# obsolete.py - obsolete markers handling

#

# Copyright 2012 Pierre-Yves David <pierre-yves.david@ens-lyon.org>

#                Logilab SA        <contact@logilab.fr>

#

# This software may be used and distributed according to the terms of the

# GNU General Public License version 2 or any later version.

"""Obsolete marker handling

An obsolete marker maps an old changeset to a list of new

changesets. If the list of new changesets is empty, the old changeset

is said to be "killed". Otherwise, the old changeset is being

"replaced" by the new changesets.

Obsolete markers can be used to record and distribute changeset graph

transformations performed by history rewrite operations, and help

building new tools to reconcile conflicting rewrite actions. To

facilitate conflict resolution, markers include various annotations

besides old and news changeset identifiers, such as creation date or

author name.

The old obsoleted changeset is called a "precursor" and possible

replacements are called "successors". Markers that used changeset X as

a precursor are called "successor markers of X" because they hold

information about the successors of X. Markers that use changeset Y as

a successors are call "precursor markers of Y" because they hold

information about the precursors of Y.

Examples:

- When changeset A is replaced by changeset A', one marker is stored:

    (A, (A',))

- When changesets A and B are folded into a new changeset C, two markers are

  stored:

    (A, (C,)) and (B, (C,))

- When changeset A is simply "pruned" from the graph, a marker is created:

    (A, ())

- When changeset A is split into B and C, a single marker are used:

    (A, (C, C))

  We use a single marker to distinguish the "split" case from the "divergence"

  case. If two independent operations rewrite the same changeset A in to A' and

  A'', we have an error case: divergent rewriting. We can detect it because

  two markers will be created independently:

  (A, (B,)) and (A, (C,))

Format

------

Markers are stored in an append-only file stored in

'.hg/store/obsstore'.

The file starts with a version header:

- 1 unsigned byte: version number, starting at zero.

The header is followed by the markers. Each marker is made of:

- 1 unsigned byte: number of new changesets "N", can be zero.

- 1 unsigned 32-bits integer: metadata size "M" in bytes.

- 1 byte: a bit field. It is reserved for flags used in common

  obsolete marker operations, to avoid repeated decoding of metadata

  entries.

- 20 bytes: obsoleted changeset identifier.

- N*20 bytes: new changesets identifiers.

- M bytes: metadata as a sequence of nul-terminated strings. Each

  string contains a key and a value, separated by a colon ':', without

  additional encoding. Keys cannot contain '\0' or ':' and values

  cannot contain '\0'.

"""

import struct

import util, base85, node

import phases

from i18n import _

_pack = struct.pack

_unpack = struct.unpack

_SEEK_END = 2 # os.SEEK_END was introduced in Python 2.5

# the obsolete feature is not mature enough to be enabled by default.

# you have to rely on third party extension extension to enable this.

_enabled = False

# data used for parsing and writing

_fm0version = 0

_fm0fixed   = '>BIB20s'

_fm0node = '20s'

_fm0fsize = struct.calcsize(_fm0fixed)

_fm0fnodesize = struct.calcsize(_fm0node)

### obsolescence marker flag

## bumpedfix flag

#

# When a changeset A' succeed to a changeset A which became public, we call A'

# "bumped" because it's a successors of a public changesets

#

# o    A' (bumped)

# |`:

# | o  A

# |/

# o    Z

#

# The way to solve this situation is to create a new changeset Ad as children

# of A. This changeset have the same content than A'. So the diff from A to A'

# is the same than the diff from A to Ad. Ad is marked as a successors of A'

#

# o   Ad

# |`:

# | x A'

# |'|

# o | A

# |/

# o Z

#

# But by transitivity Ad is also a successors of A. To avoid having Ad marked

# as bumped too, we add the `bumpedfix` flag to the marker. <A', (Ad,)>.

# This flag mean that the successors express the changes between the public and

# bumped version and fix the situation, breaking the transitivity of

# "bumped" here.

bumpedfix = 1

def _readmarkers(data):

    """Read and enumerate markers from raw data"""

    off = 0

    diskversion = _unpack('>B', data[off:off + 1])[0]

    off += 1

    if diskversion not in formats:

        raise util.Abort(_('parsing obsolete marker: unknown version %r')

                         % diskversion)

def _fm0readmarkers(data, off=0):

    """Read and enumerate markers from raw data in format version 0"""

    # Loop on markers

    l = len(data)

    while off + _fm0fsize <= l:

        # read fixed part

        cur = data[off:off + _fm0fsize]

        off += _fm0fsize

        nbsuc, mdsize, flags, pre = _unpack(_fm0fixed, cur)

        # read replacement

        sucs = ()

        if nbsuc:

            s = (_fm0fnodesize * nbsuc)

            cur = data[off:off + s]

            sucs = _unpack(_fm0node * nbsuc, cur)

            off += s

        # read metadata

        # (metadata will be decoded on demand)

        metadata = data[off:off + mdsize]

        if len(metadata) != mdsize:

            raise util.Abort(_('parsing obsolete marker: metadata is too '

