llvm-premerge-checks/scripts/metrics/repo_hist.py

#!/usr/bin/env python3
# Copyright 2019 Google LLC
#
# Licensed under the the Apache License v2.0 with LLVM Exceptions (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     https://llvm.org/LICENSE.txt
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

# generate statistics on the llvm github repository

import csv
import datetime
import git
import re
import os
from typing import Dict, Optional, List, Set
import random
import string

REVISION_REGEX = re.compile(
    r'^Differential Revision: https://reviews\.llvm\.org/(.*)$',
    re.MULTILINE)
REVERT_REGEX = re.compile(r'^Revert "(.+)"')


class MyCommit:

    SALT = ''.join(random.choices(
        string.ascii_lowercase + string.ascii_uppercase + string.digits, k=16))

    def __init__(self, commit: git.Commit):
        self.commit = commit
        self.chash = commit.hexsha  # type: str
        self.author = hash(commit.author.email + MyCommit.SALT)  # type: int
        self.author_domain = commit.author.email.rsplit("@")[-1].lower()  # type: str
        self.commiter = hash(commit.committer.email.lower() + MyCommit.SALT)  # type:int
        self.summary = commit.summary  # type: str
        self.date = datetime.datetime.fromtimestamp(
            commit.committed_date)  # type: datetime.datetime
        self.phab_revision = self._get_revision(commit)  # type: Optional[str]
        self.reverts = None   # type: Optional[MyCommit]
        self.reverted_by = None   # type: Optional[MyCommit]
        self._diff_index = None  # type: Optional[git.DiffIndex]

    @staticmethod
    def _get_revision(commit: git.Commit) -> Optional[str]:
        m = REVISION_REGEX.search(commit.message)
        if m is None:
            return None
        return m.group(1)

    @property
    def day(self) -> datetime.date:
        return self.date.date()

    def reverts_summary(self) -> Optional[str]:
        m = REVERT_REGEX.search(self.summary)
        if m is None:
            return None
        return m.group(1)

    def __str__(self):
        return self.chash

    @property
    def was_reverted(self) -> bool:
        return self.reverted_by is not None

    @property
    def was_reviewed(self) -> bool:
        return self.phab_revision is not None

    @property
    def is_revert(self) -> bool:
        return self.reverts is not None

    @property
    def week(self) -> str:
        return '{}-w{:02d}'.format(self.date.year, self.date.isocalendar()[1])

    @property
    def diff_index(self) -> git.DiffIndex:
        # expensive operation, cache the results
        if self._diff_index is None:
            self._diff_index = self.commit.diff(self.commit.parents[0], create_patch=True)
        return self._diff_index

    @property
    def num_loc(self) -> int:
        nloc = 0
        for diff in self.diff_index:
            nloc += str(diff.diff, encoding='utf8').count('\n')
        return nloc

    @property
    def modified_paths(self) -> Set[str]:
        result = set(d.b_path for d in self.diff_index if d.b_path is not None)
        result.update(d.a_path for d in self.diff_index if d.a_path is not None)
        return result

    @property
    def modified_projects(self) -> Set[str]:
        return set(p.split('/')[0] for p in self.modified_paths)


class RepoStats:

    def __init__(self, git_dir: str):
        self.repo = git.Repo(git_dir)
        self.commit_by_hash = dict()  # type: Dict[str, MyCommit]
        self.commit_by_summary = dict()  # type: Dict[str, List[MyCommit]]
        self.commit_by_week = dict()  # type: Dict[str, List[MyCommit]]
        self.commit_by_author = dict()  # type: Dict[int, List[MyCommit]]
        self.commit_by_author_domain = dict()  # type: Dict[str, List[MyCommit]]

