def compare_nc_atts(dat0, dat1, filename, var):
"""Report if the attributes of a NCDF file or variable have changed."""
# Check if any attributes added/removed
atts = set(dat0.ncattrs()).symmetric_difference(dat1.ncattrs())
if atts:
warn(FieldMissing(filename, ', '.join(atts)), stacklevel=3)
# Check if any attributes changed
for key in dat0.ncattrs():
if key in atts:
continue
if (dat0.__dict__[key] != dat1.__dict__[key]
and key not in defaults.atts_to_ignore):
warn(Regression(
filename, var + ', ' + key, 'warning',
'Changed attribute ({} vs {})'.format(dat0.__dict__[key],
dat1.__dict__[key])),
stacklevel=3)
def compare_orac_out(file0, file1):
"""Compare two NCDF files"""
import numpy as np
try:
from netCDF4 import Dataset
except ImportError:
warn('Skipping regression tests as netCDF4 unavailable',
OracWarning,
stacklevel=2)
return
try:
# Open files
dat0 = Dataset(file0, 'r')
dat1 = Dataset(file1, 'r')
# Check if any dimensions added/removed
dims = set(dat0.dimensions.keys()).symmetric_difference(
dat1.dimensions.keys())
if dims:
# Not bothering to identify which file contains the errant field
warn(FieldMissing(file1, ', '.join(dims)), stacklevel=2)
# Check if any dimensions changed
for key in dat0.dimensions.keys():
if key in dims:
continue
if dat0.dimensions[key].size != dat1.dimensions[key].size:
warn(InconsistentDim(file1, key, dat0.dimensions[key].size,
dat1.dimensions[key].size),
stacklevel=2)
# Check attributes
compare_nc_atts(dat0, dat1, file1, key)
# Check if any variables added/removed
variables = set(dat0.variables.keys()).symmetric_difference(
dat1.variables.keys())
if variables:
warn(FieldMissing(file1, ', '.join(vars)), stacklevel=2)
# Check if any variables changed
for key in dat0.variables.keys():
if key in variables:
continue
att0 = dat0.variables[key]
att1 = dat1.variables[key]
# Turn off masking, so completely NaN fields can be equal
att0.set_auto_mask(False)
att1.set_auto_mask(False)
compare_nc_atts(att0, att1, file1, key)
if att0.size != att1.size:
warn(InconsistentDim(file1, key, att0.size, att1.size),
stacklevel=2)
continue
# Check if there has been any change
if not np.allclose(att0, att1, equal_nan=True, rtol=0, atol=0):
test = False
# For floats, check if variation is acceptable
if att0.dtype.kind == 'f':
test = np.allclose(att0,
att1,
equal_nan=True,
rtol=defaults.rtol,
atol=defaults.atol)
else:
try:
if isinstance(att0.scale_factor, np.floating):
# Packed floats consider the scale factor
test = np.allclose(att0,
att1,
equal_nan=True,
rtol=defaults.rtol,
atol=max(
att0.scale_factor,
defaults.atol))
except AttributeError:
# If there is no scale factor, treat as an integer
pass
if test or key in defaults.vars_to_accept:
warn(Acceptable(file1, key), stacklevel=2)
else:
warn(RoundingError(file1, key), stacklevel=2)
except IOError:
pass
finally:
if 'dat0' in locals() or 'dat0' in globals():
dat0.close()
if 'dat1' in locals() or 'dat1' in globals():
dat1.close()