123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509 |
- from __future__ import annotations
- # Allow direct execution
- import os
- import sys
- sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
- import enum
- import itertools
- import json
- import logging
- import re
- from collections import defaultdict
- from dataclasses import dataclass
- from functools import lru_cache
- from pathlib import Path
- from devscripts.utils import read_file, run_process, write_file
- BASE_URL = 'https://github.com'
- LOCATION_PATH = Path(__file__).parent
- HASH_LENGTH = 7
- logger = logging.getLogger(__name__)
- class CommitGroup(enum.Enum):
- PRIORITY = 'Important'
- CORE = 'Core'
- EXTRACTOR = 'Extractor'
- DOWNLOADER = 'Downloader'
- POSTPROCESSOR = 'Postprocessor'
- NETWORKING = 'Networking'
- MISC = 'Misc.'
- @classmethod
- @lru_cache
- def subgroup_lookup(cls):
- return {
- name: group
- for group, names in {
- cls.MISC: {
- 'build',
- 'ci',
- 'cleanup',
- 'devscripts',
- 'docs',
- 'test',
- },
- cls.NETWORKING: {
- 'rh',
- },
- }.items()
- for name in names
- }
- @classmethod
- @lru_cache
- def group_lookup(cls):
- result = {
- 'fd': cls.DOWNLOADER,
- 'ie': cls.EXTRACTOR,
- 'pp': cls.POSTPROCESSOR,
- 'upstream': cls.CORE,
- }
- result.update({item.name.lower(): item for item in iter(cls)})
- return result
- @classmethod
- def get(cls, value: str) -> tuple[CommitGroup | None, str | None]:
- group, _, subgroup = (group.strip().lower() for group in value.partition('/'))
- if result := cls.group_lookup().get(group):
- return result, subgroup or None
- if subgroup:
- return None, value
- return cls.subgroup_lookup().get(group), group or None
- @dataclass
- class Commit:
- hash: str | None
- short: str
- authors: list[str]
- def __str__(self):
- result = f'{self.short!r}'
- if self.hash:
- result += f' ({self.hash[:HASH_LENGTH]})'
- if self.authors:
- authors = ', '.join(self.authors)
- result += f' by {authors}'
- return result
- @dataclass
- class CommitInfo:
- details: str | None
- sub_details: tuple[str, ...]
- message: str
- issues: list[str]
- commit: Commit
- fixes: list[Commit]
- def key(self):
- return ((self.details or '').lower(), self.sub_details, self.message)
- def unique(items):
- return sorted({item.strip().lower(): item for item in items if item}.values())
- class Changelog:
- MISC_RE = re.compile(r'(?:^|\b)(?:lint(?:ing)?|misc|format(?:ting)?|fixes)(?:\b|$)', re.IGNORECASE)
- ALWAYS_SHOWN = (CommitGroup.PRIORITY,)
- def __init__(self, groups, repo, collapsible=False):
- self._groups = groups
- self._repo = repo
- self._collapsible = collapsible
- def __str__(self):
- return '\n'.join(self._format_groups(self._groups)).replace('\t', ' ')
- def _format_groups(self, groups):
- first = True
- for item in CommitGroup:
- if self._collapsible and item not in self.ALWAYS_SHOWN and first:
- first = False
- yield '\n<details><summary><h3>Changelog</h3></summary>\n'
- if group := groups[item]:
- yield self.format_module(item.value, group)
- if self._collapsible:
- yield '\n</details>'
- def format_module(self, name, group):
- result = f'\n#### {name} changes\n' if name else '\n'
- return result + '\n'.join(self._format_group(group))
- def _format_group(self, group):
- sorted_group = sorted(group, key=CommitInfo.key)
