onionshare/lib/stem/descriptor/export.py
2014-05-21 14:09:41 -04:00

107 lines
3.8 KiB
Python

# Copyright 2012-2013, Damian Johnson and The Tor Project
# See LICENSE for licensing information
"""
Toolkit for exporting descriptors to other formats.
**Module Overview:**
::
export_csv - Exports descriptors to a CSV
export_csv_file - Writes exported CSV output to a file
"""
import cStringIO
import csv
import stem.descriptor
import stem.prereq
class _ExportDialect(csv.excel):
lineterminator = '\n'
def export_csv(descriptors, included_fields = (), excluded_fields = (), header = True):
"""
Provides a newline separated CSV for one or more descriptors. If simply
provided with descriptors then the CSV contains all of its attributes,
labeled with a header row. Either 'included_fields' or 'excluded_fields' can
be used for more granular control over its attributes and the order.
:param Descriptor,list descriptors: either a
:class:`~stem.descriptor.Descriptor` or list of descriptors to be exported
:param list included_fields: attributes to include in the csv
:param list excluded_fields: attributes to exclude from the csv
:param bool header: if **True** then the first line will be a comma separated
list of the attribute names (**only supported in python 2.7 and higher**)
:returns: **str** of the CSV for the descriptors, one per line
:raises: **ValueError** if descriptors contain more than one descriptor type
"""
output_buffer = cStringIO.StringIO()
export_csv_file(output_buffer, descriptors, included_fields, excluded_fields, header)
return output_buffer.getvalue()
def export_csv_file(output_file, descriptors, included_fields = (), excluded_fields = (), header = True):
"""
Similar to :func:`stem.descriptor.export.export_csv`, except that the CSV is
written directly to a file.
:param file output_file: file to be written to
:param Descriptor,list descriptors: either a
:class:`~stem.descriptor.Descriptor` or list of descriptors to be exported
:param list included_fields: attributes to include in the csv
:param list excluded_fields: attributes to exclude from the csv
:param bool header: if **True** then the first line will be a comma separated
list of the attribute names (**only supported in python 2.7 and higher**)
:returns: **str** of the CSV for the descriptors, one per line
:raises: **ValueError** if descriptors contain more than one descriptor type
"""
if isinstance(descriptors, stem.descriptor.Descriptor):
descriptors = (descriptors,)
if not descriptors:
return
descriptor_type = type(descriptors[0])
descriptor_type_label = descriptor_type.__name__
included_fields = list(included_fields)
# If the user didn't specify the fields to include then export everything,
# ordered alphabetically. If they did specify fields then make sure that
# they exist.
desc_attr = sorted(vars(descriptors[0]).keys())
if included_fields:
for field in included_fields:
if not field in desc_attr:
raise ValueError("%s does not have a '%s' attribute, valid fields are: %s" % (descriptor_type_label, field, ", ".join(desc_attr)))
else:
included_fields = [attr for attr in desc_attr if not attr.startswith('_')]
for field in excluded_fields:
try:
included_fields.remove(field)
except ValueError:
pass
writer = csv.DictWriter(output_file, included_fields, dialect = _ExportDialect(), extrasaction='ignore')
if header and stem.prereq.is_python_27():
writer.writeheader()
for desc in descriptors:
if not isinstance(desc, stem.descriptor.Descriptor):
raise ValueError("Unable to export a descriptor CSV since %s is not a descriptor." % type(desc).__name__)
elif descriptor_type != type(desc):
raise ValueError("To export a descriptor CSV all of the descriptors must be of the same type. First descriptor was a %s but we later got a %s." % (descriptor_type_label, type(desc)))
writer.writerow(vars(desc))