    def parse_repo(self,  maxage: datetime.datetime):
        for commit in self.repo.iter_commits('main'):
            if commit.committed_datetime < maxage:
                break
            mycommit = MyCommit(commit)
            self.commit_by_hash[mycommit.chash] = mycommit
            self.commit_by_summary.setdefault(mycommit.summary, [])\
                .append(mycommit)
            self.commit_by_week.setdefault(mycommit.week, []).append(mycommit)
            self.commit_by_author.setdefault(mycommit.author, [])\
                .append(mycommit)
            self.commit_by_author_domain.setdefault(mycommit.author_domain, []) \
                .append(mycommit)
        print('Read {} commits'.format(len(self.commit_by_hash)))

    def find_reverts(self):
        reverts = 0
        for commit in self.commit_by_hash.values():
            summary = commit.reverts_summary()
            if summary is None:
                continue
            if summary not in self.commit_by_summary:
                print('summary not found: {}'.format(summary))
                continue
            reverting_commit = self.commit_by_summary[summary][-1]
            commit.reverted_by = reverting_commit
            reverting_commit.reverts = commit
            reverts += 1
        print('Found {} reverts'.format(reverts))

    # https://stackoverflow.com/questions/2600775/how-to-get-week-number-in-python
    def dump_daily_stats(self):
        fieldnames = ["week", "num_commits", "num_reverts", "percentage_reverts",
                      "num_reviewed", "percentage_reviewed",
                      "# reviewed & revert", "# !reviewed & !revert", "# !reviewed & revert", "# reviewed & !revert"]
        csvfile = open('tmp/llvm-project-weekly.csv', 'w')
        writer = csv.DictWriter(csvfile, fieldnames=fieldnames,
                                dialect=csv.excel)
        writer.writeheader()
        for week in sorted(self.commit_by_week.keys()):
            commits = self.commit_by_week[week]
            num_commits = len(commits)
            num_reverts = len([c for c in commits if c.is_revert])
            percentage_reverts = 100.0*num_reverts / num_commits
            num_reviewed = len([c for c in commits if c.was_reviewed])
            percentage_reviewed = 100*num_reviewed / (num_commits - num_reverts)
            num_reviewed_revert = len([c for c in commits if c.was_reviewed and c.is_revert])
            num_reviewed_nrevert = len([c for c in commits if c.was_reviewed and not c.is_revert])
            num_nreviewed_nrevert = len([c for c in commits if not c.was_reviewed and not c.is_revert])
            num_nreviewed_revert = len([c for c in commits if not c.was_reviewed and c.is_revert])
            writer.writerow({
                "week": week,
                "num_commits": num_commits,
                "num_reverts": num_reverts,
                "percentage_reverts": percentage_reverts,
                "num_reviewed": num_reviewed,
                "percentage_reviewed": percentage_reviewed,
                "# reviewed & revert": num_reviewed_revert,
                "# !reviewed & !revert": num_nreviewed_nrevert,
                "# !reviewed & revert": num_nreviewed_revert,
                "# reviewed & !revert": num_reviewed_nrevert,
            })

    def dump_overall_stats(self):
        num_commits = len(self.commit_by_hash)
        num_reverts = len([c for c in self.commit_by_hash.values()
                           if c.is_revert])
        print('Number of commits: {}'.format(num_commits))
        print('Number of reverts: {}'.format(num_reverts))
        print('percentage of reverts: {:0.2f}'.format(
            100*num_reverts / num_commits))

        num_reviewed = len([c for c in self.commit_by_hash.values()
                            if c.was_reviewed])
        print('Number of reviewed commits: {}'.format(num_reviewed))
        print('percentage of reviewed commits: {:0.2f}'.format(
            100*num_reviewed / num_commits))