- detail_groups = itertools.groupby(sorted_group, lambda item: (item.details or '').lower())
- for _, items in detail_groups:
- items = list(items)
- details = items[0].details
- if details == 'cleanup':
- items = self._prepare_cleanup_misc_items(items)
- prefix = '-'
- if details:
- if len(items) == 1:
- prefix = f'- **{details}**:'
- else:
- yield f'- **{details}**'
- prefix = '\t-'
- sub_detail_groups = itertools.groupby(items, lambda item: tuple(map(str.lower, item.sub_details)))
- for sub_details, entries in sub_detail_groups:
- if not sub_details:
- for entry in entries:
- yield f'{prefix} {self.format_single_change(entry)}'
- continue
- entries = list(entries)
- sub_prefix = f'{prefix} {", ".join(entries[0].sub_details)}'
- if len(entries) == 1:
- yield f'{sub_prefix}: {self.format_single_change(entries[0])}'
- continue
- yield sub_prefix
- for entry in entries:
- yield f'\t{prefix} {self.format_single_change(entry)}'
- def _prepare_cleanup_misc_items(self, items):
- cleanup_misc_items = defaultdict(list)
- sorted_items = []
- for item in items:
- if self.MISC_RE.search(item.message):
- cleanup_misc_items[tuple(item.commit.authors)].append(item)
- else:
- sorted_items.append(item)
- for commit_infos in cleanup_misc_items.values():
- sorted_items.append(CommitInfo(
- 'cleanup', ('Miscellaneous',), ', '.join(
- self._format_message_link(None, info.commit.hash)
- for info in sorted(commit_infos, key=lambda item: item.commit.hash or '')),
- [], Commit(None, '', commit_infos[0].commit.authors), []))
- return sorted_items
- def format_single_change(self, info: CommitInfo):
- message, sep, rest = info.message.partition('\n')
- if '[' not in message:
- # If the message doesn't already contain markdown links, try to add a link to the commit
- message = self._format_message_link(message, info.commit.hash)
- if info.issues:
- message = f'{message} ({self._format_issues(info.issues)})'
- if info.commit.authors:
- message = f'{message} by {self._format_authors(info.commit.authors)}'
- if info.fixes:
- fix_message = ', '.join(f'{self._format_message_link(None, fix.hash)}' for fix in info.fixes)
- authors = sorted({author for fix in info.fixes for author in fix.authors}, key=str.casefold)
- if authors != info.commit.authors:
- fix_message = f'{fix_message} by {self._format_authors(authors)}'
- message = f'{message} (With fixes in {fix_message})'
- return message if not sep else f'{message}{sep}{rest}'
- def _format_message_link(self, message, commit_hash):
- assert message or commit_hash, 'Improperly defined commit message or override'
- message = message if message else commit_hash[:HASH_LENGTH]
- return f'[{message}]({self.repo_url}/commit/{commit_hash})' if commit_hash else message
- def _format_issues(self, issues):
- return ', '.join(f'[#{issue}]({self.repo_url}/issues/{issue})' for issue in issues)
- @staticmethod
- def _format_authors(authors):
- return ', '.join(f'[{author}]({BASE_URL}/{author})' for author in authors)
- @property
- def repo_url(self):
- return f'{BASE_URL}/{self._repo}'
- class CommitRange:
- COMMAND = 'git'
- COMMIT_SEPARATOR = '-----'
- AUTHOR_INDICATOR_RE = re.compile(r'Authored by:? ', re.IGNORECASE)
- MESSAGE_RE = re.compile(r'''
- (?:\[(?P<prefix>[^\]]+)\]\ )?
- (?:(?P<sub_details>`?[\w.-]+`?): )?
- (?P<message>.+?)
- (?:\ \((?P<issues>\#\d+(?:,\ \#\d+)*)\))?
