#!/usr/bin/env python
#
# Copyright (c) 2016 Mark Nauwelaerts
#

from mercurial import hg, ui, commands, util
from mercurial import context, subrepo
try:
    # hg >= 5.8
    from mercurial.utils import urlutil
except ImportError:
    from mercurial import util as urlutil

import re
import sys
import os
import subprocess
import argparse
import textwrap
import logging
import threading

# thanks go to git-remote-helper for some helper functions
# likewise so for python2/3 compatibility

# generic
class basecompat:
    @staticmethod
    def char(c):
      assert len(c) == 1
      return c[0]

if sys.version_info[0] == 3:
    import locale
    class compat(basecompat):
        # sigh ... wonderful python3 ... as taken from Mercurial's pycompat
        @staticmethod
        def decode_sysarg(arg):
            if os.name == r'nt':
                return arg.encode("mbcs", "ignore")
            else:
                enc = (
                    locale.getlocale()[1]
                    or locale.getdefaultlocale()[1]
                    or sys.getfilesystemencoding()
                )
                return arg.encode(enc, "surrogateescape")
        # mostly used for straight 'cast' (not real unicode content)
        @staticmethod
        def to_b(s, *args):
            if isinstance(s, str):
                args = args or ['latin-1']
                return s.encode(*args)
            return s
        stdin = sys.stdin.buffer
        stdout = sys.stdout.buffer
        stderr = sys.stderr.buffer
        getcwd = os.getcwdb
        @staticmethod
        def getenvb(val, default):
            result = os.getenv(val.decode(), default.decode() if hasattr(default, 'decode') else default)
             # if result is a string, get bytes instead
            result = result.encode() if hasattr(result, 'encode') else result
            return result
        getenv = os.getenvb if os.supports_bytes_environ else getenvb
else:
    class compat(basecompat):
        # life was simple in those days ...
        @staticmethod
        def to_b(s, *args):
            return s
        decode_sysarg = to_b
        stdin = sys.stdin
        stdout = sys.stdout
        stderr = sys.stderr
        getcwd = staticmethod(os.getcwd)
        getenv = staticmethod(os.getenv)

def puts(msg = b''):
    compat.stdout.write(msg)
    compat.stdout.write(b'\n')

def die(msg):
    compat.stderr.write(b'ERROR: %s\n' % compat.to_b(msg, 'utf-8'))
    sys.exit(1)

def warn(msg):
    compat.stderr.write(b'WARNING: %s\n' % compat.to_b(msg, 'utf-8'))
    compat.stderr.flush()

def info(msg, *args):
    logger.info(msg, *args)

def debug(msg, *args):
    logger.debug(msg, *args)

def log(msg, *args):
    logger.log(logging.LOG, msg, *args)

# new style way to import a source file
def _imp_load_source(module_name, file_path):
    import importlib.util
    loader = importlib.machinery.SourceFileLoader(module_name, file_path)
    spec = importlib.util.spec_from_loader(module_name, loader)
    module = importlib.util.module_from_spec(spec)
    sys.modules[module_name] = module
    spec.loader.exec_module(module)
    return module

def import_sibling(mod, filename):
    mydir = os.path.dirname(__file__)
    sys.dont_write_bytecode = True
    vi = sys.version_info
    ff = os.path.join(mydir, filename)
    if vi.major >= 3 and vi.minor >= 5:
        return _imp_load_source(mod, ff)
    else:
        import imp
        return imp.load_source(mod, ff)


class GitHgRepo:

    def __init__(self, topdir=None, gitdir=None):
        if gitdir != None:
            self.gitdir = gitdir
            self.topdir = os.path.join(gitdir, b'..') # will have to do
        else:
            self.topdir = None
            if not topdir:
                topdir = self.run_cmd(['rev-parse', '--show-cdup']).strip()
                if not topdir:
                    if not os.path.exists('.git'):
                        # now we lost where we are
                        raise Exception('failed to determine topdir')
                    topdir = b'.'
            self.topdir = topdir
            self.gitdir = self.run_cmd(['rev-parse', '--git-dir']).strip()
            if not self.gitdir:
                raise Exception('failed to determine gitdir')
            # the above was run in topdir
            if not os.path.isabs(self.gitdir):
                self.gitdir = os.path.join(self.topdir, self.gitdir)
        self.hg_repos = {}

    def identity(self):
        return b'[%s|%s]' % (compat.getcwd(), self.topdir or b'')

    def start_cmd(self, args, **kwargs):
        cmd = ['git'] + args
        popen_options = { 'cwd': self.topdir,
            'stdout': subprocess.PIPE, 'stderr': subprocess.PIPE }
        popen_options.update(kwargs)
        log('%s running cmd %s with options %s', self.identity(),
            cmd, popen_options)
        process = subprocess.Popen(cmd, **popen_options)
        return process