                               'short, %d bytes expected, got %d')

                             % (mdsize, len(metadata)))

        off += mdsize

        meta = decodemeta(metadata)

        try:

            when, offset = decodemeta(metadata).pop('date', '0 0').split(' ')

            date = float(when), int(offset)

        except ValueError:

            date = (0., 0)

        parents = None

        if 'p2' in meta:

            parents = (meta.pop('p1', None), meta.pop('p2', None))

        elif 'p1' in meta:

            parents = (meta.pop('p1', None),)

        elif 'p0' in meta:

            parents = ()

        if parents is not None:

            try:

                parents = tuple(node.bin(p) for p in parents)

                # if parent content is not a nodeid, drop the data

                for p in parents:

                    if len(p) != 20:

                        parents = None

                        break

            except TypeError:

                # if content cannot be translated to nodeid drop the data.

                parents = None

        metadata = encodemeta(meta)

        yield (pre, sucs, flags, metadata, date, parents)

def _fm0encodeonemarker(marker):

    pre, sucs, flags, metadata, date, parents = marker

    metadata = decodemeta(metadata)

    metadata['date'] = '%d %i' % date

    if parents is not None:

        if not parents:

            # mark that we explicitly recorded no parents

            metadata['p0'] = ''

        for i, p in enumerate(parents):

            metadata['p%i' % (i + 1)] = node.hex(p)

    metadata = encodemeta(metadata)

    nbsuc = len(sucs)

    format = _fm0fixed + (_fm0node * nbsuc)

    data = [nbsuc, len(metadata), flags, pre]

    data.extend(sucs)

    return _pack(format, *data) + metadata

# mapping to read/write various marker formats

# <version> -> (decoder, encoder)

formats = {0: (_fm0readmarkers, _fm0encodeonemarker)}

def encodemeta(meta):

    """Return encoded metadata string to string mapping.

    Assume no ':' in key and no '\0' in both key and value."""

    for key, value in meta.iteritems():

        if ':' in key or '\0' in key:

            raise ValueError("':' and '\0' are forbidden in metadata key'")

        if '\0' in value:

            raise ValueError("':' is forbidden in metadata value'")

    return '\0'.join(['%s:%s' % (k, meta[k]) for k in sorted(meta)])

def decodemeta(data):

    """Return string to string dictionary from encoded version."""

    d = {}

    for l in data.split('\0'):

        if l:

            key, value = l.split(':')

            d[key] = value

    return d

class marker(object):

    """Wrap obsolete marker raw data"""

    def __init__(self, repo, data):

        # the repo argument will be used to create changectx in later version

        self._repo = repo

        self._data = data

        self._decodedmeta = None

    def __hash__(self):

        return hash(self._data)

    def __eq__(self, other):

        if type(other) != type(self):

            return False

        return self._data == other._data

    def precnode(self):

        """Precursor changeset node identifier"""

        return self._data[0]

    def succnodes(self):

        """List of successor changesets node identifiers"""

        return self._data[1]

    def parentnodes(self):

        """Parents of the precursors (None if not recorded)"""

        return self._data[5]

    def metadata(self):

        """Decoded metadata dictionary"""

        if self._decodedmeta is None:

            self._decodedmeta = decodemeta(self._data[3])

        return self._decodedmeta

    def date(self):

        """Creation date as (unixtime, offset)"""

        return self._data[4]

    def flags(self):

        """The flags field of the marker"""

        return self._data[2]

class obsstore(object):

    """Store obsolete markers

    Markers can be accessed with two mappings:

    - precursors[x] -> set(markers on precursors edges of x)

    - successors[x] -> set(markers on successors edges of x)

    - children[x]   -> set(markers on precursors edges of children(x)

    """

    fields = ('prec', 'succs', 'flag', 'meta', 'date', 'parents')

    # prec:    nodeid, precursor changesets

    # succs:   tuple of nodeid, successor changesets (0-N length)

    # flag:    integer, flag field carrying modifier for the markers (see doc)

    # meta:    binary blob, encoded metadata dictionary

    # date:    (float, int) tuple, date of marker creation

    # parents: (tuple of nodeid) or None, parents of precursors

    #          None is used when no data has been recorded

    def __init__(self, sopener):

        # caches for various obsolescence related cache

        self.caches = {}

        self._all = []

        self.precursors = {}

        self.successors = {}

        self.children = {}

        self.sopener = sopener

        data = sopener.tryread('obsstore')

        if data:

    def __iter__(self):

        return iter(self._all)

    def __len__(self):

        return len(self._all)

    def __nonzero__(self):

        return bool(self._all)

    def create(self, transaction, prec, succs=(), flag=0, parents=None,

               date=None, metadata=None):

        """obsolete: add a new obsolete marker

        * ensuring it is hashable

        * check mandatory metadata

        * encode metadata

        If you are a human writing code creating marker you want to use the

        `createmarkers` function in this module instead.

        return True if a new marker have been added, False if the markers

        already existed (no op).

