fedora-coreos-config/versionary

#!/usr/bin/python3 -u

# This file originally lived in
# https://github.com/coreos/fedora-coreos-releng-automation. See that repo for
# archeological git research.

'''
    Implements the Fedora CoreOS versioning scheme as per:
        https://github.com/coreos/fedora-coreos-tracker/issues/81
        https://github.com/coreos/fedora-coreos-tracker/issues/211
    And also the RHCOS/SCOS versioning scheme such as:
        9.8.20260125-0
'''

import argparse
import dotenv
import json
import os
import platform
import re
import subprocess
import sys
import time
import yaml

from datetime import datetime

# https://github.com/coreos/fedora-coreos-tracker/issues/211#issuecomment-543547587
FCOS_STREAM_TO_NUM = {
    'next': 1,
    'testing': 2,
    'stable': 3,
    'next-devel': 10,
    'testing-devel': 20,
    'rawhide': 91,
    'branched': 92,
    'bodhi-updates-testing': 93,
    'bodhi-updates': 94,
}


def main():
    args = parse_args()
    if args.workdir is not None:
        os.chdir(args.workdir)
    assert os.path.isdir('builds'), 'Missing builds/ dir'

    # Initialize all the components of our versions
    x, y, z, n = (None, None, None, None)

    # Pick up values from our build-args.
    config = dotenv.dotenv_values(args.build_args)

    # Grab the current datetime object representing the timestamp
    # for the timestamp component of our version.
    dt = get_timestamp()

    # The base version in FCOS is a single number (i.e. 43) while
    # in RHCOS/SCOS it's two numbers separated by . (i.e. 9.8 or 10.0)
    x, y = split_base_version(config['VERSION'])

    # The y component in FCOS is the timestamp, while in RHCOS/SCOS
    # it's the z component. We'll convert to a YYYYMMDD formatted string.
    if y is None:
        y = int(dt.strftime('%Y%m%d'))
    else:
        z = int(dt.strftime('%Y%m%d'))

    # At this point if z isn't defined then we're FCOS
    if z is None:
        z = FCOS_STREAM_TO_NUM[config['STREAM']]

    # For !FCOS and in dev mode we'll default to getting the build ID
    # n component by incrementing on top of the last build. For FCOS
    # not in dev mode we'll calculate the n by looking at git history.
    if args.dev or config['ID'] != 'fedora':
        n = get_next_iteration_from_builds(x, y, z)
    else:
        n = get_next_iteration_from_git(str(dt))

    # On !FCOS the delimeter for the `n` component is a -
    n_delimiter = '.' if config['ID'] == 'fedora' else '-'

    # Now we can compute the final version. Note we prepend the
    # `dev` string for the n component if --dev was passed.
    dev = 'dev' if args.dev else ''
    new_version = f'{x}.{y}.{z}{n_delimiter}{dev}{n}'
    eprint(f'VERSIONARY: selected new version for build: {new_version}')

    # sanity check the new version by trying to re-parse it
    assert parse_version(new_version) is not None
    print(new_version)


def parse_args():
    parser = argparse.ArgumentParser()
    parser.add_argument('--build-args', help="path to build-args.conf",
                        default='src/config/build-args.conf')
    parser.add_argument('--workdir', help="path to cosa workdir")
    parser.add_argument(
        "--dev", action="store_true", help="generate a developer version"
    )
    return parser.parse_args()


def get_timestamp():
    """
        Get the timestamp from either the lockfiles, or use the
        current time if no lockfiles exist.
    """

    # XXX: should sanity check that the lockfiles for all the basearches have
    # matching timestamps
    exts = ['json', 'yaml']
    basearch = platform.machine()
    for ext in exts:
        try:
            with open(f"src/config/manifest-lock.{basearch}.{ext}") as f:
                lockfile = yaml.safe_load(f)
                generated = lockfile.get('metadata', {}).get('generated')
                if not generated:
                    raise Exception("Missing 'metadata.generated' key "
                                    f"from {lockfile}")
                dt = datetime.strptime(generated, '%Y-%m-%dT%H:%M:%SZ')
                msg_src = "from lockfile"
                break
        except FileNotFoundError:
            continue
    else:
        msg_src = "from datetime.now()"
        dt = datetime.now()

    eprint(f"timestamp: {dt.strftime('%Y%m%d')} ({msg_src})")
    return dt


def split_base_version(base_version):
    components = base_version.split('.')
    if len(components) == 1:
        return int(components[0]), None
    else:
        return int(components[0]), int(components[1])


def get_next_iteration_from_builds(x, y, z):
    try:
        with open('builds/builds.json') as f:
            builds = json.load(f)
    except FileNotFoundError:
        builds = {'builds': []}

    if len(builds['builds']) == 0:
        eprint("n: 0 (no previous builds)")
        return 0

    last_buildid = builds['builds'][0]['id']
    last_version_tuple = parse_version(last_buildid)
    if not last_version_tuple:
        eprint(f"n: 0 (previous version {last_buildid} does not match scheme)")
        return 0

    if (x, y, z) != last_version_tuple[:3]:
        eprint(f"n: 0 (previous version {last_buildid} x.y.z does not match)")
        return 0

    n = last_version_tuple[3] + 1
    eprint(f"n: {n} (incremented from previous version {last_buildid})")
    return n


def get_next_iteration_from_git(timestamp):
    """
    Compute the next iteration number based on git commit history.

    Given the Y component of the version (YYYY-MM-DD HH:MM:SS date),
    this counts all commits from the start of that date up to HEAD.
    This guarantees that multiple builds on the same day each receive a
    unique `.n` value, even if several changes occur.

    See: https://github.com/coreos/fedora-coreos-tracker/issues/2015
    """
    try:
        # Count commits after that point
        commit_count_since_change = subprocess.check_output(
            ['git', 'rev-list', '--count', '--after', timestamp, 'HEAD'],
            cwd="src/config", text=True
        ).strip()
        eprint(
            f"n: {commit_count_since_change} "
            "(calculated using git commit history)"
        )
        return int(commit_count_since_change)
    except subprocess.CalledProcessError as err:
        msg = (
            "Git command failed: unable to determine the next "
            f"iteration value ({err})"
        )
        raise RuntimeError(msg) from err


def parse_version(version):
    # Note that (?:pattern) os a non-matching group in python regex so
    # it won't show up in the matched m.groups()
    m = re.match(r'^([0-9]+)\.([0-9]+)\.([0-9]+)(?:\.|-)(?:dev)?([0-9]+)$', version)
    if m is None:
        return None
    # sanity-check date. The time could be in the y component or z component
    # so we have to look for it in either.
    timegroup = 2
    if len(m.group(3)) == 8:
        timegroup = 3
    try:
        time.strptime(m.group(timegroup), '%Y%m%d')
    except ValueError:
        return None
    return tuple(map(int, m.groups()))


def eprint(*args):
    print(*args, file=sys.stderr)


if __name__ == "__main__":
    sys.exit(main())