annotate contrib/phabricator.py @ 39506:b66ea3fc3a86

sparse-revlog: set max delta chain length to on thousand

The new snapshot system used in the sparse-revlog case gave us some small size
benefit so far. However its most important property is to gracefully handle
harder limit on delta chainlength.

Long delta chain has a very detrimental impact on read (and write) performance
in revlog. Being able to shorter them provide a great boost. However, shorting
delta used to result significantly lower compression ratio. The intermediate
snapshots effectively suppress most of this effect (even all in some case).

# Effect on the test repository

The repository we use for test is not "realistic" but can still show this in
action using an unreasonably low chain limit. Limiting the chain length show a
sizeable increase but stay under control: +6% for limit=15; +15% for limit=10.
Without the snapshot system the increase is significantly bigger: +45% for
limit=15; +80% for limit=10. Even slightly larger than without delta chain
limit, the resulting size is still smaller than before we started doing
snapshots.

Here is a table for comparison. *Since the repository is not branchy, the
initial sparse-revlog version does not bring much benefit compare to the
non-sparse one):

       chain length limit |     none   |   limit=15  |   limit=10  |
    without sparse-revlog | 62 818 987 | 112 664 615 | 131 222 574 |
    without snapshot      | 74 365 490 | 108 211 410 | 133 857 764 |
    with    snapshot      | 59 230 936 |  63 002 924 |  68 415 329 |


# Effect On Real Life Repositories

The series provides significant benefits on all kind of repositories.

Using `hg debugupgraderepo -o redeltaparent --run`, we recomputed delta chain
for various repositories with different settings:
  - delta chain length: unlimited or 1000 limit
  - sparse-revlog:      enabled or disabled
  - this series:        applied or not applied

We can observe multiple types of effect:

 - On very branchy repositories:
   * The delta chain limit as low impact on the repo size.
   * Intermediate snapshot greatly reduces manifest size:
     - pypy:     -80%
     - netbeans: -95%
   * The delta chain limit is effective, without a size impact:
     - netbeans average:      613 -> 282
     - private #1 average:  1 068 -> 307

 - On more linear repository:
   * Intermediate snapshot limit the impact of delta chain limit:
     - mozilla:
         without the series: +360%
         with the series:     +25%
   * The delta chain limit provides large improvement:
     - mozilla's average chain length:
        unlimited: 15 338
        limited:      469
   * Despite the chain length limit, the manifest size is reduced:
     - mercurial: -25%
     - mozilla:   -30%

It is clear that the use of chains of intermediate snapshots provide large
benefits both in storage size and delta chains quality. We should now switch our
effort toward making sure the write performance are acceptable. Then,
`sparse-revlog` will be a suitable format for all new repository.

# Raw Statistic

  * no-sparse:   general delta repository not using sparse-revlog
  * no-snapshot: sparse-revlog repository not using this series
  * snapshot:    sparse-revlog repository using this series

mercurial

    Manifest Size:
        limit       | none        |        1000
        ------------|-------------|------------
        no-sparse   |   8 021 373 |  8 199 366
        no-snapshot |   8 103 561 |  8 259 719
        snapshot    |   6 137 116 |  6 126 433

    Manifest Chain length data
        limit       ||       none        ||       1000        ||
        value       || average |   max   || average |   max   ||
        ------------||---------|---------||---------|---------||
        no-sparse   ||     307 |    1456 ||     279 |    1000 ||
        no-snapshot ||     312 |    1456 ||     283 |    1000 ||
        snapshot    ||     248 |    1208 ||     241 |    1000 ||

    Full Store Size
        limit       | none        |        1000
        ------------|-------------|------------
        no-sparse   |  51 013 198 | 51 201 574
        no-snapshot |  50 930 795 | 51 141 006
        snapshot    |  48 072 037 | 48 093 572

pypy

    Manifest Size:
        limit       | none        |        1000
        ------------|-------------|------------
        no-sparse   | 193 987 784 | 193 987 784
        no-snapshot | 163 171 745 | 163 312 229
        snapshot    |  34 605 900 |  34 600 750

    Manifest Chain length data
        limit       ||       none        ||       1000        ||
        value       || average |   max   || average |   max   ||
        ------------||---------|---------||---------|---------||
        no-sparse   ||     101 |     692 ||     101 |     692 ||
        no-snapshot ||     151 |    1307 ||     148 |    1000 ||
        snapshot    ||     128 |    1309 ||     125 |    1000 ||

    Full Store Size
        limit       | none        |        1000
        ------------|-------------|------------
        no-sparse   | 495 931 473 | 495 931 473
        no-snapshot | 465 441 017 | 465 581 501
        snapshot    | 355 467 301 | 355 472 451

Mozilla

    Manifest Size:
        limit       | none           |        1000
        ------------|----------------|---------------
        no-sparse   |    416 757 148 |  1 869 009 668
        no-snapshot |    401 592 370 |  1 843 493 795
        snapshot    |    224 359 521 |    284 615 500

    Manifest Chain length data
        limit       ||       none        ||       1000        ||
        value       || average |   max   || average |   max   ||
        ------------||---------|---------||---------|---------||
        no-sparse   ||  15 333 |  58 980 ||     468 |   1 000 ||
        no-snapshot ||  15 336 |  58 980 ||     469 |   1 000 ||
        snapshot    ||  15 338 |  58 983 ||     469 |   1 000 ||

