diff --git a/src/fediblockhole/__init__.py b/src/fediblockhole/__init__.py index 945e29c..893787d 100755 --- a/src/fediblockhole/__init__.py +++ b/src/fediblockhole/__init__.py @@ -11,7 +11,7 @@ import os.path import sys import urllib.request as urlr -from .blocklist_parser import parse_blocklist +from .blocklists import Blocklist, parse_blocklist from .const import DomainBlock, BlockSeverity from importlib.metadata import version @@ -178,41 +178,71 @@ def fetch_from_instances(blocklists: dict, sources: dict, save_intermediate_blocklist(blocklists[itemsrc], domain, savedir, export_fields) return blocklists -def merge_blocklists(blocklists: dict, mergeplan: str='max') -> dict: +def merge_blocklists(blocklists: list[Blocklist], mergeplan: str='max', threshold: int=0) -> dict: """Merge fetched remote blocklists into a bulk update @param blocklists: A dict of lists of DomainBlocks, keyed by source. Each value is a list of DomainBlocks @param mergeplan: An optional method of merging overlapping block definitions 'max' (the default) uses the highest severity block found 'min' uses the lowest severity block found + @param threshold: An integer percentage [0-100]. + If a domain is not present in this pct or more of the blocklists, + it will not get merged into the final list. @param returns: A dict of DomainBlocks keyed by domain """ merged = {} - for key, blist in blocklists.items(): - log.debug(f"processing blocklist from: {key} ...") - for newblock in blist: - domain = newblock.domain - # If the domain has two asterisks in it, it's obfuscated - # and we can't really use it, so skip it and do the next one - if '*' in domain: + num_blocklists = len(blocklists) + + # Create a domain keyed list of blocks for each domain + domain_blocks = {} + + for bl in blocklists: + for block in bl.values(): + if '*' in block.domain: log.debug(f"Domain '{domain}' is obfuscated. Skipping it.") continue - - elif domain in merged: - log.debug(f"Overlapping block for domain {domain}. Merging...") - blockdata = apply_mergeplan(merged[domain], newblock, mergeplan) - + elif block.domain in domain_blocks: + domain_blocks[block.domain].append(block) else: - # New block - blockdata = newblock + domain_blocks[block.domain] = [block,] - # end if - log.debug(f"blockdata is: {blockdata}") - merged[domain] = blockdata - # end for + # Only merge items if there are more than `threshold` pct of them + for domain in domain_blocks: + pct = len(domain_blocks[domain]) / num_blocklists + if pct >= threshold: + # Add first block in the list to merged + merged[domain] = domain_blocks[domain][0] + # Merge the others with this record + for block in domain_blocks[domain][1:]: + merged[domain] = apply_mergeplan(merged[domain], block, mergeplan) + return merged + # for key, blist in blocklists.items(): + # log.debug(f"processing blocklist from: {key} ...") + # for newblock in blist: + # domain = newblock.domain + # # If the domain has two asterisks in it, it's obfuscated + # # and we can't really use it, so skip it and do the next one + # if '*' in domain: + # log.debug(f"Domain '{domain}' is obfuscated. Skipping it.") + # continue + + # elif domain in merged: + # log.debug(f"Overlapping block for domain {domain}. Merging...") + # blockdata = apply_mergeplan(merged[domain], newblock, mergeplan) + + # else: + # # New block + # blockdata = newblock + + # # end if + # log.debug(f"blockdata is: {blockdata}") + # merged[domain] = blockdata + # # end for + # return merged + def apply_mergeplan(oldblock: DomainBlock, newblock: DomainBlock, mergeplan: str='max') -> dict: """Use a mergeplan to decide how to merge two overlapping block definitions diff --git a/src/fediblockhole/blocklist_parser.py b/src/fediblockhole/blocklists.py similarity index 87% rename from src/fediblockhole/blocklist_parser.py rename to src/fediblockhole/blocklists.py index d5d8394..33b71a2 100644 --- a/src/fediblockhole/blocklist_parser.py +++ b/src/fediblockhole/blocklists.py @@ -1,14 +1,42 @@ """Parse various blocklist data formats """ -from typing import Iterable -from .const import DomainBlock, BlockSeverity - import csv import json +from typing import Iterable +from dataclasses import dataclass, field + +from .const import DomainBlock, BlockSeverity import logging log = logging.getLogger('fediblockhole') +@dataclass +class Blocklist: + """ A Blocklist object + + A Blocklist is a list of DomainBlocks from an origin + """ + origin: str = None + blocks: dict[str, DomainBlock] = field(default_factory=dict) + + def __len__(self): + return len(self.blocks) + + def __class_getitem__(cls, item): + return dict[str, DomainBlock] + + def __getitem__(self, item): + return self.blocks[item] + + def __iter__(self): + return self.blocks.