    # run a git cmd in repo dir, captures stdout and stderr by default
    # override in kwargs if otherwise desired
    def run_cmd(self, args, check=False, **kwargs):
        process = self.start_cmd(args, **kwargs)
        output = process.communicate()[0]
        if check and process.returncode != 0:
            die(b'git command failed: %s' % b' '.join([compat.to_b(a) for a in args]))
        return output

    def get_config(self, config, getall=False):
        get = { True : '--get-all', False: '--get' }
        cmd = ['git', 'config', get[getall] , config]
        return self.run_cmd(['config', get[getall] , config], stderr=None)

    def get_config_bool(self, config, default=False):
        value = self.get_config(config).rstrip()
        if value == b"true":
            return True
        elif value == b"false":
            return False
        else:
            return default

    def get_hg_repo_url(self, remote):
        url = self.get_config(b'remote.%s.url' % (remote))
        if url and url[0:4] == b'hg::':
            url = url[4:].strip()
        else:
            url = None
        return url

    def get_hg_rev(self, commit):
        hgrev = self.run_cmd(['notes', '--ref', 'refs/notes/hg', 'show', commit])
        return hgrev

    def rev_parse(self, ref):
        args = [ref] if not isinstance(ref, list) else ref
        args[0:0] = ['rev-parse', '--verify', '-q']
        return self.run_cmd(args).strip()

    def update_ref(self, ref, value):
        self.run_cmd(['update-ref', '-m', 'update by helper', ref, value])
        # let's check it happened
        return git_rev_parse(ref) == git_rev_parse(value)

    def cat_file(self,  ref):
        return self.run_cmd(['cat-file', '-p', ref])

    def get_git_commit(self, rev):
        remotehg = import_sibling('remotehg', 'git-remote-hg')
        for r in self.get_hg_repos():
            try:
                hgpath = remotehg.select_marks_dir(r, self.gitdir, False)
                m = remotehg.Marks(os.path.join(hgpath, b'marks-hg'), None)
                mark = m.from_rev(rev)
                m = GitMarks(os.path.join(hgpath, b'marks-git'))
                return m.to_rev(mark)
            except:
                pass

    # returns dict: (alias: local hg repo dir)
    def get_hg_repos(self):
        # minor caching
        if self.hg_repos:
            return self.hg_repos

        # check any local hg repo to see if rev is in there
        shared_path = os.path.join(self.gitdir, b'hg')
        hg_path = os.path.join(shared_path, b'.hg')
        if os.path.exists(shared_path):
            repos = os.listdir(shared_path)
            for r in repos:
                # skip the shared repo
                if r == b'.hg':
                    continue
                # only dirs
                if not os.path.isdir(os.path.join(shared_path, r)):
                    continue
                local_path = os.path.join(shared_path, r, b'clone')
                local_hg = os.path.join(local_path, b'.hg')
                if not os.path.exists(local_hg):
                    # could be a local repo without proxy, fetch url
                    local_path = self.get_hg_repo_url(r)
                    if not local_path:
                        warn(b'failed to find local hg for remote %s' % (r))
                        continue
                else:
                    npath = os.path.abspath(hg_path)
                    # use relative path if possible
                    if check_version(4, 2):
                        npath = os.path.join(b'..', b'..', b'..', b'.hg')
                    # make sure the shared path is always up-to-date
                    util.writefile(os.path.join(local_hg, b'sharedpath'), npath)
                self.hg_repos[r] = os.path.join(local_path)

        log('%s determined hg_repos %s', self.identity(), self.hg_repos)
        return self.hg_repos

    # returns hg repo object
    def get_hg_repo(self, r):
        repos = self.get_hg_repos()
        if r in repos:
            local_path = repos[r]
            hushui = ui.ui.load() if hasattr(ui.ui, 'load') else ui.ui()
            hushui.setconfig(b'ui', b'interactive', b'off')
            hushui.fout = open(os.devnull, 'wb')
            return hg.repository(hushui, local_path)

    def find_hg_repo(self, rev):
        repos = self.get_hg_repos()
        for r in repos:
            srepo = self.get_hg_repo(r)
            # if this one had it, we are done
            if srepo and rev in srepo and srepo[rev]:
                return srepo

    def rev_describe(self, rev):
        result = self.run_cmd(['describe', rev]) or \
            self.run_cmd(['describe', '--tags', rev]) or \
            self.run_cmd(['describe', '--contains', rev]) or \
            self.run_cmd(['describe', '--all', '--always', rev])
        return result.strip()

    class dictmemctx(context.memctx):

        def __init__(self, repo, files):
            p1, p2, data = repo[None], '0' * 40, ''
            context.memctx.__init__(self, repo, (p1, p2),
                data, list(files.keys()), self.getfilectx)
            self.files = files
            self.remotehg = import_sibling('remotehg', 'git-remote-hg')
            self.remotehg.hg_version = hg_version

        def __contains__(self, item):
            return item in self.files

        def getfilectx(self, repo, memctx, path):
            is_exec = is_link = rename = False
            return self.remotehg.make_memfilectx(repo, memctx, path, self.files[path],
                is_link, is_exec, rename)

    def read(self, relpath, rev=None):
        rev = rev if rev else b':0'
        obj = b'%s:%s' % (rev, relpath)
        # might complain bitterly to stderr if no subrepos so let's not show that
        return self.run_cmd(['show', obj])

