ssg/xccdf.py
"""A couple generic XCCDF utilities used by build_all_guides.py andbuild_all_remediations.py Author: Martin Preisler <mpreisle@redhat.com>""" from __future__ import absolute_importfrom __future__ import print_function import re from .constants import XCCDF12_NS # if a profile ID ends with a string listed here we skip itPROFILE_ID_SKIPLIST = ["test", "index", "default"]# filler XCCDF 1.2 prefix which we will strip to avoid very long filenamesPROFILE_ID_PREFIX = ("^xccdf_org.*content_profile_") def get_benchmark_id_title_map(input_tree): """ Extracts a mapping of benchmark IDs to their titles from an XML tree. Args: input_tree (xml.etree.ElementTree.ElementTree): The XML tree containing benchmark data. Returns: dict: A dictionary where the keys are benchmark IDs (str) and the values are benchmark titles (str). """ input_root = input_tree.getroot() ret = {} candidates = [] scrape_benchmarks(input_root, XCCDF12_NS, candidates) Similar blocks of code found in 2 locations. Consider refactoring. for _, elem in candidates: _id = elem.get("id") if _id is None: continue title = "<unknown>" for element in elem.findall("{%s}title" % (XCCDF12_NS)): title = element.text break ret[_id] = title return ret Function `get_profile_choices_for_input` has a Cognitive Complexity of 19 (exceeds 7 allowed). Consider refactoring.
Refactor this function to reduce its Cognitive Complexity from 19 to the 15 allowed.def get_profile_choices_for_input(input_tree, benchmark_id, tailoring_tree): """ Returns a dictionary that maps profile_ids to their respective titles. Args: input_tree (ElementTree): The XML tree containing the benchmark profiles. benchmark_id (str): The ID of the benchmark to filter profiles. tailoring_tree (ElementTree, optional): An optional XML tree containing tailored profiles. Returns: dict: A dictionary where keys are profile IDs and values are profile titles. """ # Ideally oscap would have a command line to do this, but as of now it # doesn't so we have to implement it ourselves. Importing openscap Python # bindings is nasty and overkill for this. ret = {} def scrape_profiles(root_element, namespace, dest): candidates = \ list(root_element.findall(".//{%s}Benchmark" % (namespace))) if root_element.tag == "{%s}Benchmark" % (namespace): candidates.append(root_element) for benchmark in candidates: if benchmark.get("id") != benchmark_id: continue Similar blocks of code found in 2 locations. Consider refactoring. for elem in benchmark.findall(".//{%s}Profile" % (namespace)): id_ = elem.get("id") if id_ is None: continue title = "<unknown>" for element in elem.findall("{%s}title" % (namespace)): title = element.text break dest[id_] = title input_root = input_tree.getroot() scrape_profiles( input_root, XCCDF12_NS, ret ) if tailoring_tree is not None: tailoring_root = tailoring_tree.getroot() scrape_profiles( tailoring_root, XCCDF12_NS, ret ) return ret def get_profile_short_id(long_id): """ Shortens the given profile ID if it matches the XCCDF 1.2 long ID format. Args: long_id (str): The long profile ID to be shortened. Returns: str: The shortened profile ID if the long ID matches the XCCDF 1.2 format, otherwise returns the original long ID. """ if re.search(PROFILE_ID_PREFIX, long_id): return long_id[re.search(PROFILE_ID_PREFIX, long_id).end():] return long_id def scrape_benchmarks(root, namespace, dest): """ Add all benchmark elements in root to dest list. This function searches for all elements with the tag 'Benchmark' within the given XML root element, using the specified namespace. It then adds these elements to the destination list 'dest' along with their namespace. If the root element itself is a 'Benchmark', it is also added to the list. Args: root (xml.etree.ElementTree.Element): The root XML element to search within. namespace (str): The XML namespace to use when searching for 'Benchmark' elements.Line too long (100 > 99 characters) dest (list): The list to which found 'Benchmark' elements and their namespace will be added. Returns: None """ dest.extend([ (namespace, elem) for elem in list(root.findall(".//{%s}Benchmark" % (namespace))) ]) if root.tag == "{%s}Benchmark" % (namespace): dest.append((namespace, root))