        num_reviewed_reverted = len([c for c in self.commit_by_hash.values()
                                     if c.was_reviewed and c.was_reverted])
        num_not_reviewed_reverted = len([c for c in self.commit_by_hash.values()
                                         if not c.was_reviewed and
                                         c.was_reverted])
        print('Number of reviewed that were reverted: {}'.format(num_reviewed_reverted))
        print('Number of NOT reviewed that were reverted: {}'.format(num_not_reviewed_reverted))
        print('percentage of reviewed that were reverted: {:0.2f}'.format(
            100*num_reviewed_reverted / num_reviewed))
        print('percentage of NOT reviewed that were reverted: {:0.2f}'.format(
            100*num_not_reviewed_reverted / (num_commits-num_reviewed)))

        num_foreign_committer = len([c for c in self.commit_by_hash.values()
                                     if c.author != c.commiter])
        print('Number of commits where author != committer: {}'.format(
            num_foreign_committer))
        print('Percentage of commits where author != committer: {:0.2f}'.format(
            100*num_foreign_committer/num_commits))

    def dump_author_stats(self):
        print('Number of authors: {}'.format(len(self.commit_by_author)))
        fieldnames = ["author", "num_commits", "num_reverts", "percentage_reverts",
                      "num_reviewed", "percentage_reviewed"]
        csvfile = open('tmp/llvm-project-authors.csv', 'w')
        writer = csv.DictWriter(csvfile, fieldnames=fieldnames,
                                dialect=csv.excel)
        writer.writeheader()
        for author, commits in self.commit_by_author.items():
            num_commits = len(commits)
            num_reverts = len([c for c in commits if c.was_reverted])
            percentage_reverts = 100 * num_reverts / num_commits
            num_reviewed = len([c for c in commits if c.was_reviewed])
            percentage_reviewed = 100*num_reviewed / num_commits
            writer.writerow({
                "author": author,
                "num_commits": num_commits,
                "num_reverts": num_reverts,
                "percentage_reverts": percentage_reverts,
                "num_reviewed": num_reviewed,
                "percentage_reviewed": percentage_reviewed,
            })

    def dump_author_domain_stats(self):
        print('Number of authors: {}'.format(len(self.commit_by_author)))
        fieldnames = ["author_domain", "num_commits", "num_committers"]
        csvfile = open('tmp/llvm-project-author_domains.csv', 'w')
        writer = csv.DictWriter(csvfile, fieldnames=fieldnames,
                                dialect=csv.excel)
        writer.writeheader()
        for author_domain, commits in self.commit_by_author_domain.items():
            num_commits = len(commits)
            committers = set(c.author for c in commits)
            writer.writerow({
                "author_domain": author_domain,
                "num_commits": num_commits,
                "num_committers": len(committers),
            })

    def dump_unreviewed_paths(self, maxage: datetime.datetime):
        # TODO: this is really slow. Maybe parallelize?
        path_count = {
            True: {},
            False: {},
        }  # type: Dict[bool, Dict[str, int]]
        for commit in self.repo.iter_commits('main'):
            if commit.committed_datetime < maxage:
                break
            mycommit = MyCommit(commit)
            for prefix in set(p.split('/')[0] for p in mycommit.modified_paths):
                path_count[mycommit.was_reviewed].setdefault(prefix, 0)
                path_count[mycommit.was_reviewed][prefix] += 1
        fieldnames = ['was_reviewed']
        all_paths = set(path_count[True].keys())
        all_paths.update(path_count[False].keys())
        fieldnames.extend(sorted(all_paths))
        csvfile = open('tmp/llvm-project-unreviewed-paths.csv', 'w')
        writer = csv.DictWriter(csvfile, fieldnames=fieldnames,
                                dialect=csv.excel)
        writer.writeheader()
        for reviewed in [True, False]:
            row = {'was_reviewed': reviewed}
            for path, count in path_count[reviewed].items():
                row[path] = count
            writer.writerow(row)
        csvfile.close()

