def prefix_validator(key, data, errors, context):
out = {}
extras = data.get(('__extras',), {})
# values passed as lists of dicts will have been flattened into __junk
junk = df.unflatten(data.get(('__junk',), {}))
for field_name in junk:
if not field_name.startswith(prefix):
continue
extras[field_name] = junk[field_name]
for field_name in list(extras):
if not field_name.startswith(prefix):
continue
data[(field_name,)] = extras.pop(field_name)
for v in validator_fns:
try:
df.convert(v, (field_name,), data, errors, context)
except df.StopOnError:
break
out[field_name[len(prefix):]] = data.pop((field_name,))
data[(prefix,)] = out
def subfields_validator(key, data, errors, context):
# If the field is coming from the API the value will be set directly.
value = data.get(key)
if not value:
# ... otherwise, it's a form submission so our values are stuck
# unrolled in __extras.
# If we're working on a package field, the key will look like:
# (<field name>,)
# and if we're working on a resource it'll be:
# ('resources', <resource #>, <field name>)
_junk = data.get(key[:-1] + ('__junk',), {})
# Group our unrolled fields by their index.
values = defaultdict(dict)
for k in _junk.keys():
if k[0] == key[0]:
name = k[2]
index = k[1]
# Always pop, we don't want handled values to remain in
# __extras or they'll end up on the model.
values[index][name] = _junk.pop(k)
# ... then turn it back into an ordered list.
value = [v for k, v in sorted(values.items())]
elif isinstance(value, six.string_types):
value = json.loads(value)
if not isinstance(value, list):
# We treat all subfields as repeatable when processing, even
# when they aren't defined that way in the schema.
value = [value]
for subfield in field.get('repeating_subfields', field.get('simple_subfields')):
validators = _field_create_validators(subfield, schema, False)
for entry in value:
# This right here is why we recommend globally unique field
# names, else you risk trampling values from the top-level
# schema. Some validators like require_when_published require
# other top-level fields.
entry_as_data = {(k,): v for k, v in entry.items()}
entry_as_data.update(data)
entry_errors = defaultdict(list)
for v in validators:
convert(
v,
(subfield['field_name'],),
entry_as_data,
entry_errors,
context
)
# Any subfield errors should be added as errors to the parent
# since this is the only way we have to let other plugins know
# of issues.
errors[key].extend(
itertools.chain.from_iterable(
v for v in entry_errors.itervalues()
)
)
# Pull our potentially modified fields back. What if validators
# modified other fields such as a top-level field? Is this
# "allowed" in CKAN validators? We might have to replace
# entry_as_data with a write-tracing dict to capture all
# changes.
for k in entry.keys():
entry[k] = entry_as_data[(k,)]
# It would be preferable to just always store as a list, but some plugins
# such as ckanext-restricted make assumptions on how values are stored.
if 'repeating_subfields' in field:
data[key] = json.dumps(value)
elif value:
data[key] = json.dumps(value[0])