    # see also subrepo.state
    def state(self, remote=b'origin', rev=None):
        """return a state dict, mapping subrepo paths configured in .hgsub
        to tuple: (source from .hgsub, revision from .hgsubstate, kind
        (key in types dict))
        """

        # obtain relevant files' content from specified revision
        files = { }
        for f in (b'.hgsub', b'.hgsubstate'):
            files[f] = self.read(f)
        log('state files for %s in revision %s:\n%s', remote, rev, files)

        # wrap in a context and delegate to subrepo
        # (rather than duplicating the admittedly simple parsing here)
        repo = self.get_hg_repo(remote)
        if not repo:
            die(b'no hg repo for alias %s' % remote)
        ctx = self.dictmemctx(repo, files)
        # helpers moved around 4.6
        if hasattr(subrepo, 'state'):
          state = subrepo.state(ctx, ui.ui())
        else:
          from mercurial import subrepoutil
          state = subrepoutil.state(ctx, ui.ui())
        log('obtained state %s', state)

        # now filter on type and resolve relative urls
        import posixpath
        resolved = {}
        for s in state:
            src, rev, kind = state[s]
            if not kind in (b'hg', b'git'):
                warn('skipping unsupported subrepo type %s' % kind)
                continue
            if not urlutil.url(src).isabs():
                parent = self.get_hg_repo_url(remote)
                if not parent:
                    die(b'could not determine repo url of %s' % remote)
                parent = urlutil.url(parent)
                parent.path = posixpath.join(parent.path or b'', src)
                parent.path = posixpath.normpath(parent.path)
                src = bytes(parent)
            # translate to git view url
            if kind == b'hg':
                src = b'hg::' + src
            resolved[s] = (src.strip(), rev or b'', kind)
        log('resolved state %s', resolved)
        return resolved


class SubCommand:

    def __init__(self, subcmdname, githgrepo):
        self.subcommand = subcmdname
        self.githgrepo = githgrepo
        self.argparser = self.argumentparser()
        # list of str
        self.args = []

    def argumentparser(self):
        return argparse.ArgumentParser()

    def get_remote(self, args):
        if len(args):
            assert isinstance(args[0], bytes)
            return (args[0], args[1:])
        else:
            self.usage('missing argument: <remote-alias>')

    def get_remote_url_hg(self, remote):
        url = self.githgrepo.get_hg_repo_url(remote)
        if not url:
            self.usage('%s is not a remote hg repository' % (remote))
        return url

    def execute(self, args):
        (self.options, self.args) = self.argparser.parse_known_args(args)
        self.do(self.options, [compat.decode_sysarg(a) for a in self.args])

    def usage(self, msg):
        if msg:
            self.argparser.error(msg)
        else:
            self.argparser.print_usage(sys.stderr)
            sys.exit(2)

    # args: list of bytes
    def do(self, options, args):
        pass


class HgRevCommand(SubCommand):

    def argumentparser(self):
        usage = '%%(prog)s %s [options] <commit-ish>' % (self.subcommand)
        p = argparse.ArgumentParser(usage=usage)
        p.epilog = textwrap.dedent("""\
        Determines the hg revision corresponding to <commit-ish>.
        """)
        return p

    def do(self, options, args):
        if len(args):
            hgrev = self.githgrepo.get_hg_rev(args[0])
            if hgrev:
                puts(hgrev)


class GitMarks:

    def __init__(self, path):
      self.path = path
      self.clear()
      self.load()

    def clear(self):
        self.marks = {}
        self.rev_marks = {}

    def load(self):
        if not os.path.exists(self.path):
            return

        for l in open(self.path, 'rb'):
            m, c = l.strip().split(b' ', 2)
            m = int(m[1:])
            self.marks[c] = m
            self.rev_marks[m] = c

    def store(self):
        marks = list(self.rev_marks.keys())
        marks.sort()
        with open(self.path, 'wb') as f:
            for m in marks:
                f.write(b':%d %s\n' % (m, self.rev_marks[m]))

    def from_rev(self, rev):
        return self.marks[rev]

    def to_rev(self, mark):
        return self.rev_marks[mark]


class GitRevCommand(SubCommand):

    def argumentparser(self):
        usage = '%%(prog)s %s [options] <revision>' % (self.subcommand)
        p = argparse.ArgumentParser(usage=usage)
        p.epilog = textwrap.dedent("""\
        Determines the git commit id corresponding to hg <revision>.
        """)
        return p