__iter__() + + def items(self): + return self.blocks.items() + + def values(self): + return self.blocks.values() + class BlocklistParser(object): """ Base class for parsing blocklists @@ -30,7 +58,7 @@ class BlocklistParser(object): """ raise NotImplementedError - def parse_blocklist(self, blockdata) -> dict[DomainBlock]: + def parse_blocklist(self, blockdata, origin:str=None) -> Blocklist: """Parse an iterable of blocklist items @param blocklist: An Iterable of blocklist items @returns: A dict of DomainBlocks, keyed by domain @@ -38,9 +66,10 @@ class BlocklistParser(object): if self.preparse: blockdata = self.preparse(blockdata) - parsed_list = [] + parsed_list = Blocklist(origin) for blockitem in blockdata: - parsed_list.append(self.parse_item(blockitem)) + block = self.parse_item(blockitem) + parsed_list.blocks[block.domain] = block return parsed_list def parse_item(self, blockitem) -> DomainBlock: @@ -178,6 +207,7 @@ FORMAT_PARSERS = { # helper function to select the appropriate Parser def parse_blocklist( blockdata, + origin, format="csv", import_fields: list=['domain', 'severity'], max_severity: str='suspend'): @@ -185,4 +215,4 @@ def parse_blocklist( """ parser = FORMAT_PARSERS[format](import_fields, max_severity) log.debug(f"parsing {format} blocklist with import_fields: {import_fields}...") - return parser.parse_blocklist(blockdata) \ No newline at end of file + return parser.parse_blocklist(blockdata, origin) \ No newline at end of file diff --git a/src/fediblockhole/const.py b/src/fediblockhole/const.py index 93cf2ef..7ed9f48 100644 --- a/src/fediblockhole/const.py +++ b/src/fediblockhole/const.py @@ -123,7 +123,8 @@ class DomainBlock(object): reject_media: bool=False, reject_reports: bool=False, obfuscate: bool=False, - id: int=None): + id: int=None, + count: int=0): """Initialize the DomainBlock """ self.domain = domain @@ -134,6 +135,7 @@ class DomainBlock(object): self.reject_reports = reject_reports self.obfuscate = obfuscate self.id = id + self.count = 0 @property def severity(self): diff --git a/tests/test_mergeplan.py b/tests/test_mergeplan.py index 55f3914..42d2816 100644 --- a/tests/test_mergeplan.py +++ b/tests/test_mergeplan.py @@ -1,7 +1,7 @@ """Various mergeplan tests """ -from fediblockhole.blocklist_parser import parse_blocklist +from fediblockhole.blocklists import parse_blocklist from fediblockhole import merge_blocklists, merge_comments, apply_mergeplan from fediblockhole.const import SeverityLevel, DomainBlock @@ -22,20 +22,19 @@ import_fields = [ def load_test_blocklist_data(datafiles): - blocklists = {} + blocklists = [] for df in datafiles: with open(df) as fp: data = fp.read() - bl = parse_blocklist(data, 'csv', import_fields) - blocklists[df] = bl + bl = parse_blocklist(data, df, 'csv', import_fields) + blocklists.append(bl) return blocklists def test_mergeplan_max(): """Test 'max' mergeplan""" blocklists = load_test_blocklist_data([datafile01, datafile02]) - bl = merge_blocklists(blocklists, 'max') assert len(bl) == 13 diff --git a/tests/test_parser_csv.py b/tests/test_parser_csv.py index c817e16..703fe95 100644 --- a/tests/test_parser_csv.py +++ b/tests/test_parser_csv.py @@ -1,22 +1,24 @@ """Tests of the CSV parsing """ -from fediblockhole.blocklist_parser import BlocklistParserCSV, parse_blocklist -from fediblockhole.const import DomainBlock, BlockSeverity, SeverityLevel +from fediblockhole.blocklists import BlocklistParserCSV, parse_blocklist +from fediblockhole.const import SeverityLevel def test_single_line(): csvdata = "example.org" + origin = "csvfile" parser = BlocklistParserCSV() - bl = parser.parse_blocklist(csvdata) + bl = parser.parse_blocklist(csvdata, origin) assert len(bl) == 0 def test_header_only(): csvdata = "domain,severity,public_comment" + origin = "csvfile" parser = BlocklistParserCSV() - bl = parser.parse_blocklist(csvdata) + bl = parser.parse_blocklist(csvdata, origin) assert len(bl) == 0 def test_2_blocks(): @@ -24,12 +26,13 @@ def test_2_blocks(): example.org,silence example2.org,suspend """ + origin = "csvfile" parser = BlocklistParserCSV() - bl = parser.parse_blocklist(csvdata) + bl = parser.parse_blocklist(csvdata, origin) assert len(bl) == 2 - assert bl[0].domain == 'example.org' + assert 'example.org' in bl def test_4_blocks(): csvdata = """domain,severity,public_comment @@ -38,20 +41,21 @@ example2.org,suspend,"test 2" example3.org,noop,"test 3" example4.org,suspend,"test 4" """ + origin = "csvfile" parser = BlocklistParserCSV() - bl = parser.parse_blocklist(csvdata) + bl = parser.parse_blocklist(csvdata, origin) assert len(bl) == 4 - assert bl[0].domain == 'example.org' - assert bl[1].domain == 'example2.org' - assert bl[2].domain == 'example3.org' - assert bl[3].domain == 'example4.org' + assert 'example.org' in bl + assert 'example2.org' in bl + assert 'example3.org' in bl + assert 'example4.org' in bl - assert bl[0].severity.level == SeverityLevel.SILENCE - assert bl[1].severity.level == SeverityLevel.SUSPEND - assert bl[2].severity.level == SeverityLevel.NONE - assert bl[3].severity.level == SeverityLevel.SUSPEND + assert bl['example.org'].severity.level == SeverityLevel.SILENCE + assert bl['example2.org'].severity.level == SeverityLevel.SUSPEND + assert bl['example3.org'].severity.level == SeverityLevel.NONE + assert bl['example4.org'].severity.level == SeverityLevel.SUSPEND def test_ignore_comments(): csvdata = """domain,severity,public_comment,private_comment @@ -60,18 +64,18 @@ example2.org,suspend,"test 2","ignote me also" example3.org,noop,"test 3","and me" example4.org,suspend,"test 4","also me" """ + origin = "csvfile" parser = BlocklistParserCSV() - bl = parser.parse_blocklist(csvdata) + bl = parser.parse_blocklist(csvdata, origin) assert len(bl) == 4 - assert bl[0].domain == 'example.org' - assert bl[1].domain == 'example2.org' - assert bl[2].domain == 'example3.org' - assert bl[3].domain == 'example4.org' + assert 'example.org' in bl + assert 'example2.org' in bl + assert 'example3.org' in bl + assert 'example4.org' in bl - assert bl[0].public_comment == '' - assert bl[0].private_comment == '' - - assert bl[2].public_comment == '' - assert bl[2].private_comment == '' \ No newline at end of file + assert bl['example.org'].public_comment == '' + assert bl['example.org'].private_comment == '' + assert bl['example3.org'].public_comment == '' + assert bl['example4.org'].private_comment == '' \ No newline at end of file diff --git a/tests/test_parser_json.py b/tests/test_parser_json.py index 8bf17df..b2fb0a1 100644 --- a/tests/test_parser_json.py +++ b/tests/test_parser_json.py @@ -1,8 +1,8 @@ """Tests of the CSV parsing """ -from fediblockhole.blocklist_parser import BlocklistParserJSON, parse_blocklist -from fediblockhole.const import DomainBlock, BlockSeverity, SeverityLevel +from fediblockhole.blocklists import BlocklistParserJSON, parse_blocklist +from fediblockhole.const import SeverityLevel datafile = 'data-mastodon.json' @@ -14,33 +14,32 @@ def test_json_parser(): data = load_data() parser = BlocklistParserJSON() - bl = parser.parse_blocklist(data) + bl = parser.parse_blocklist(data, 'test_json') assert len(bl) == 10 - assert bl[0].domain == 'example.org' - assert bl[1].domain == 'example2.org' - assert bl[2].domain == 'example3.org' - assert bl[3].domain == 'example4.org' + assert 'example.org' in bl + assert 'example2.org' in bl + assert 'example3.org' in bl + assert 'example4.org' in bl - assert bl[0].severity.level == SeverityLevel.SUSPEND - assert bl[1].severity.level == SeverityLevel.SILENCE - assert bl[2].severity.level == SeverityLevel.SUSPEND - assert bl[3].severity.level == SeverityLevel.NONE + assert bl['example.org'].severity.level == SeverityLevel.SUSPEND + assert bl['example2.org'].severity.level == SeverityLevel.SILENCE + assert bl['example3.org'].severity.level == SeverityLevel.SUSPEND + assert bl['example4.org'].severity.level == SeverityLevel.NONE def test_ignore_comments(): data = load_data() parser = BlocklistParserJSON() - bl = parser.parse_blocklist(data) + bl = parser.parse_blocklist(data, 'test_json') assert len(bl) == 10 - assert bl[0].domain == 'example.org' - assert bl[1].domain == 'example2.org' - assert bl[2].domain == 'example3.org' - assert bl[3].domain == 'example4.org' + assert 'example.org' in bl + assert 'example2.org' in bl + assert 'example3.org' in bl + assert 'example4.org' in bl - assert bl[0].public_comment == '' - assert bl[0].private_comment == '' - - assert bl[2].public_comment == '' - assert bl[2].private_comment == '' \ No newline at end of file + assert bl['example.org'].public_comment == '' + assert bl['example.org'].private_comment == '' + assert bl['example3.org'].public_comment == '' + assert bl['example4.org'].private_comment == '' \ No newline at end of file diff --git a/tests/test_parser_rapidblockcsv.py b/tests/test_parser_rapidblockcsv.py index edb8d1e..65d579d 100644 --- a/tests/test_parser_rapidblockcsv.py +++ b/tests/test_parser_rapidblockcsv.py @@ -1,7 +1,7 @@ """Tests of the Rapidblock CSV parsing """ -from fediblockhole.blocklist_parser import RapidBlockParserCSV, parse_blocklist +from fediblockhole.blocklists import RapidBlockParserCSV, parse_blocklist from fediblockhole.const import DomainBlock, BlockSeverity, SeverityLevel csvdata = """example.org\r\nsubdomain.example.org\r\nanotherdomain.org\r\ndomain4.org\r\n""" @@ -11,13 +11,13 @@ def test_basic_rapidblock(): bl = parser.parse_blocklist(csvdata) assert len(bl) == 4 - assert bl[0].domain == 'example.org' - assert bl[1].domain == 'subdomain.example.org' - assert bl[2].domain == 'anotherdomain.org' - assert bl[3].domain == 'domain4.org' + assert 'example.org' in bl + assert 'subdomain.example.org' in bl + assert 'anotherdomain.org' in bl + assert 'domain4.org' in bl def test_severity_is_suspend(): bl = parser.parse_blocklist(csvdata) - for block in bl: + for block in bl.values(): assert block.severity.level == SeverityLevel.SUSPEND \ No newline at end of file diff --git a/tests/test_parser_rapidblockjson.py b/tests/test_parser_rapidblockjson.py index 8ccca0f..ad13811 100644 --- a/tests/test_parser_rapidblockjson.py +++ b/tests/test_parser_rapidblockjson.py @@ -1,6 +1,6 @@ """Test parsing the RapidBlock JSON format """ -from fediblockhole.blocklist_parser import parse_blocklist +from fediblockhole.blocklists import parse_blocklist from fediblockhole.const import SeverityLevel @@ -9,26 +9,26 @@ rapidblockjson = "data-rapidblock.json" def test_parse_rapidblock_json(): with open(rapidblockjson) as fp: data = fp.read() - bl = parse_blocklist(data, 'rapidblock.json') + bl = parse_blocklist(data, 'pytest', 'rapidblock.json') - assert bl[0].domain == '101010.pl' - assert bl[0].severity.level == SeverityLevel.SUSPEND - assert bl[0].public_comment == '' + assert '101010.pl' in bl + assert bl['101010.pl'].severity.level == SeverityLevel.SUSPEND + assert bl['101010.pl'].public_comment == '' - assert bl[10].domain == 'berserker.town' - assert bl[10].severity.level == SeverityLevel.SUSPEND - assert bl[10].public_comment == '' - assert bl[10].private_comment == '' + assert 'berserker.town' in bl + assert bl['berserker.town'].severity.level == SeverityLevel.SUSPEND + assert bl['berserker.town'].public_comment == '' + assert bl['berserker.town'].private_comment == '' def test_parse_with_comments(): with open(rapidblockjson) as fp: data = fp.read() - bl = parse_blocklist(data, 'rapidblock.json', ['domain', 'severity', 'public_comment', 'private_comment']) + bl = parse_blocklist(data, 'pytest', 'rapidblock.json', ['domain', 'severity', 'public_comment', 'private_comment']) - assert bl[0].domain == '101010.pl' - assert bl[0].severity.level == SeverityLevel.SUSPEND - assert bl[0].public_comment == 'cryptomining javascript, white supremacy' + assert '101010.pl' in bl + assert bl['101010.pl'].severity.level == SeverityLevel.SUSPEND + assert bl['101010.pl'].public_comment == 'cryptomining javascript, white supremacy' - assert bl[10].domain == 'berserker.town' - assert bl[10].severity.level == SeverityLevel.SUSPEND - assert bl[10].public_comment == 'freeze peach' \ No newline at end of file + assert 'berserker.town' in bl + assert bl['berserker.town'].severity.level == SeverityLevel.SUSPEND + assert bl['berserker.town'].public_comment == 'freeze peach' \ No newline at end of file