    def dump_loc_commits(self, maxage: datetime.datetime):
        # TODO: this is really slow. Maybe parallelize?
        buckets = list(range(0, 2001, 100))
        review_dict = {
            True: {b: 0 for b in buckets},
            False: {b: 0 for b in buckets},
        }  # type: Dict[bool, Dict[int, int]]
        reverted_dict = {
            True: {b: 0 for b in buckets},
            False: {b: 0 for b in buckets},
        }  # type: Dict[bool, Dict[int, int]]
        for commit in self.repo.iter_commits('main'):
            if commit.committed_datetime < maxage:
                break
            mycommit = self.commit_by_hash[commit.hexsha]
            review_dict[mycommit.was_reviewed][self._find_bucket(mycommit.num_loc, buckets)] += 1
            reverted_dict[mycommit.was_reverted][self._find_bucket(mycommit.num_loc, buckets)] += 1
        fieldnames = ['was_reviewed']
        for i in range(0, len(buckets)-1):
            fieldnames.append('{}-{}'.format(buckets[i], buckets[i+1]-1))
        fieldnames.append('>={}'.format(buckets[-1]))
        csvfile = open('tmp/llvm-project-unreviewed-loc.csv', 'w')
        writer = csv.DictWriter(csvfile, fieldnames=fieldnames,
                                dialect=csv.excel)
        writer.writeheader()
        for reviewed in [True, False]:
            row = {'was_reviewed': reviewed}
            for i in range(0, len(buckets)):
                row[fieldnames[i+1]] = review_dict[reviewed][buckets[i]]
            writer.writerow(row)

        writer.writerow({'was_reviewed': 'reverted'})
        for reverted in [True, False]:
            row = {'was_reviewed': reverted}
            for i in range(0, len(buckets)):
                row[fieldnames[i+1]] = reverted_dict[reverted][buckets[i]]
            writer.writerow(row)
        csvfile.close()

    @staticmethod
    def _find_bucket(number: int, buckets: List[int]) -> int:
        for bucket in buckets:
            if number < bucket:
                return bucket
        return buckets[-1]

    def export_commits(self):

        print('starting export...')
        csvfile = open('tmp/llvm-project-export.csv', 'w')
        fieldnames = ['timestamp', 'hash', 'reviewed', 'was_reverted', 'is_revert', '# LOC changed',
                      'modified projects',  'author domain', 'revision']
        writer = csv.DictWriter(csvfile, fieldnames=fieldnames,
                                dialect=csv.excel)
        writer.writeheader()
        # did not work with multiprocessing.map, gave recursion error from gitpython...
        # so using normal map function
        for row in map(_create_row, self.commit_by_hash.values()):
            writer.writerow(row)
        csvfile.close()


def _create_row(mycommit: MyCommit) -> Dict:
    try:
        return {
            'timestamp': mycommit.date.isoformat(),
            'hash': mycommit.chash,
            'reviewed': mycommit.was_reviewed,
            'was_reverted': mycommit.was_reverted,
            'is_revert': mycommit.is_revert,
            '# LOC changed': mycommit.num_loc,
            'modified projects': (';'.join(mycommit.modified_projects)),
            'author domain': mycommit.author_domain,
            'revision': mycommit.phab_revision if mycommit.phab_revision is not None else "",
        }
    except Exception as e:
        print(e)
        return {}


if __name__ == '__main__':
    max_age = datetime.datetime(year=2019, month=10, day=1,
                                tzinfo=datetime.timezone.utc)
    now = datetime.datetime.now(tz=datetime.timezone.utc)
    rs = RepoStats(os.path.expanduser('~/git/llvm-project'))
    # TODO: make the path configurable, and `git clone/pull`
    rs.parse_repo(max_age)
    rs.find_reverts()
    rs.dump_daily_stats()
    rs.dump_overall_stats()
    rs.dump_author_stats()
    rs.dump_author_domain_stats()
    # disabled as it's quite slow
    # rs.dump_unreviewed_paths(now - datetime.timedelta(days=100))
    # rs.dump_loc_commits(now - datetime.timedelta(days=100))
    rs.export_commits()
    print('Done.')