- ''', re.VERBOSE | re.DOTALL)
- EXTRACTOR_INDICATOR_RE = re.compile(r'(?:Fix|Add)\s+Extractors?', re.IGNORECASE)
- REVERT_RE = re.compile(r'(?:\[[^\]]+\]\s+)?(?i:Revert)\s+([\da-f]{40})')
- FIXES_RE = re.compile(r'(?i:(?:bug\s*)?fix(?:es)?(?:\s+bugs?)?(?:\s+in|\s+for)?|Improve)\s+([\da-f]{40})')
- UPSTREAM_MERGE_RE = re.compile(r'Update to ytdl-commit-([\da-f]+)')
- def __init__(self, start, end, default_author=None):
- self._start, self._end = start, end
- self._commits, self._fixes = self._get_commits_and_fixes(default_author)
- self._commits_added = []
- def __iter__(self):
- return iter(itertools.chain(self._commits.values(), self._commits_added))
- def __len__(self):
- return len(self._commits) + len(self._commits_added)
- def __contains__(self, commit):
- if isinstance(commit, Commit):
- if not commit.hash:
- return False
- commit = commit.hash
- return commit in self._commits
- def _get_commits_and_fixes(self, default_author):
- result = run_process(
- self.COMMAND, 'log', f'--format=%H%n%s%n%b%n{self.COMMIT_SEPARATOR}',
- f'{self._start}..{self._end}' if self._start else self._end).stdout
- commits, reverts = {}, {}
- fixes = defaultdict(list)
- lines = iter(result.splitlines(False))
- for i, commit_hash in enumerate(lines):
- short = next(lines)
- skip = short.startswith('Release ') or short == '[version] update'
- fix_commitish = None
- if match := self.FIXES_RE.search(short):
- fix_commitish = match.group(1)
- authors = [default_author] if default_author else []
- for line in iter(lambda: next(lines), self.COMMIT_SEPARATOR):
- if match := self.AUTHOR_INDICATOR_RE.match(line):
- authors = sorted(map(str.strip, line[match.end():].split(',')), key=str.casefold)
- if not fix_commitish and (match := self.FIXES_RE.fullmatch(line)):
- fix_commitish = match.group(1)
- commit = Commit(commit_hash, short, authors)
- if skip and (self._start or not i):
- logger.debug(f'Skipped commit: {commit}')
- continue
- elif skip:
- logger.debug(f'Reached Release commit, breaking: {commit}')
- break
- if match := self.REVERT_RE.fullmatch(commit.short):
- reverts[match.group(1)] = commit
- continue
- if fix_commitish:
- fixes[fix_commitish].append(commit)
- commits[commit.hash] = commit
- for commitish, revert_commit in reverts.items():
- if reverted := commits.pop(commitish, None):
- logger.debug(f'{commitish} fully reverted {reverted}')
- else:
- commits[revert_commit.hash] = revert_commit
- for commitish, fix_commits in fixes.items():
- if commitish in commits:
- hashes = ', '.join(commit.hash[:HASH_LENGTH] for commit in fix_commits)
- logger.info(f'Found fix(es) for {commitish[:HASH_LENGTH]}: {hashes}')
- for fix_commit in fix_commits:
- del commits[fix_commit.hash]
- else:
- logger.debug(f'Commit with fixes not in changes: {commitish[:HASH_LENGTH]}')
- return commits, fixes
- def apply_overrides(self, overrides):
- for override in overrides:
- when = override.get('when')
- if when and when not in self and when != self._start:
- logger.debug(f'Ignored {when!r} override')
- continue
- override_hash = override.get('hash') or when
- if override['action'] == 'add':
- commit = Commit(override.get('hash'), override['short'], override.get('authors') or [])
- logger.info(f'ADD {commit}')
- self._commits_added.append(commit)
- elif override['action'] == 'remove':
- if override_hash in self._commits:
- logger.info(f'REMOVE {self._commits[override_hash]}')
- del self._commits[override_hash]
- elif override['action'] == 'change':
- if override_hash not in self._commits:
- continue
- commit = Commit(override_hash, override['short'], override.get('authors') or [])
- logger.info(f'CHANGE {self._commits[commit.hash]} -> {commit}')
- self._commits[commit.hash] = commit
- self._commits = dict(reversed(self._commits.items()))
- def groups(self):
- group_dict = defaultdict(list)
- for commit in self:
- upstream_re = self.UPSTREAM_MERGE_RE.search(commit.short)
- if upstream_re:
- commit.short = f'[upstream] Merged with youtube-dl {upstream_re.group(1)}'
- match = self.MESSAGE_RE.fullmatch(commit.short)
- if not match:
- logger.error(f'Error parsing short commit message: {commit.short!r}')
- continue
- prefix, sub_details_alt, message, issues = match.groups()
- issues = [issue.strip()[1:] for issue in issues.split(',')] if issues else []
- if prefix:
- groups, details, sub_details = zip(*map(self.details_from_prefix, prefix.split(',')))
- group = next(iter(filter(None, groups)), None)
- details = ', '.join(unique(details))
- sub_details = list(itertools.chain.from_iterable(sub_details))
- else:
- group = CommitGroup.CORE
- details = None
- sub_details = []
- if sub_details_alt:
- sub_details.append(sub_details_alt)
- sub_details = tuple(unique(sub_details))
- if not group:
- if self.EXTRACTOR_INDICATOR_RE.search(commit.short):
- group = CommitGroup.EXTRACTOR
- logger.error(f'Assuming [ie] group for {commit.short!r}')
- else:
- group = CommitGroup.CORE
- commit_info = CommitInfo(
- details, sub_details, message.strip(),
- issues, commit, self._fixes[commit.hash])
- logger.debug(f'Resolved {commit.short!r} to {commit_info!r}')
- group_dict[group].append(commit_info)
- return group_dict
- @staticmethod
- def details_from_prefix(prefix):
- if not prefix:
- return CommitGroup.CORE, None, ()
- prefix, *sub_details = prefix.split(':')
- group, details = CommitGroup.get(prefix)
- if group is CommitGroup.PRIORITY and details:
- details = details.partition('/')[2].strip()
- if details and '/' in details:
- logger.error(f'Prefix is overnested, using first part: {prefix}')
- details = details.partition('/')[0].strip()
- if details == 'common':
- details = None
- elif group is CommitGroup.NETWORKING and details == 'rh':
- details = 'Request Handler'
- return group, details, sub_details
- def get_new_contributors(contributors_path, commits):
- contributors = set()
- if contributors_path.exists():
- for line in read_file(contributors_path).splitlines():
- author, _, _ = line.strip().partition(' (')
- authors = author.split('/')
- contributors.update(map(str.casefold, authors))
- new_contributors = set()
- for commit in commits:
- for author in commit.authors:
- author_folded = author.casefold()
- if author_folded not in contributors:
- contributors.add(author_folded)
- new_contributors.add(author)
- return sorted(new_contributors, key=str.casefold)
- def create_changelog(args):
- logging.basicConfig(
- datefmt='%Y-%m-%d %H-%M-%S', format='{asctime} | {levelname:<8} | {message}',
- level=logging.WARNING - 10 * args.verbosity, style='{', stream=sys.stderr)
- commits = CommitRange(None, args.commitish, args.default_author)
- if not args.no_override:
- if args.override_path.exists():
- overrides = json.loads(read_file(args.override_path))
- commits.apply_overrides(overrides)
- else:
- logger.warning(f'File {args.override_path.as_posix()} does not exist')
- logger.info(f'Loaded {len(commits)} commits')
- if new_contributors := get_new_contributors(args.contributors_path, commits):
- if args.contributors:
- write_file(args.contributors_path, '\n'.join(new_contributors) + '\n', mode='a')
- logger.info(f'New contributors: {", ".join(new_contributors)}')
- return Changelog(commits.groups(), args.repo, args.collapsible)
- def create_parser():
- import argparse
- parser = argparse.ArgumentParser(
- description='Create a changelog markdown from a git commit range')
- parser.add_argument(
- 'commitish', default='HEAD', nargs='?',
- help='The commitish to create the range from (default: %(default)s)')
- parser.add_argument(
- '-v', '--verbosity', action='count', default=0,
- help='increase verbosity (can be used twice)')
- parser.add_argument(
- '-c', '--contributors', action='store_true',
- help='update CONTRIBUTORS file (default: %(default)s)')
- parser.add_argument(
- '--contributors-path', type=Path, default=LOCATION_PATH.parent / 'CONTRIBUTORS',
- help='path to the CONTRIBUTORS file')
- parser.add_argument(
- '--no-override', action='store_true',
- help='skip override json in commit generation (default: %(default)s)')
- parser.add_argument(
- '--override-path', type=Path, default=LOCATION_PATH / 'changelog_override.json',
- help='path to the changelog_override.json file')
- parser.add_argument(
- '--default-author', default='pukkandan',
- help='the author to use without a author indicator (default: %(default)s)')
- parser.add_argument(
- '--repo', default='yt-dlp/yt-dlp',
- help='the github repository to use for the operations (default: %(default)s)')
- parser.add_argument(
- '--collapsible', action='store_true',
- help='make changelog collapsible (default: %(default)s)')
- return parser
- if __name__ == '__main__':
- print(create_changelog(create_parser().parse_args()))
|