    Full Store Size
        limit       | none           |        1000
        ------------|----------------|---------------
        no-sparse   |  2 712 477 887 | 4 164 995 451
        no-snapshot |  2 698 887 835 | 4 141 054 304
        snapshot    |  2 518 130 385 | 2 578 587 596

Netbeans

    Manifest Size:
        limit       | none           |        1000
        ------------|----------------|---------------
        no-sparse   |  4 766 794 101 |  4 870 642 687
        no-snapshot |  4 334 806 082 |  4 428 681 309
        snapshot    |    232 659 666 |    240 330 665

    Manifest Chain length data
        limit       ||       none        ||       1000        ||
        value       || average |   max   || average |   max   ||
        ------------||---------|---------||---------|---------||
        no-sparse   ||     597 |   6802  ||     254 |   1 000 ||
        no-snapshot ||     648 |   6 802 ||     305 |   1 000 ||
        snapshot    ||     613 |   6 804 ||     282 |   1 000 ||

    Full Store Size
        limit       | none           |        1000
        ------------|----------------|---------------
        no-sparse   |  5 807 347 998 |  5 911 196 584
        no-snapshot |  5 375 398 602 |  5 469 273 829
        snapshot    |  1 282 519 928 |  1 290 190 927

Private repo #1

    Manifest Size:
        limit       | none            |        1000
        ------------|-----------------|---------------
        no-sparse   | 41 389 010 840  | 41 398 162 091
        no-snapshot |  9 737 319 435  | 10 223 773 150
        snapshot    |    744 215 807  |    747 961 822

    Manifest Chain length data
        limit       ||       none        ||       1000        ||
        value       || average |   max   || average |   max   ||
        ------------||---------|---------||---------|---------||
        no-sparse   ||     245 |   8 885 ||     81  |   1 000 ||
        no-snapshot ||   1 225 |   8 885 ||    336  |   1 000 ||
        snapshot    ||   1 068 |   7 909 ||    307  |   1 000 ||

    Full Store Size
        limit       | none           |        1000
        ------------|----------------|---------------
        no-sparse   | 49 646 065 126 | 49 655 216 377
        no-snapshot | 17 924 862 856 | 18 411 316 571
        snapshot    |  9 009 024 710 |  9 012 770 725

Private repo #2

  We currently have less data available for that repository.
    * Before is a sparse-revlog repository without this series
    * After is a sparse-revlog repository with this series + 1000 chain limit


  Manifest Size:
    Before: 1 531 485 040 bytes
    After:  1 091 422 451 bytes
  Manifest Chain:
    Before: 2 218 avg; 6 575 Max
    After:    442 avg; 1 000 Max

  Full Store Size
    Before: 15 203 955 615
    after:   8 207 180 693

# phabricator.py - simple Phabricator integration
#
# Copyright 2017 Facebook, Inc.
#
# This software may be used and distributed according to the terms of the
# GNU General Public License version 2 or any later version.
"""simple Phabricator integration

This extension provides a ``phabsend`` command which sends a stack of
changesets to Phabricator, and a ``phabread`` command which prints a stack of
revisions in a format suitable for :hg:`import`, and a ``phabupdate`` command
to update statuses in batch.

By default, Phabricator requires ``Test Plan`` which might prevent some
changeset from being sent. The requirement could be disabled by changing
``differential.require-test-plan-field`` config server side.

Config::

    [phabricator]
    # Phabricator URL
    url = https://phab.example.com/

    # Repo callsign. If a repo has a URL https://$HOST/diffusion/FOO, then its
    # callsign is "FOO".
    callsign = FOO

    # curl command to use. If not set (default), use builtin HTTP library to
    # communicate. If set, use the specified curl command. This could be useful
    # if you need to specify advanced options that is not easily supported by
    # the internal library.
    curlcmd = curl --connect-timeout 2 --retry 3 --silent

    [auth]
    example.schemes = https
    example.prefix = phab.example.com

    # API token. Get it from https://$HOST/conduit/login/
    example.phabtoken = cli-xxxxxxxxxxxxxxxxxxxxxxxxxxxx
"""

from __future__ import absolute_import

import itertools
import json
import operator
import re

from mercurial.node import bin, nullid
from mercurial.i18n import _
from mercurial import (
    cmdutil,
    context,
    encoding,
    error,
    httpconnection as httpconnectionmod,
    mdiff,
    obsutil,
    parser,
    patch,
    registrar,
    scmutil,
    smartset,
    tags,
    url as urlmod,
    util,
)
from mercurial.utils import (
    procutil,
    stringutil,
)

cmdtable = {}
command = registrar.command(cmdtable)

configtable = {}
configitem = registrar.configitem(configtable)

# developer config: phabricator.batchsize
configitem(b'phabricator', b'batchsize',
    default=12,
)
configitem(b'phabricator', b'callsign',
    default=None,
)
configitem(b'phabricator', b'curlcmd',
    default=None,
)
# developer config: phabricator.repophid
configitem(b'phabricator', b'repophid',
    default=None,
)
configitem(b'phabricator', b'url',
    default=None,
)
configitem(b'phabsend', b'confirm',
    default=False,
)

colortable = {
    b'phabricator.action.created': b'green',
    b'phabricator.action.skipped': b'magenta',
    b'phabricator.action.updated': b'magenta',
    b'phabricator.desc': b'',
    b'phabricator.drev': b'bold',
    b'phabricator.node': b'',
}

def urlencodenested(params):
    """like urlencode, but works with nested parameters.