    def do(self, options, args):
        if len(args):
            rev = args[0]
            gitcommit = self.githgrepo.get_git_commit(rev)
            if gitcommit:
                puts(gitcommit)


class GcCommand(SubCommand):

    def argumentparser(self):
        usage = '%%(prog)s %s [options] <remote>...' % (self.subcommand)
        p = argparse.ArgumentParser(usage=usage, \
            formatter_class=argparse.RawDescriptionHelpFormatter)
        p.add_argument('--check-hg', action='store_true',
            help='also prune invalid hg revisions')
        p.add_argument('-n', '--dry-run', action='store_true',
            help='do not actually update any metadata files')
        p.epilog = textwrap.dedent("""\
        Performs cleanup on <remote>'s marks files and ensures these are consistent
        (never affecting or touching any git repository objects or history).
        The marks files are considered consistent if they "join"
        on the :mark number (along with a valid git commit id).

        This command can be useful in following scenarios:
        * following a git gc command;
          this could prune objects and lead to (then) invalid commit ids in marks
          (in which case git-fast-export or git-fast-import would complain bitterly).
          Such pruning is more likely to happen with remote hg repos with multiple heads.
        * cleaning marks-git of a fetch-only remote;
          git-fast-import (used during fetch) also dumps non-commit SHA-1 in the marks file,
          so the latter can become pretty large. It will reduce in size either by a push
          (git-fast-export only dumps commit objects) or by running this helper command.
        """)
        return p

    def print_commits(self, gm, dest):
        for c in gm.marks.keys():
            dest.write(c + b'\n')
        dest.flush()
        dest.close()

    def do(self, options, args):
        remotehg = import_sibling('remotehg', 'git-remote-hg')

        hg_repos = self.githgrepo.get_hg_repos()
        if not args:
            self.usage('no remote specified')
        for remote in args:
            if not remote in hg_repos:
                self.usage('%s is not a valid hg remote' % (remote))
            hgpath = remotehg.select_marks_dir(remote, self.githgrepo.gitdir, False)
            puts(b"Loading hg marks ...")
            hgm = remotehg.Marks(os.path.join(hgpath, b'marks-hg'), None)
            puts(b"Loading git marks ...")
            gm = GitMarks(os.path.join(hgpath, b'marks-git'))
            repo = hg.repository(ui.ui(), hg_repos[remote]) if options.check_hg else None
            # git-gc may have dropped unreachable commits
            # (in particular due to multiple hg head cases)
            # need to drop those so git-fast-export or git-fast-import does not complain
            puts(b"Performing garbage collection on git commits ...")
            process = self.githgrepo.start_cmd(['cat-file', '--batch-check'], \
                stdin=subprocess.PIPE)
            thread = threading.Thread(target=self.print_commits, args=(gm, process.stdin))
            thread.start()
            git_marks = set({})
            for l in process.stdout:
                sp = l.strip().split(b' ', 2)
                if sp[1] == 'commit':
                    git_marks.add(gm.from_rev(sp[0]))
            thread.join()
            # reduce down to marks that are common to both
            puts(b"Computing marks intersection ...")
            common_marks = set(hgm.rev_marks.keys()).intersection(git_marks)
            hg_rev_marks = {}
            git_rev_marks = {}
            for m in common_marks:
                if repo and not hgm.rev_marks[m] in repo:
                        continue
                hg_rev_marks[m] = hgm.rev_marks[m]
                git_rev_marks[m] = gm.rev_marks[m]
            # common marks will not not include any refs/notes/hg
            # let's not discard those casually, though they are not vital
            puts(b"Including notes commits ...")
            revlist = self.githgrepo.start_cmd(['rev-list', 'refs/notes/hg'])
            for l in revlist.stdout.readlines():
                c = l.strip()
                m = gm.marks.get(c, 0)
                if m:
                    git_rev_marks[m] = c
            # also save last-note mark
            if hgm.last_note:
                git_rev_marks[hgm.last_note] = gm.rev_marks[hgm.last_note]
            # some status report
            if len(hgm.rev_marks) != len(hg_rev_marks):
                puts(b"Trimmed hg marks from #%d down to #%d" % (len(hgm.rev_marks), len(hg_rev_marks)))
            if len(gm.rev_marks) != len(git_rev_marks):
                puts(b"Trimmed git marks from #%d down to #%d" % (len(gm.rev_marks), len(git_rev_marks)))
            # marks-hg tips irrelevant nowadays
            # now update and store
            if not options.dry_run:
                # hg marks
                puts(b"Writing hg marks ...")
                hgm.rev_marks = hg_rev_marks
                hgm.marks = {}
                for mark, rev in hg_rev_marks.items():
                    hgm.marks[rev] = mark
                hgm.store()
                # git marks
                puts(b"Writing git marks ...")
                gm.rev_marks = git_rev_marks
                gm.marks = {}
                for mark, rev in git_rev_marks.items():
                    gm.marks[rev] = mark
                gm.store()


class MapFileCommand(SubCommand):

    def argumentparser(self):
        usage = '%%(prog)s %s [options] <remote>' % (self.subcommand)
        p = argparse.ArgumentParser(usage=usage)
        p.add_argument('--output', required=True,
            help='mapfile to write')
        p.epilog = textwrap.dedent("""\
        Writes a so-called git-mapfile, as used internally by hg-git.
        This files consists of lines of format `<githexsha> <hghexsha>`.

