile raising RecursionError when q is a Quantity
- Fix numpy.tensordot, numpy.cross and the other multiplicative
functions rejecting positional arguments after the two arrays,
e.g. numpy.tensordot(a, b, 1)
- Fix numpy.mod, numpy.remainder, numpy.fmod and numpy.floor_divide
ignoring the units of the divisor when both operands have the
same dimensionality, e.g. np.mod(7 m, 200 cm) returned 7 m
- Fix numpy.unwrap raising TypeError on the period keyword; a
Quantity period or discont is converted to the units of the input
- Fix numpy.diff and numpy.ediff1d not converting the units of
Quantity prepend/append and to_end/to_begin values
- Fix numpy.ptp returning an absolute temperature instead of a
delta for offset units
Bugfixes
- Fix converting logarithmic units defined with a compound
reference, e.g. volt / ampere, which dropped all but one
reference unit
- Fix compound logarithmic units, e.g. Quantity(-161, "dBm/Hz"),
becoming an undefined delta_ unit that could not be converted
- Fix logarithmic unit conversion and hashing with Decimal
magnitudes and non_int_type=Decimal, with or without NumPy
- Fix UnitRegistry.wraps() converting only the first variadic
positional argument and failing on empty *args; preserve
keyword-only arguments and defaults
- Fix UnitRegistry.check() validating the wrong arguments of
functions with *args, **kwargs or keyword-only parameters, and
failing to call functions with positional-only defaults
- Fix 1 * unit skipping auto_reduce_dimensions and
autoconvert_to_preferred, so it disagreed with every other
magnitude
- Fix to_preferred raising a spurious DimensionalityError for some
preferred units
- Fix RecursionError in find_connected_nodes for long chains of
contexts
- Fix in-place **= changing a Quantity's units when the magnitude
update fails, e.g. a fractional power of an integer array
- Fix Quantity.dimensionality being stale after an in-place
operation changes the units
- Fix Quantity.to_compact() raising OffsetUnitCalculusError for
offset units like degree_Celsius with magnitude below 1
Documentation
- Document is_compatible_with in the tutorial
- Document defining units with a custom Converter
- Document that Unit equality compares definitions, not values,
e.g. kph != km/hour (#1433, #1876)
- Document how to compare angles that differ by a full turn, e.g.
-140 degree and 220 degree, using the modulo operator
Development and packaging
- Remove the babel <= 2.8 upper bound from the babel extra
- Fix CI never comparing the matplotlib test figures with the
baseline images: --mpl was lost with the Travis config in 2021
- Fix CI installing the numpy matrix pin outside the pixi env on
Windows/macOS, so those legs ran the no-numpy test suite
Signed-off-by: Leon Anavi <leon.anavi@konsulko.com>
Signed-off-by: Khem Raj <khem.raj@oss.qualcomm.com>
Leon Anavi
6540dceb7b