    For example, if params is {'a': ['b', 'c'], 'd': {'e': 'f'}}, it will be
    flattened to {'a[0]': 'b', 'a[1]': 'c', 'd[e]': 'f'} and then passed to
    urlencode. Note: the encoding is consistent with PHP's http_build_query.
    """
    flatparams = util.sortdict()
    def process(prefix, obj):
        items = {list: enumerate, dict: lambda x: x.items()}.get(type(obj))
        if items is None:
            flatparams[prefix] = obj
        else:
            for k, v in items(obj):
                if prefix:
                    process(b'%s[%s]' % (prefix, k), v)
                else:
                    process(k, v)
    process(b'', params)
    return util.urlreq.urlencode(flatparams)

printed_token_warning = False

def readlegacytoken(repo, url):
    """Transitional support for old phabricator tokens.

    Remove before the 4.7 release.
    """
    groups = {}
    for key, val in repo.ui.configitems(b'phabricator.auth'):
        if b'.' not in key:
            repo.ui.warn(_(b"ignoring invalid [phabricator.auth] key '%s'\n")
                         % key)
            continue
        group, setting = key.rsplit(b'.', 1)
        groups.setdefault(group, {})[setting] = val

    token = None
    for group, auth in groups.iteritems():
        if url != auth.get(b'url'):
            continue
        token = auth.get(b'token')
        if token:
            break

    global printed_token_warning

    if token and not printed_token_warning:
        printed_token_warning = True
        repo.ui.warn(_(b'phabricator.auth.token is deprecated - please '
                       b'migrate to auth.phabtoken.\n'))
    return token

def readurltoken(repo):
    """return conduit url, token and make sure they exist

    Currently read from [auth] config section. In the future, it might
    make sense to read from .arcconfig and .arcrc as well.
    """
    url = repo.ui.config(b'phabricator', b'url')
    if not url:
        raise error.Abort(_(b'config %s.%s is required')
                          % (b'phabricator', b'url'))

    res = httpconnectionmod.readauthforuri(repo.ui, url, util.url(url).user)
    token = None

    if res:
        group, auth = res

        repo.ui.debug(b"using auth.%s.* for authentication\n" % group)

        token = auth.get(b'phabtoken')

    if not token:
        token = readlegacytoken(repo, url)
        if not token:
            raise error.Abort(_(b'Can\'t find conduit token associated to %s')
                              % (url,))

    return url, token

def callconduit(repo, name, params):
    """call Conduit API, params is a dict. return json.loads result, or None"""
    host, token = readurltoken(repo)
    url, authinfo = util.url(b'/'.join([host, b'api', name])).authinfo()
    repo.ui.debug(b'Conduit Call: %s %s\n' % (url, params))
    params = params.copy()
    params[b'api.token'] = token
    data = urlencodenested(params)
    curlcmd = repo.ui.config(b'phabricator', b'curlcmd')
    if curlcmd:
        sin, sout = procutil.popen2(b'%s -d @- %s'
                                    % (curlcmd, procutil.shellquote(url)))
        sin.write(data)
        sin.close()
        body = sout.read()
    else:
        urlopener = urlmod.opener(repo.ui, authinfo)
        request = util.urlreq.request(url, data=data)
        body = urlopener.open(request).read()
    repo.ui.debug(b'Conduit Response: %s\n' % body)
    parsed = json.loads(body)
    if parsed.get(r'error_code'):
        msg = (_(b'Conduit Error (%s): %s')
               % (parsed[r'error_code'], parsed[r'error_info']))
        raise error.Abort(msg)
    return parsed[r'result']

@command(b'debugcallconduit', [], _(b'METHOD'))
def debugcallconduit(ui, repo, name):
    """call Conduit API

    Call parameters are read from stdin as a JSON blob. Result will be written
    to stdout as a JSON blob.
    """
    params = json.loads(ui.fin.read())
    result = callconduit(repo, name, params)
    s = json.dumps(result, sort_keys=True, indent=2, separators=(b',', b': '))
    ui.write(b'%s\n' % s)

def getrepophid(repo):
    """given callsign, return repository PHID or None"""
    # developer config: phabricator.repophid
    repophid = repo.ui.config(b'phabricator', b'repophid')
    if repophid:
        return repophid
    callsign = repo.ui.config(b'phabricator', b'callsign')
    if not callsign:
        return None
    query = callconduit(repo, b'diffusion.repository.search',
                        {b'constraints': {b'callsigns': [callsign]}})
    if len(query[r'data']) == 0:
        return None
    repophid = encoding.strtolocal(query[r'data'][0][r'phid'])
    repo.ui.setconfig(b'phabricator', b'repophid', repophid)
    return repophid

_differentialrevisiontagre = re.compile(b'\AD([1-9][0-9]*)\Z')
_differentialrevisiondescre = re.compile(
    b'^Differential Revision:\s*(?P<url>(?:.*)D(?P<id>[1-9][0-9]*))$', re.M)

def getoldnodedrevmap(repo, nodelist):
    """find previous nodes that has been sent to Phabricator

    return {node: (oldnode, Differential diff, Differential Revision ID)}
    for node in nodelist with known previous sent versions, or associated
    Differential Revision IDs. ``oldnode`` and ``Differential diff`` could
    be ``None``.

    Examines commit messages like "Differential Revision:" to get the
    association information.