        As such, the result could be used to coax hg-git in some manner.
        However, as git-remote-hg and hg-git may (likely) produce different
        commits (either git or hg), mixed use of both tools is not recommended.
        """)
        return p

    def do(self, options, args):
        remotehg = import_sibling('remotehg', 'git-remote-hg')

        if not args or len(args) != 1:
            self.usage('expect 1 remote')

        remote = args[0]
        hgpath = remotehg.select_marks_dir(remote, self.githgrepo.gitdir, False)
        puts(b"Loading hg marks ...")
        hgm = remotehg.Marks(os.path.join(hgpath, b'marks-hg'), None)
        puts(b"Loading git marks ...")
        gm = GitMarks(os.path.join(hgpath, b'marks-git'))
        puts(b"Writing mapfile ...")
        with open(options.output, 'wb') as f:
            for c, m in gm.marks.items():
                hgc = hgm.rev_marks.get(m, None)
                if hgc:
                    f.write(b'%s %s\n' % (c, hgc))


class SubRepoCommand(SubCommand):

    def writestate(repo, state):
        """rewrite .hgsubstate in (outer) repo with these subrepo states"""
        lines = [b'%s %s\n' % (state[s][1], s) for s in sorted(state)
                                                    if state[s][1] != nullstate[1]]
        repo.wwrite(b'.hgsubstate', b''.join(lines), b'')

    def argumentparser(self):
        #usage = '%%(prog)s %s [options] <remote>...' % (self.subcommand)
        # argparse.ArgumentParser(parents=[common])
        # top-level
        p = argparse.ArgumentParser(
            formatter_class=argparse.RawDescriptionHelpFormatter)
        p.add_argument('--quiet', '-q', action='store_true')
        p.add_argument('--recursive', '-r', action='store_true',
            help='recursive execution in each submodule')
        p.add_argument('--remote', metavar='ALIAS', default='origin',
            help='remote alias to use as base for relative url')
        p.epilog = textwrap.dedent("""\
        A group of (sub-)subcommands that aid in handling Mercurial subrepos
        as managed by .hgsub and .hgsubstate.  The commands are (where applicable)
        loosely modeled after the corresponding and similar git-submodule command,
        though otherwise there is no relation whatsoever with the git-submodule system.
        Each command supports recursive execution into subrepos of subrepos,
        and the considered (top-level) repos can be restricted by a list of subrepo
        paths as arguments.  For most commands, the state files are examined
        in the parent repo's index.

        Though depending on your workflow, the \'update\' command will likely
        suffice, along with possibly an occasional \'upstate\'.
        """)
        # subparsers
        sub = p.add_subparsers()
        # common
        common = argparse.ArgumentParser(add_help=False)
        common.add_argument('paths', nargs=argparse.REMAINDER)
        # update
        p_update = sub.add_parser('update', parents=[common],
            help='update subrepos to state as specified in parent repo')
        p_update.set_defaults(func=self.cmd_update)
        p_update.add_argument('--rebase', action='store_true',
            help='rebase the subrepo current branch onto recorded superproject commit')
        p_update.add_argument('--merge', action='store_true',
            help='merge recorded superproject commit into the current branch of subrepo')
        p_update.add_argument('--force', action='store_true',
            help='throw away local changes when switching to a different commit')
        p_update.epilog = textwrap.dedent("""\
        Brings all subrepos in the state as specified by the parent repo.
        If not yet present in the subpath, the subrepo is cloned from the URL
        specified by the parent repo (resolving relative path wrt a parent repo URL).
        If already present, and target state revision is ancestor of current HEAD,
        no update is done.  Otherwise, a checkout to the required revision is done
        (optionally forcibly so).  Alternatively, a rebase or merge is performed if
        so requested by the corresponding option.
        """)
        # foreach
        p_foreach = sub.add_parser('foreach',
            help='evaluate shell command in each checked out subrepo')
        p_foreach.set_defaults(func=self.cmd_foreach)
        p_foreach.add_argument('command', help='shell command')
        p_foreach.epilog = textwrap.dedent("""\
        The command is executed in the subrepo directory and has access to
        the variables $path, $toplevel, $sha1, $rev and $kind.
        $path is the subrepo directory relative to the parent project.
        $toplevel is the absolute path to the top-level of the parent project.
        $rev is the state revision info as set in .hgsubstate whereas
        $sha1 is the git commit corresponding to it (which may be identical
        if subrepo is a git repo).  $kind is the type of subrepo, e.g. git or hg.
        Unless given --quiet, the (recursively collected) submodule path is printed
        before evaluating the command.  A non-zero return from the command in
        any submodule causes the processing to terminate.
        """)
        # add state
        p_upstate = sub.add_parser('upstate', parents=[common],
            help='update .hgsubstate to current HEAD of subrepo')
        p_upstate.set_defaults(func=self.cmd_upstate)
        p_upstate.epilog = textwrap.dedent("""\
        Rather than dealing with the index the .hgsubstate file is simply and only
        edited to reflect the current HEAD of (all or selected) subrepos
        (and any adding to index and committing is left to user's discretion).
        """)
        # status
        p_status = sub.add_parser('status', parents=[common],
            help='show the status of the subrepos')
        p_status.add_argument('--cached', action='store_true',
            help='show index .hgsubstate revision if != subrepo HEAD')
        p_status.set_defaults(func=self.cmd_status)
        p_status.epilog = textwrap.dedent("""\
        This will print the SHA-1 of the currently checked out commit for each
        subrepo, along with the subrepo path and the output of git describe for
        the SHA-1.  Each SHA-1 will be prefixed with - if the submodule is not
        yet set up (i.e. cloned) and + if the currently checked out submodule commit
        does not match the SHA-1 found in the parent repo index state.
        """)
        # sync
        p_sync = sub.add_parser('sync', parents=[common],
            help='synchronize subrepo\'s remote URL configuration with parent configuration')
        p_sync.set_defaults(func=self.cmd_sync)
        p_sync.epilog = textwrap.dedent("""\
        The subrepo's .git/config URL configuration is set to the value specified in
        the parent's .hgstate (with relative path suitable resolved wrt parent project).
        """)
        return p