        """

        if metadata is None:

            metadata = {}

        if date is None:

            if 'date' in metadata:

                # as a courtesy for out-of-tree extensions

                date = util.parsedate(metadata.pop('date'))

            else:

                date = util.makedate()

        if len(prec) != 20:

            raise ValueError(prec)

        for succ in succs:

            if len(succ) != 20:

                raise ValueError(succ)

        if prec in succs:

            raise ValueError(_('in-marker cycle with %s') % node.hex(prec))

        marker = (str(prec), tuple(succs), int(flag), encodemeta(metadata),

                  date, parents)

        return bool(self.add(transaction, [marker]))

    def add(self, transaction, markers):

        """Add new markers to the store

        Take care of filtering duplicate.

        Return the number of new marker."""

        if not _enabled:

            raise util.Abort('obsolete feature is not enabled on this repo')

        known = set(self._all)

        new = []

        for m in markers:

            if m not in known:

                known.add(m)

                new.append(m)

        if new:

            f = self.sopener('obsstore', 'ab')

            try:

                # Whether the file's current position is at the begin or at

                # the end after opening a file for appending is implementation

                # defined. So we must seek to the end before calling tell(),

                # or we may get a zero offset for non-zero sized files on

                # some platforms (issue3543).

                f.seek(0, _SEEK_END)

                offset = f.tell()

                transaction.add('obsstore', offset)

                # offset == 0: new file - add the version header

                for bytes in _encodemarkers(new, offset == 0):

                    f.write(bytes)

            finally:

                # XXX: f.close() == filecache invalidation == obsstore rebuilt.

                # call 'filecacheentry.refresh()'  here

                f.close()

            self._load(new)

            # new marker *may* have changed several set. invalidate the cache.

            self.caches.clear()

        return len(new)

    def mergemarkers(self, transaction, data):

        """merge a binary stream of markers inside the obsstore

        Returns the number of new markers added."""

        return self.add(transaction, markers)

    def _load(self, markers):

        for mark in markers:

            self._all.append(mark)

            pre, sucs = mark[:2]

            self.successors.setdefault(pre, set()).add(mark)

            for suc in sucs:

                self.precursors.setdefault(suc, set()).add(mark)

            parents = mark[5]

            if parents is not None:

                for p in parents:

                    self.children.setdefault(p, set()).add(mark)

        if node.nullid in self.precursors:

            raise util.Abort(_('bad obsolescence marker detected: '

                               'invalid successors nullid'))

    def relevantmarkers(self, nodes):

        """return a set of all obsolescence markers relevant to a set of nodes.

        "relevant" to a set of nodes mean:

        - marker that use this changeset as successor

        - prune marker of direct children on this changeset

        - recursive application of the two rules on precursors of these markers

        It is a set so you cannot rely on order."""

        pendingnodes = set(nodes)

        seenmarkers = set()

        seennodes = set(pendingnodes)

        precursorsmarkers = self.precursors

        children = self.children

        while pendingnodes:

            direct = set()

            for current in pendingnodes:

                direct.update(precursorsmarkers.get(current, ()))

                pruned = [m for m in children.get(current, ()) if not m[1]]

                direct.update(pruned)

            direct -= seenmarkers

            pendingnodes = set([m[0] for m in direct])

            seenmarkers |= direct

            pendingnodes -= seennodes

            seennodes |= pendingnodes

        return seenmarkers

def _encodemarkers(markers, addheader=False, version=_fm0version):

    # Kept separate from flushmarkers(), it will be reused for

    # markers exchange.

    encodeone = formats[version][1]

    if addheader:

        yield _pack('>B', _fm0version)

    for marker in markers:

        yield encodeone(marker)

# arbitrary picked to fit into 8K limit from HTTP server

# you have to take in account:

# - the version header

# - the base85 encoding

_maxpayload = 5300

def _pushkeyescape(markers):

    """encode markers into a dict suitable for pushkey exchange

    - binary data is base85 encoded

    - split in chunks smaller than 5300 bytes"""

    keys = {}

    parts = []

    currentlen = _maxpayload * 2  # ensure we create a new part

    for marker in markers:

        nextdata = _fm0encodeonemarker(marker)

        if (len(nextdata) + currentlen > _maxpayload):

            currentpart = []

            currentlen = 0

            parts.append(currentpart)

        currentpart.append(nextdata)

        currentlen += len(nextdata)

    for idx, part in enumerate(reversed(parts)):

        data = ''.join([_pack('>B', _fm0version)] + part)

        keys['dump%i' % idx] = base85.b85encode(data)

    return keys

def listmarkers(repo):

    """List markers over pushkey"""

    if not repo.obsstore:

        return {}

    return _pushkeyescape(repo.obsstore)

def pushmarker(repo, key, old, new):

    """Push markers over pushkey"""

    if not key.startswith('dump'):

        repo.ui.warn(_('unknown key: %r') % key)