    If such commit message line is not found, examines all precursors and their
    tags. Tags with format like "D1234" are considered a match and the node
    with that tag, and the number after "D" (ex. 1234) will be returned.

    The ``old node``, if not None, is guaranteed to be the last diff of
    corresponding Differential Revision, and exist in the repo.
    """
    url, token = readurltoken(repo)
    unfi = repo.unfiltered()
    nodemap = unfi.changelog.nodemap

    result = {} # {node: (oldnode?, lastdiff?, drev)}
    toconfirm = {} # {node: (force, {precnode}, drev)}
    for node in nodelist:
        ctx = unfi[node]
        # For tags like "D123", put them into "toconfirm" to verify later
        precnodes = list(obsutil.allpredecessors(unfi.obsstore, [node]))
        for n in precnodes:
            if n in nodemap:
                for tag in unfi.nodetags(n):
                    m = _differentialrevisiontagre.match(tag)
                    if m:
                        toconfirm[node] = (0, set(precnodes), int(m.group(1)))
                        continue

        # Check commit message
        m = _differentialrevisiondescre.search(ctx.description())
        if m:
            toconfirm[node] = (1, set(precnodes), int(m.group(b'id')))

    # Double check if tags are genuine by collecting all old nodes from
    # Phabricator, and expect precursors overlap with it.
    if toconfirm:
        drevs = [drev for force, precs, drev in toconfirm.values()]
        alldiffs = callconduit(unfi, b'differential.querydiffs',
                               {b'revisionIDs': drevs})
        getnode = lambda d: bin(encoding.unitolocal(
            getdiffmeta(d).get(r'node', b''))) or None
        for newnode, (force, precset, drev) in toconfirm.items():
            diffs = [d for d in alldiffs.values()
                     if int(d[r'revisionID']) == drev]

            # "precursors" as known by Phabricator
            phprecset = set(getnode(d) for d in diffs)

            # Ignore if precursors (Phabricator and local repo) do not overlap,
            # and force is not set (when commit message says nothing)
            if not force and not bool(phprecset & precset):
                tagname = b'D%d' % drev
                tags.tag(repo, tagname, nullid, message=None, user=None,
                         date=None, local=True)
                unfi.ui.warn(_(b'D%s: local tag removed - does not match '
                               b'Differential history\n') % drev)
                continue

            # Find the last node using Phabricator metadata, and make sure it
            # exists in the repo
            oldnode = lastdiff = None
            if diffs:
                lastdiff = max(diffs, key=lambda d: int(d[r'id']))
                oldnode = getnode(lastdiff)
                if oldnode and oldnode not in nodemap:
                    oldnode = None

            result[newnode] = (oldnode, lastdiff, drev)

    return result

def getdiff(ctx, diffopts):
    """plain-text diff without header (user, commit message, etc)"""
    output = util.stringio()
    for chunk, _label in patch.diffui(ctx.repo(), ctx.p1().node(), ctx.node(),
                                      None, opts=diffopts):
        output.write(chunk)
    return output.getvalue()

def creatediff(ctx):
    """create a Differential Diff"""
    repo = ctx.repo()
    repophid = getrepophid(repo)
    # Create a "Differential Diff" via "differential.createrawdiff" API
    params = {b'diff': getdiff(ctx, mdiff.diffopts(git=True, context=32767))}
    if repophid:
        params[b'repositoryPHID'] = repophid
    diff = callconduit(repo, b'differential.createrawdiff', params)
    if not diff:
        raise error.Abort(_(b'cannot create diff for %s') % ctx)
    return diff

def writediffproperties(ctx, diff):
    """write metadata to diff so patches could be applied losslessly"""
    params = {
        b'diff_id': diff[r'id'],
        b'name': b'hg:meta',
        b'data': json.dumps({
            b'user': ctx.user(),
            b'date': b'%d %d' % ctx.date(),
            b'node': ctx.hex(),
            b'parent': ctx.p1().hex(),
        }),
    }
    callconduit(ctx.repo(), b'differential.setdiffproperty', params)

    params = {
        b'diff_id': diff[r'id'],
        b'name': b'local:commits',
        b'data': json.dumps({
            ctx.hex(): {
                b'author': stringutil.person(ctx.user()),
                b'authorEmail': stringutil.email(ctx.user()),
                b'time': ctx.date()[0],
            },
        }),
    }
    callconduit(ctx.repo(), b'differential.setdiffproperty', params)

def createdifferentialrevision(ctx, revid=None, parentrevid=None, oldnode=None,
                               olddiff=None, actions=None):
    """create or update a Differential Revision

    If revid is None, create a new Differential Revision, otherwise update
    revid. If parentrevid is not None, set it as a dependency.

    If oldnode is not None, check if the patch content (without commit message
    and metadata) has changed before creating another diff.