    def git_hg_repo_try(self, path):
        try:
            return GitHgRepo(topdir=path)
        except:
            return None

    def run_cmd(self, options, repo, *args, **kwargs):
        # no output if quiet, otherwise show output by default
        # unless the caller specified something explicitly
        if hasattr(options, 'quiet') and options.quiet:
            kwargs['stdout'] = os.devnull
        elif 'stdout' not in kwargs:
            kwargs['stdout'] = None
        # show errors unless caller decide some way
        if 'stderr' not in kwargs:
            kwargs['stderr'] = None
        repo.run_cmd(*args, **kwargs)

    class subcontext(dict):
        __getattr__= dict.__getitem__
        __setattr__= dict.__setitem__
        __delattr__= dict.__delitem__
        def __init__(self, repo):
            # recursion level
            self.level = 0
            # super repo object
            self.repo = repo
            # sub repo object (if any)
            self.subrepo = None
            # (recursively) collected path of subrepo
            self.subpath = None
            # relative path of subrepo wrt super repo
            self.relpath = None

    def do(self, options, args):
        if args:
            # all arguments are registered, so should not have leftover
            # could be that main arguments were given to subcommands
            warn(b'unparsed arguments: %s' % b' '.join(args))
        options.remote = compat.decode_sysarg(options.remote)
        log('running subcmd options %s, args %s', options, args)
        # establish initial operation ctx
        ctx = self.subcontext(self.githgrepo)
        self.do_operation(options, args, ctx)

    def do_operation(self, options, args, ctx):
        log('running %s with options %s in context %s', \
            options.func, options, ctx)
        subrepos = ctx.repo.state(options.remote)
        paths = list(subrepos.keys())
        selabspaths = None
        if ctx.level == 0 and hasattr(options, 'paths') and options.paths:
            selabspaths = [ os.path.abspath(compat.decode_sysarg(p)) for p in options.paths ]
        log('level %s selected paths %s', ctx.level, selabspaths)
        for p in paths:
            # prep context
            ctx.relpath = p
            ctx.subpath = os.path.join(ctx.repo.topdir, p)
            # check if selected
            abspath = os.path.abspath(ctx.subpath)
            if selabspaths != None and abspath not in selabspaths:
                log('skipping subrepo abspath %s' % abspath)
                continue
            ctx.subrepo = self.git_hg_repo_try(ctx.subpath)
            ctx.state = subrepos[p]
            # perform operation
            log('exec for context %s', ctx)
            options.func(options, args, ctx)
            if not ctx.subrepo:
                ctx.subrepo = self.git_hg_repo_try(ctx.subpath)
            # prep recursion (only into git-hg subrepos)
            if ctx.subrepo and options.recursive and ctx.state[2] == b'hg':
                newctx = self.subcontext(ctx.subrepo)
                newctx.level = ctx.level + 1
                self.do_operation(options, args, newctx)

    def get_git_commit(self, ctx):
        src, rev, kind = ctx.state
        if kind == b'hg':
            gitcommit = ctx.subrepo.get_git_commit(rev)
            if not gitcommit:
                die(b'could not determine git commit for %s; a fetch may update notes' % rev)
        else:
            gitcommit = rev
        return gitcommit

    def cmd_upstate(self, options, args, ctx):
        if not ctx.subrepo:
            return
        src, orig, kind = ctx.state
        gitcommit = ctx.subrepo.rev_parse(b'HEAD')
        if not gitcommit:
            die(b'could not determine current HEAD state in %s' % ctx.subrepo.topdir)
        rev = gitcommit
        if kind == b'hg':
            rev = ctx.subrepo.get_hg_rev(gitcommit)
            if not rev:
                die(b'could not determine hg changeset for commit %s' % gitcommit)
        else:
            rev = gitcommit
        # obtain state from index
        state_path = os.path.join(ctx.repo.topdir, b'.hgsubstate')
        # should have this, since we have subrepo (state) in the first place ...
        if not os.path.exists(state_path):
            die(b'no .hgsubstate found in repo %s' % ctx.repo.topdir)
        if orig != rev:
            short = ctx.subrepo.rev_parse(['--short', gitcommit])
            puts(b"Updating %s to %s [git %s]" % (ctx.subpath, rev, short))
            # replace and update index
            with open(state_path, 'rb') as f:
                state = f.read()
            state = re.sub(b'.{40} %s' % (ctx.relpath), b'%s %s' % (rev, ctx.relpath), state)
            with open(state_path, 'wb') as f:
                f.write(state)

    def cmd_foreach(self, options, args, ctx):
        if not ctx.subrepo:
            return
        if not options.quiet:
            puts(b'Entering %s' % ctx.subpath)
            sys.stdout.flush()
        newenv = os.environ.copy()
        newenv['path'] = ctx.relpath
        newenv['sha1'] = self.get_git_commit(ctx)
        newenv['toplevel'] = os.path.abspath(ctx.repo.topdir)
        newenv['rev'] = ctx.state[1]
        newenv['kind'] = ctx.state[2]
        proc = subprocess.Popen(options.command, shell=True, cwd=ctx.subpath, env=newenv)
        proc.wait()
        if proc.returncode != 0:
            die(b'stopping at %s; script returned non-zero status' % ctx.subpath)

    def cmd_update(self, options, args, ctx):
        if not ctx.subrepo:
            src, _, _ = ctx.state
            self.run_cmd(options, ctx.repo, ['clone', src, ctx.subpath], cwd=None)
            ctx.subrepo = self.git_hg_repo_try(ctx.subpath)
            if not ctx.subrepo:
                die(b'subrepo %s setup clone failed' % ctx.subpath)
            # force (detached) checkout of target commit following clone
            cmd = [ 'checkout', '-q' ]
        else:
            self.run_cmd(options, ctx.subrepo, ['fetch', 'origin'], check=True)
            cmd = []
        # check if subrepo is up-to-date,
        # i.e. if target commit is ancestor of HEAD
        # (output never to be shown)
        gitcommit = self.get_git_commit(ctx)
        newrev = ctx.subrepo.run_cmd(['rev-list', gitcommit, '^HEAD']).strip()
        if newrev and not cmd:
            if options.force:
                self.run_cmd(options, ctx.subrepo, ['reset', '--hard', 'HEAD'],
                    check=True)
            if options.rebase:
                cmd = [ 'rebase' ]
            elif options.merge:
                cmd = [ 'merge' ]
            else:
                # we know about the detached consequences ... keep it a bit quiet
                cmd = [ 'checkout', '-q' ]
        if cmd:
            cmd.append(gitcommit)
            self.run_cmd(options, ctx.subrepo, cmd, check=True)

    def cmd_status(self, options, args, ctx):
        if not ctx.subrepo:
            state = b'-'
            revname = b''
            _, gitcommit, kind = ctx.state
            if kind != b'git':
                gitcommit += b'[hg] '
        else:
            gitcommit = self.get_git_commit(ctx)
            head = ctx.subrepo.rev_parse(b'HEAD')
            if head == gitcommit:
                state = b' '
            else:
                state = b'+'
                # option determines what to print
                if not options.cached:
                    gitcommit = head
            revname = ctx.subrepo.rev_describe(gitcommit)
            if revname:
                revname = b' (%s)' % revname
        puts(b"%s%s %s%s" % (state, gitcommit, ctx.subpath, revname))

    def cmd_sync(self, options, args, ctx):
        if not ctx.subrepo:
            return
        src, _, _ = ctx.state
        self.run_cmd(options, ctx.subrepo, \
            ['config', b'remote.%s.url' % (options.remote), src])


class RepoCommand(SubCommand):

    def argumentparser(self):
        usage = '%%(prog)s %s [options] <remote>...' % (self.subcommand)
        p = argparse.ArgumentParser(usage=usage)
        p.epilog = textwrap.dedent("""\
        Determines the local hg repository of <remote>.
        This can either be a separate and independent local hg repository
        or a local proxy repo (within the .git directory).
        """)
        return p