    If actions is not None, they will be appended to the transaction.
    """
    repo = ctx.repo()
    if oldnode:
        diffopts = mdiff.diffopts(git=True, context=32767)
        oldctx = repo.unfiltered()[oldnode]
        neednewdiff = (getdiff(ctx, diffopts) != getdiff(oldctx, diffopts))
    else:
        neednewdiff = True

    transactions = []
    if neednewdiff:
        diff = creatediff(ctx)
        transactions.append({b'type': b'update', b'value': diff[r'phid']})
    else:
        # Even if we don't need to upload a new diff because the patch content
        # does not change. We might still need to update its metadata so
        # pushers could know the correct node metadata.
        assert olddiff
        diff = olddiff
    writediffproperties(ctx, diff)

    # Use a temporary summary to set dependency. There might be better ways but
    # I cannot find them for now. But do not do that if we are updating an
    # existing revision (revid is not None) since that introduces visible
    # churns (someone edited "Summary" twice) on the web page.
    if parentrevid and revid is None:
        summary = b'Depends on D%s' % parentrevid
        transactions += [{b'type': b'summary', b'value': summary},
                         {b'type': b'summary', b'value': b' '}]

    if actions:
        transactions += actions

    # Parse commit message and update related fields.
    desc = ctx.description()
    info = callconduit(repo, b'differential.parsecommitmessage',
                       {b'corpus': desc})
    for k, v in info[r'fields'].items():
        if k in [b'title', b'summary', b'testPlan']:
            transactions.append({b'type': k, b'value': v})

    params = {b'transactions': transactions}
    if revid is not None:
        # Update an existing Differential Revision
        params[b'objectIdentifier'] = revid

    revision = callconduit(repo, b'differential.revision.edit', params)
    if not revision:
        raise error.Abort(_(b'cannot create revision for %s') % ctx)

    return revision, diff

def userphids(repo, names):
    """convert user names to PHIDs"""
    query = {b'constraints': {b'usernames': names}}
    result = callconduit(repo, b'user.search', query)
    # username not found is not an error of the API. So check if we have missed
    # some names here.
    data = result[r'data']
    resolved = set(entry[r'fields'][r'username'] for entry in data)
    unresolved = set(names) - resolved
    if unresolved:
        raise error.Abort(_(b'unknown username: %s')
                          % b' '.join(sorted(unresolved)))
    return [entry[r'phid'] for entry in data]

@command(b'phabsend',
         [(b'r', b'rev', [], _(b'revisions to send'), _(b'REV')),
          (b'', b'amend', True, _(b'update commit messages')),
          (b'', b'reviewer', [], _(b'specify reviewers')),
          (b'', b'confirm', None, _(b'ask for confirmation before sending'))],
         _(b'REV [OPTIONS]'))
def phabsend(ui, repo, *revs, **opts):
    """upload changesets to Phabricator

    If there are multiple revisions specified, they will be send as a stack
    with a linear dependencies relationship using the order specified by the
    revset.

    For the first time uploading changesets, local tags will be created to
    maintain the association. After the first time, phabsend will check
    obsstore and tags information so it can figure out whether to update an
    existing Differential Revision, or create a new one.

    If --amend is set, update commit messages so they have the
    ``Differential Revision`` URL, remove related tags. This is similar to what
    arcanist will do, and is more desired in author-push workflows. Otherwise,
    use local tags to record the ``Differential Revision`` association.

    The --confirm option lets you confirm changesets before sending them. You
    can also add following to your configuration file to make it default
    behaviour::

        [phabsend]
        confirm = true

    phabsend will check obsstore and the above association to decide whether to
    update an existing Differential Revision, or create a new one.
    """
    revs = list(revs) + opts.get(b'rev', [])
    revs = scmutil.revrange(repo, revs)

    if not revs:
        raise error.Abort(_(b'phabsend requires at least one changeset'))
    if opts.get(b'amend'):
        cmdutil.checkunfinished(repo)

    # {newnode: (oldnode, olddiff, olddrev}
    oldmap = getoldnodedrevmap(repo, [repo[r].node() for r in revs])

    confirm = ui.configbool(b'phabsend', b'confirm')
    confirm |= bool(opts.get(b'confirm'))
    if confirm:
        confirmed = _confirmbeforesend(repo, revs, oldmap)
        if not confirmed:
            raise error.Abort(_(b'phabsend cancelled'))

    actions = []
    reviewers = opts.get(b'reviewer', [])
    if reviewers:
        phids = userphids(repo, reviewers)
        actions.append({b'type': b'reviewers.add', b'value': phids})

    drevids = [] # [int]
    diffmap = {} # {newnode: diff}

    # Send patches one by one so we know their Differential Revision IDs and
    # can provide dependency relationship
    lastrevid = None
    for rev in revs:
        ui.debug(b'sending rev %d\n' % rev)
        ctx = repo[rev]

        # Get Differential Revision ID
        oldnode, olddiff, revid = oldmap.get(ctx.node(), (None, None, None))
        if oldnode != ctx.node() or opts.get(b'amend'):
            # Create or update Differential Revision
            revision, diff = createdifferentialrevision(
                ctx, revid, lastrevid, oldnode, olddiff, actions)
            diffmap[ctx.node()] = diff
            newrevid = int(revision[r'object'][r'id'])
            if revid:
                action = b'updated'
            else:
                action = b'created'

            # Create a local tag to note the association, if commit message
            # does not have it already
            m = _differentialrevisiondescre.search(ctx.description())
            if not m or int(m.group(b'id')) != newrevid:
                tagname = b'D%d' % newrevid
                tags.tag(repo, tagname, ctx.node(), message=None, user=None,
                         date=None, local=True)
        else:
            # Nothing changed. But still set "newrevid" so the next revision
            # could depend on this one.
            newrevid = revid
            action = b'skipped'

        actiondesc = ui.label(
            {b'created': _(b'created'),
             b'skipped': _(b'skipped'),
             b'updated': _(b'updated')}[action],
            b'phabricator.action.%s' % action)
        drevdesc = ui.label(b'D%s' % newrevid, b'phabricator.drev')
        nodedesc = ui.label(bytes(ctx), b'phabricator.node')
        desc = ui.label(ctx.description().split(b'\n')[0], b'phabricator.desc')
        ui.write(_(b'%s - %s - %s: %s\n') % (drevdesc, actiondesc, nodedesc,
                                             desc))
        drevids.append(newrevid)
        lastrevid = newrevid

    # Update commit messages and remove tags
    if opts.get(b'amend'):
        unfi = repo.unfiltered()
        drevs = callconduit(repo, b'differential.query', {b'ids': drevids})
        with repo.wlock(), repo.lock(), repo.transaction(b'phabsend'):
            wnode = unfi[b'.'].node()
            mapping = {} # {oldnode: [newnode]}
            for i, rev in enumerate(revs):
                old = unfi[rev]
                drevid = drevids[i]
                drev = [d for d in drevs if int(d[r'id']) == drevid][0]
                newdesc = getdescfromdrev(drev)
                newdesc = encoding.unitolocal(newdesc)
                # Make sure commit message contain "Differential Revision"
                if old.description() != newdesc:
                    parents = [
                        mapping.get(old.p1().node(), (old.p1(),))[0],
                        mapping.get(old.p2().node(), (old.p2(),))[0],
                    ]
                    new = context.metadataonlyctx(
                        repo, old, parents=parents, text=newdesc,
                        user=old.user(), date=old.date(), extra=old.extra())

                    newnode = new.commit()

                    mapping[old.node()] = [newnode]
                    # Update diff property
                    writediffproperties(unfi[newnode], diffmap[old.node()])
                # Remove local tags since it's no longer necessary
                tagname = b'D%d' % drevid
                if tagname in repo.tags():
                    tags.tag(repo, tagname, nullid, message=None, user=None,
                             date=None, local=True)
            scmutil.cleanupnodes(repo, mapping, b'phabsend', fixphase=True)
            if wnode in mapping:
                unfi.setparents(mapping[wnode][0])

# Map from "hg:meta" keys to header understood by "hg import". The order is
# consistent with "hg export" output.
_metanamemap = util.sortdict([(r'user', b'User'), (r'date', b'Date'),
                              (r'node', b'Node ID'), (r'parent', b'Parent ')])

def _confirmbeforesend(repo, revs, oldmap):
    url, token = readurltoken(repo)
    ui = repo.ui
    for rev in revs:
        ctx = repo[rev]
        desc = ctx.description().splitlines()[0]
        oldnode, olddiff, drevid = oldmap.get(ctx.node(), (None, None, None))
        if drevid:
            drevdesc = ui.label(b'D%s' % drevid, b'phabricator.drev')
        else:
            drevdesc = ui.label(_(b'NEW'), b'phabricator.drev')

        ui.write(_(b'%s - %s: %s\n')
                 % (drevdesc,
                    ui.label(bytes(ctx), b'phabricator.node'),
                    ui.label(desc, b'phabricator.desc')))

    if ui.promptchoice(_(b'Send the above changes to %s (yn)?'
                         b'$$ &Yes $$ &No') % url):
        return False

    return True

_knownstatusnames = {b'accepted', b'needsreview', b'needsrevision', b'closed',
                     b'abandoned'}

def _getstatusname(drev):
    """get normalized status name from a Differential Revision"""
    return drev[r'statusName'].replace(b' ', b'').lower()

# Small language to specify differential revisions. Support symbols: (), :X,
# +, and -.

_elements = {
    # token-type: binding-strength, primary, prefix, infix, suffix
    b'(':      (12, None, (b'group', 1, b')'), None, None),
    b':':      (8, None, (b'ancestors', 8), None, None),
    b'&':      (5,  None, None, (b'and_', 5), None),
    b'+':      (4,  None, None, (b'add', 4), None),
    b'-':      (4,  None, None, (b'sub', 4), None),
    b')':      (0,  None, None, None, None),
    b'symbol': (0, b'symbol', None, None, None),
    b'end':    (0, None, None, None, None),
}

def _tokenize(text):
    view = memoryview(text) # zero-copy slice
    special = b'():+-& '
    pos = 0
    length = len(text)
    while pos < length:
        symbol = b''.join(itertools.takewhile(lambda ch: ch not in special,
                                              view[pos:]))
        if symbol:
            yield (b'symbol', symbol, pos)
            pos += len(symbol)
        else: # special char, ignore space
            if text[pos] != b' ':
                yield (text[pos], None, pos)
            pos += 1
    yield (b'end', None, pos)

def _parse(text):
    tree, pos = parser.parser(_elements).parse(_tokenize(text))
    if pos != len(text):
        raise error.ParseError(b'invalid token', pos)
    return tree

def _parsedrev(symbol):
    """str -> int or None, ex. 'D45' -> 45; '12' -> 12; 'x' -> None"""
    if symbol.startswith(b'D') and symbol[1:].isdigit():
        return int(symbol[1:])
    if symbol.isdigit():
        return int(symbol)

def _prefetchdrevs(tree):
    """return ({single-drev-id}, {ancestor-drev-id}) to prefetch"""
    drevs = set()
    ancestordrevs = set()
    op = tree[0]
    if op == b'symbol':
        r = _parsedrev(tree[1])
        if r:
            drevs.add(r)
    elif op == b'ancestors':
        r, a = _prefetchdrevs(tree[1])
        drevs.update(r)
        ancestordrevs.update(r)
        ancestordrevs.update(a)
    else:
        for t in tree[1:]:
            r, a = _prefetchdrevs(t)
            drevs.update(r)
            ancestordrevs.update(a)
    return drevs, ancestordrevs

def querydrev(repo, spec):
    """return a list of "Differential Revision" dicts

    spec is a string using a simple query language, see docstring in phabread
    for details.