    def do(self, options, args):
        (remote, args) = self.get_remote(args)
        repos = self.githgrepo.get_hg_repos()
        if remote in repos:
            puts(repos[remote].rstrip(b'/'))


class HgCommand(SubCommand):

    def argumentparser(self):
        usage = '%%(prog)s %s <hg-command>...' % (self.subcommand)
        p = argparse.ArgumentParser(usage=usage)
        hgdir = self.githgrepo.get_hg_repos()[self.subcommand]
        p.epilog = textwrap.dedent("""\
        Executes <hg-command> on the backing repository of %s (%s)
        (by supplying it with the standard -R option).
        """ % (self.subcommand, hgdir))
        return p

    def do(self, options, args):
        # subcommand name is already a known valid alias of hg repo
        remote = self.subcommand
        repos = self.githgrepo.get_hg_repos()
        if len(args) and remote in repos:
            if args[0].find(b'hg') < 0:
                args.insert(0, b'hg')
            args[1:1] = ['-R', repos[remote]]
            p = subprocess.Popen(args, stdout=None)
            p.wait()
        else:
            if len(args):
                self.usage('invalid repo: %s' % remote)
            else:
                self.usage('missing command')


class HelpCommand(SubCommand):

    def do(self, options, args):
        if len(args):
            cmd = args[0]
            if cmd in subcommands:
                p = subcommands[cmd].argumentparser()
                p.print_help(sys.stderr)
                return
        do_usage()


def get_subcommands():
    commands = {
        b'hg-rev': HgRevCommand,
        b'git-rev': GitRevCommand,
        b'repo': RepoCommand,
        b'gc':  GcCommand,
        b'sub': SubRepoCommand,
        b'mapfile': MapFileCommand,
        b'help' : HelpCommand
    }
    # add remote named subcommands
    repos = githgrepo.get_hg_repos()
    for r in repos:
        if not r in commands:
            commands[r] = HgCommand
    # now turn into instances
    for c in commands:
        commands[c] = commands[c](c, githgrepo)
    return commands


def do_usage():
    usage = textwrap.dedent("""
    git-hg-helper subcommands:

    hg-rev  \t show hg revision corresponding to a git revision
    git-rev \t find git revision corresponding to a hg revision
    gc      \t perform maintenance and consistency cleanup on repo tracking marks
    sub     \t manage subrepos
    repo    \t show local hg repo backing a remote
    mapfile \t dump a hg-git git-mapfile

    If the subcommand is the name of a remote hg repo, then any remaining arguments
    are considered a "hg command", e.g. hg heads, or thg, and it is then executed
    with -R set appropriately to the local hg repo backing the specified remote.
    Do note, however, that the local proxy repos are not maintained as exact mirrors
    of their respective remote, and also use shared storage.  As such, depending
    on the command, the result may not be exactly as could otherwise be expected
    (e.g. might involve more heads, etc).

    Available hg remotes:
    """)
    usage = compat.to_b(usage)
    for r in githgrepo.get_hg_repos():
        usage += b'\t%s\n' % (r)
    usage += b'\n'
    compat.stderr.write(usage)
    compat.stderr.flush()
    sys.exit(2)

def init_git(gitdir=None):
    global githgrepo

    try:
        githgrepo = GitHgRepo(gitdir=gitdir)
    except Exception as e:
        die(str(e))

def init_logger():
    global logger

    # setup logging
    logging.LOG = 5
    logging.addLevelName(logging.LOG, 'LOG')
    envlevel = os.environ.get('GIT_HG_HELPER_DEBUG', 'WARN')
    loglevel = logging.getLevelName(envlevel)
    logging.basicConfig(level=loglevel, \
        format='%(asctime)-15s %(levelname)s %(message)s')
    logger = logging.getLogger()

def init_version():
    global hg_version

    try:
        version, _, extra = util.version().partition(b'+')
        version = list(int(e) for e in version.split(b'.'))
        if extra:
            version[-1] += 1
        hg_version = tuple(version)
    except:
        hg_version = None

def check_version(*check):
    if not hg_version:
        return True
    return hg_version >= check

def main(argv):
    global subcommands

    # as an alias, cwd is top dir, change again to original directory
    reldir = compat.getenv(b'GIT_PREFIX', None)
    if reldir:
        os.chdir(reldir)

    # init repo dir
    # we will take over dir management ...
    gitdir = compat.getenv(b'GIT_DIR', None)
    os.environ.pop('GIT_DIR', None)
    init_git(gitdir)

    subcommands = get_subcommands()

    cmd = ''
    if len(argv) > 1:
        cmd = compat.decode_sysarg(argv[1])
    argv = argv[2:]
    if cmd in subcommands:
        c = subcommands[cmd]
        c.execute(argv)
    else:
        do_usage()

init_logger()
init_version()
if __name__ == '__main__':
    sys.exit(main(sys.argv))