    A "Differential Revision dict" looks like:

        {
            "id": "2",
            "phid": "PHID-DREV-672qvysjcczopag46qty",
            "title": "example",
            "uri": "https://phab.example.com/D2",
            "dateCreated": "1499181406",
            "dateModified": "1499182103",
            "authorPHID": "PHID-USER-tv3ohwc4v4jeu34otlye",
            "status": "0",
            "statusName": "Needs Review",
            "properties": [],
            "branch": null,
            "summary": "",
            "testPlan": "",
            "lineCount": "2",
            "activeDiffPHID": "PHID-DIFF-xoqnjkobbm6k4dk6hi72",
            "diffs": [
              "3",
              "4",
            ],
            "commits": [],
            "reviewers": [],
            "ccs": [],
            "hashes": [],
            "auxiliary": {
              "phabricator:projects": [],
              "phabricator:depends-on": [
                "PHID-DREV-gbapp366kutjebt7agcd"
              ]
            },
            "repositoryPHID": "PHID-REPO-hub2hx62ieuqeheznasv",
            "sourcePath": null
        }
    """
    def fetch(params):
        """params -> single drev or None"""
        key = (params.get(r'ids') or params.get(r'phids') or [None])[0]
        if key in prefetched:
            return prefetched[key]
        drevs = callconduit(repo, b'differential.query', params)
        # Fill prefetched with the result
        for drev in drevs:
            prefetched[drev[r'phid']] = drev
            prefetched[int(drev[r'id'])] = drev
        if key not in prefetched:
            raise error.Abort(_(b'cannot get Differential Revision %r')
                              % params)
        return prefetched[key]

    def getstack(topdrevids):
        """given a top, get a stack from the bottom, [id] -> [id]"""
        visited = set()
        result = []
        queue = [{r'ids': [i]} for i in topdrevids]
        while queue:
            params = queue.pop()
            drev = fetch(params)
            if drev[r'id'] in visited:
                continue
            visited.add(drev[r'id'])
            result.append(int(drev[r'id']))
            auxiliary = drev.get(r'auxiliary', {})
            depends = auxiliary.get(r'phabricator:depends-on', [])
            for phid in depends:
                queue.append({b'phids': [phid]})
        result.reverse()
        return smartset.baseset(result)

    # Initialize prefetch cache
    prefetched = {} # {id or phid: drev}

    tree = _parse(spec)
    drevs, ancestordrevs = _prefetchdrevs(tree)

    # developer config: phabricator.batchsize
    batchsize = repo.ui.configint(b'phabricator', b'batchsize')

    # Prefetch Differential Revisions in batch
    tofetch = set(drevs)
    for r in ancestordrevs:
        tofetch.update(range(max(1, r - batchsize), r + 1))
    if drevs:
        fetch({r'ids': list(tofetch)})
    validids = sorted(set(getstack(list(ancestordrevs))) | set(drevs))

    # Walk through the tree, return smartsets
    def walk(tree):
        op = tree[0]
        if op == b'symbol':
            drev = _parsedrev(tree[1])
            if drev:
                return smartset.baseset([drev])
            elif tree[1] in _knownstatusnames:
                drevs = [r for r in validids
                         if _getstatusname(prefetched[r]) == tree[1]]
                return smartset.baseset(drevs)
            else:
                raise error.Abort(_(b'unknown symbol: %s') % tree[1])
        elif op in {b'and_', b'add', b'sub'}:
            assert len(tree) == 3
            return getattr(operator, op)(walk(tree[1]), walk(tree[2]))
        elif op == b'group':
            return walk(tree[1])
        elif op == b'ancestors':
            return getstack(walk(tree[1]))
        else:
            raise error.ProgrammingError(b'illegal tree: %r' % tree)

    return [prefetched[r] for r in walk(tree)]

def getdescfromdrev(drev):
    """get description (commit message) from "Differential Revision"

    This is similar to differential.getcommitmessage API. But we only care
    about limited fields: title, summary, test plan, and URL.
    """
    title = drev[r'title']
    summary = drev[r'summary'].rstrip()
    testplan = drev[r'testPlan'].rstrip()
    if testplan:
        testplan = b'Test Plan:\n%s' % testplan
    uri = b'Differential Revision: %s' % drev[r'uri']
    return b'\n\n'.join(filter(None, [title, summary, testplan, uri]))

def getdiffmeta(diff):
    """get commit metadata (date, node, user, p1) from a diff object

    The metadata could be "hg:meta", sent by phabsend, like:

        "properties": {
          "hg:meta": {
            "date": "1499571514 25200",
            "node": "98c08acae292b2faf60a279b4189beb6cff1414d",
            "user": "Foo Bar <[e-mail]>",
            "parent": "6d0abad76b30e4724a37ab8721d630394070fe16"
          }
        }

    Or converted from "local:commits", sent by "arc", like:

        "properties": {
          "local:commits": {
            "98c08acae292b2faf60a279b4189beb6cff1414d": {
              "author": "Foo Bar",
              "time": 1499546314,
              "branch": "default",
              "tag": "",
              "commit": "98c08acae292b2faf60a279b4189beb6cff1414d",
              "rev": "98c08acae292b2faf60a279b4189beb6cff1414d",
              "local": "1000",
              "parents": ["6d0abad76b30e4724a37ab8721d630394070fe16"],
              "summary": "...",
              "message": "...",
              "authorEmail": "[e-mail]"
            }
          }
        }

    Note: metadata extracted from "local:commits" will lose time zone
    information.
    """
    props = diff.get(r'properties') or {}
    meta = props.get(r'hg:meta')
    if not meta and props.get(r'local:commits'):
        commit = sorted(props[r'local:commits'].values())[0]
        meta = {
            r'date': r'%d 0' % commit[r'time'],
            r'node': commit[r'rev'],
            r'user': r'%s <%s>' % (commit[r'author'], commit[r'authorEmail']),
        }
        if len(commit.get(r'parents', ())) >= 1:
            meta[r'parent'] = commit[r'parents'][0]
    return meta or {}

def readpatch(repo, drevs, write):
    """generate plain-text patch readable by 'hg import'

    write is usually ui.write. drevs is what "querydrev" returns, results of
    "differential.query".
    """
    # Prefetch hg:meta property for all diffs
    diffids = sorted(set(max(int(v) for v in drev[r'diffs']) for drev in drevs))
    diffs = callconduit(repo, b'differential.querydiffs', {b'ids': diffids})

    # Generate patch for each drev
    for drev in drevs:
        repo.ui.note(_(b'reading D%s\n') % drev[r'id'])

        diffid = max(int(v) for v in drev[r'diffs'])
        body = callconduit(repo, b'differential.getrawdiff',
                           {b'diffID': diffid})
        desc = getdescfromdrev(drev)
        header = b'# HG changeset patch\n'

        # Try to preserve metadata from hg:meta property. Write hg patch
        # headers that can be read by the "import" command. See patchheadermap
        # and extract in mercurial/patch.py for supported headers.
        meta = getdiffmeta(diffs[str(diffid)])
        for k in _metanamemap.keys():
            if k in meta:
                header += b'# %s %s\n' % (_metanamemap[k], meta[k])

        content = b'%s%s\n%s' % (header, desc, body)
        write(encoding.unitolocal(content))

@command(b'phabread',
         [(b'', b'stack', False, _(b'read dependencies'))],
         _(b'DREVSPEC [OPTIONS]'))
def phabread(ui, repo, spec, **opts):
    """print patches from Phabricator suitable for importing

    DREVSPEC could be a Differential Revision identity, like ``D123``, or just
    the number ``123``. It could also have common operators like ``+``, ``-``,
    ``&``, ``(``, ``)`` for complex queries. Prefix ``:`` could be used to
    select a stack.

    ``abandoned``, ``accepted``, ``closed``, ``needsreview``, ``needsrevision``
    could be used to filter patches by status. For performance reason, they
    only represent a subset of non-status selections and cannot be used alone.

    For example, ``:D6+8-(2+D4)`` selects a stack up to D6, plus D8 and exclude
    D2 and D4. ``:D9 & needsreview`` selects "Needs Review" revisions in a
    stack up to D9.

    If --stack is given, follow dependencies information and read all patches.
    It is equivalent to the ``:`` operator.
    """
    if opts.get(b'stack'):
        spec = b':(%s)' % spec
    drevs = querydrev(repo, spec)
    readpatch(repo, drevs, ui.write)

@command(b'phabupdate',
         [(b'', b'accept', False, _(b'accept revisions')),
          (b'', b'reject', False, _(b'reject revisions')),
          (b'', b'abandon', False, _(b'abandon revisions')),
          (b'', b'reclaim', False, _(b'reclaim revisions')),
          (b'm', b'comment', b'', _(b'comment on the last revision')),
          ], _(b'DREVSPEC [OPTIONS]'))
def phabupdate(ui, repo, spec, **opts):
    """update Differential Revision in batch

    DREVSPEC selects revisions. See :hg:`help phabread` for its usage.
    """
    flags = [n for n in b'accept reject abandon reclaim'.split() if opts.get(n)]
    if len(flags) > 1:
        raise error.Abort(_(b'%s cannot be used together') % b', '.join(flags))

    actions = []
    for f in flags:
        actions.append({b'type': f, b'value': b'true'})

    drevs = querydrev(repo, spec)
    for i, drev in enumerate(drevs):
        if i + 1 == len(drevs) and opts.get(b'comment'):
            actions.append({b'type': b'comment', b'value': opts[b'comment']})
        if actions:
            params = {b'objectIdentifier': drev[r'phid'],
                      b'transactions': actions}
            callconduit(repo, b'differential.revision.edit', params)

templatekeyword = registrar.templatekeyword()

@templatekeyword(b'phabreview', requires={b'ctx'})
def template_review(context, mapping):
    """:phabreview: Object describing the review for this changeset.
    Has attributes `url` and `id`.
    """
    ctx = context.resource(mapping, b'ctx')
    m = _differentialrevisiondescre.search(ctx.description())
    if m:
        return {
            b'url': m.group(b'url'),
            b'id': b"D{}".format(m.group(b'id')),
        }