# nodemap.py - nodemap related code and utilities

#

# Copyright 2019 Pierre-Yves David <pierre-yves.david@octobus.net>

# Copyright 2019 George Racinet <georges.racinet@octobus.net>

#

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

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

from __future__ import absolute_import

import errno

import os

import re

import struct

from ..node import hex

from .. import (

    error,

    util,

)

class NodeMap(dict):

    def __missing__(self, x):

        raise error.RevlogError(b'unknown node: %s' % x)

def persisted_data(revlog):

    """read the nodemap for a revlog from disk"""

    if revlog.nodemap_file is None:

        return None

    pdata = revlog.opener.tryread(revlog.nodemap_file)

    if not pdata:

        return None

    offset = 0

    (version,) = S_VERSION.unpack(pdata[offset : offset + S_VERSION.size])

    if version != ONDISK_VERSION:

        return None

    offset += S_VERSION.size

    headers = S_HEADER.unpack(pdata[offset : offset + S_HEADER.size])

    uid_size, tip_rev, data_length, data_unused, tip_node_size = headers

    offset += S_HEADER.size

    docket = NodeMapDocket(pdata[offset : offset + uid_size])

    offset += uid_size

    docket.tip_rev = tip_rev

    docket.tip_node = pdata[offset : offset + tip_node_size]

    docket.data_length = data_length

    docket.data_unused = data_unused

    filename = _rawdata_filepath(revlog, docket)

    use_mmap = revlog.opener.options.get(b"persistent-nodemap.mmap")

    try:

        with revlog.opener(filename) as fd:

            if use_mmap:

                data = util.buffer(util.mmapread(fd, data_length))

            else:

                data = fd.read(data_length)

    except (IOError, OSError) as e:

        if e.errno == errno.ENOENT:

            return None

        else:

            raise

    if len(data) < data_length:

        return None

    return docket, data

def setup_persistent_nodemap(tr, revlog):

    """Install whatever is needed transaction side to persist a nodemap on disk

    (only actually persist the nodemap if this is relevant for this revlog)

    """

    if revlog._inline:

        return  # inlined revlog are too small for this to be relevant

    if revlog.nodemap_file is None:

        return  # we do not use persistent_nodemap on this revlog

    # we need to happen after the changelog finalization, in that use "cl-"

    callback_id = b"nm-revlog-persistent-nodemap-%s" % revlog.nodemap_file

    if tr.hasfinalize(callback_id):

        return  # no need to register again

    tr.addpending(

        callback_id, lambda tr: persist_nodemap(tr, revlog, pending=True)

    )

    tr.addfinalize(callback_id, lambda tr: persist_nodemap(tr, revlog))

class _NoTransaction(object):

    """transaction like object to update the nodemap outside a transaction"""

    def __init__(self):

        self._postclose = {}

    def addpostclose(self, callback_id, callback_func):

        self._postclose[callback_id] = callback_func

    def registertmp(self, *args, **kwargs):

        pass

    def addbackup(self, *args, **kwargs):

        pass

    def add(self, *args, **kwargs):

        pass

    def addabort(self, *args, **kwargs):

        pass

    def _report(self, *args):

        pass

def update_persistent_nodemap(revlog):

    """update the persistent nodemap right now

    To be used for updating the nodemap on disk outside of a normal transaction

    setup (eg, `debugupdatecache`).

    """

    if revlog._inline:

        return  # inlined revlog are too small for this to be relevant

    if revlog.nodemap_file is None:

        return  # we do not use persistent_nodemap on this revlog

    notr = _NoTransaction()

    persist_nodemap(notr, revlog)

    for k in sorted(notr._postclose):

        notr._postclose[k](None)

def delete_nodemap(tr, repo, revlog):

    """ Delete nodemap data on disk for a given revlog"""

    if revlog.nodemap_file is None:

        msg = "calling persist nodemap on a revlog without the feature enabled"

        raise error.ProgrammingError(msg)

    repo.svfs.unlink(revlog.nodemap_file)

def persist_nodemap(tr, revlog, pending=False, force=False):

    """Write nodemap data on disk for a given revlog"""

    if getattr(revlog, 'filteredrevs', ()):

        raise error.ProgrammingError(

            "cannot persist nodemap of a filtered changelog"

        )

    if revlog.nodemap_file is None:

        if force:

            revlog.nodemap_file = get_nodemap_file(

                revlog.opener, revlog.indexfile

            )

        else:

            msg = "calling persist nodemap on a revlog without the feature enabled"

            raise error.ProgrammingError(msg)

    can_incremental = util.safehasattr(revlog.index, "nodemap_data_incremental")

    ondisk_docket = revlog._nodemap_docket

    feed_data = util.safehasattr(revlog.index, "update_nodemap_data")

    use_mmap = revlog.opener.options.get(b"persistent-nodemap.mmap")

    data = None

    # first attemp an incremental update of the data

    if can_incremental and ondisk_docket is not None:

        target_docket = revlog._nodemap_docket.copy()

        (

            src_docket,

            data_changed_count,

            data,

        ) = revlog.index.nodemap_data_incremental()

        new_length = target_docket.data_length + len(data)

        new_unused = target_docket.data_unused + data_changed_count

        if src_docket != target_docket:

            data = None

        elif new_length <= (new_unused * 10):  # under 10% of unused data

            data = None

        else:

            datafile = _rawdata_filepath(revlog, target_docket)

            # EXP-TODO: if this is a cache, this should use a cache vfs, not a

            # store vfs

            tr.add(datafile, target_docket.data_length)

            with revlog.opener(datafile, b'r+') as fd:

                fd.seek(target_docket.data_length)

                fd.write(data)

                if feed_data:

                    if use_mmap:

                        fd.seek(0)

                        new_data = fd.read(new_length)

                    else:

                        fd.flush()

                        new_data = util.buffer(util.mmapread(fd, new_length))

            target_docket.data_length = new_length

            target_docket.data_unused = new_unused

    if data is None:

        # otherwise fallback to a full new export

        target_docket = NodeMapDocket()

        datafile = _rawdata_filepath(revlog, target_docket)

        if util.safehasattr(revlog.index, "nodemap_data_all"):

            data = revlog.index.nodemap_data_all()

        else:

            data = persistent_data(revlog.index)

        # EXP-TODO: if this is a cache, this should use a cache vfs, not a

        # store vfs

        tryunlink = revlog.opener.tryunlink

        def abortck(tr):

            tryunlink(datafile)

        callback_id = b"delete-%s" % datafile

        # some flavor of the transaction abort does not cleanup new file, it

        # simply empty them.

        tr.addabort(callback_id, abortck)

        with revlog.opener(datafile, b'w+') as fd:

            fd.write(data)

            if feed_data:

                if use_mmap:

                    new_data = data

                else:

                    fd.flush()

                    new_data = util.buffer(util.mmapread(fd, len(data)))

        target_docket.data_length = len(data)

    target_docket.tip_rev = revlog.tiprev()

    target_docket.tip_node = revlog.node(target_docket.tip_rev)

    # EXP-TODO: if this is a cache, this should use a cache vfs, not a

    # store vfs

    file_path = revlog.nodemap_file

    if pending:

        file_path += b'.a'

        tr.registertmp(file_path)

    else:

        tr.addbackup(file_path)

    with revlog.opener(file_path, b'w', atomictemp=True) as fp:

        fp.write(target_docket.serialize())

    revlog._nodemap_docket = target_docket

    if feed_data:

        revlog.index.update_nodemap_data(target_docket, new_data)

    # search for old index file in all cases, some older process might have

    # left one behind.

    olds = _other_rawdata_filepath(revlog, target_docket)

    if olds:

        realvfs = getattr(revlog, '_realopener', revlog.opener)

        def cleanup(tr):

            for oldfile in olds:

                realvfs.tryunlink(oldfile)

        callback_id = b"revlog-cleanup-nodemap-%s" % revlog.nodemap_file

        tr.addpostclose(callback_id, cleanup)

### Nodemap docket file

#

# The nodemap data are stored on disk using 2 files:

#

# * a raw data files containing a persistent nodemap

#   (see `Nodemap Trie` section)

#

# * a small "docket" file containing medatadata

#

# While the nodemap data can be multiple tens of megabytes, the "docket" is

# small, it is easy to update it automatically or to duplicated its content

# during a transaction.

#

# Multiple raw data can exist at the same time (The currently valid one and a

# new one beind used by an in progress transaction). To accomodate this, the

# filename hosting the raw data has a variable parts. The exact filename is

# specified inside the "docket" file.

#

# The docket file contains information to find, qualify and validate the raw

# data. Its content is currently very light, but it will expand as the on disk

# nodemap gains the necessary features to be used in production.

ONDISK_VERSION = 1

S_VERSION = struct.Struct(">B")

S_HEADER = struct.Struct(">BQQQQ")

ID_SIZE = 8

def _make_uid():

    """return a new unique identifier.

    The identifier is random and composed of ascii characters."""

    return hex(os.urandom(ID_SIZE))

class NodeMapDocket(object):

    """metadata associated with persistent nodemap data

    The persistent data may come from disk or be on their way to disk.

    """

    def __init__(self, uid=None):

        if uid is None:

            uid = _make_uid()

        # a unique identifier for the data file:

        #   - When new data are appended, it is preserved.

        #   - When a new data file is created, a new identifier is generated.

        self.uid = uid

        # the tipmost revision stored in the data file. This revision and all

        # revision before it are expected to be encoded in the data file.

        self.tip_rev = None

        # the node of that tipmost revision, if it mismatch the current index

        # data the docket is not valid for the current index and should be

        # discarded.

        #

        # note: this method is not perfect as some destructive operation could

        # preserve the same tip_rev + tip_node while altering lower revision.

        # However this multiple other caches have the same vulnerability (eg:

        # brancmap cache).

        self.tip_node = None

        # the size (in bytes) of the persisted data to encode the nodemap valid

        # for `tip_rev`.

        #   - data file shorter than this are corrupted,

        #   - any extra data should be ignored.

        self.data_length = None

        # the amount (in bytes) of "dead" data, still in the data file but no

        # longer used for the nodemap.

        self.data_unused = 0

    def copy(self):

        new = NodeMapDocket(uid=self.uid)

        new.tip_rev = self.tip_rev

        new.tip_node = self.tip_node

        new.data_length = self.data_length

        new.data_unused = self.data_unused

        return new

    def __cmp__(self, other):

        if self.uid < other.uid:

            return -1

        if self.uid > other.uid:

            return 1

        elif self.data_length < other.data_length:

            return -1

        elif self.data_length > other.data_length:

            return 1

        return 0

    def __eq__(self, other):

        return self.uid == other.uid and self.data_length == other.data_length

    def serialize(self):

        """return serialized bytes for a docket using the passed uid"""

        data = []

        data.append(S_VERSION.pack(ONDISK_VERSION))

        headers = (

            len(self.uid),

            self.tip_rev,

            self.data_length,

            self.data_unused,

            len(self.tip_node),

        )

        data.append(S_HEADER.pack(*headers))

        data.append(self.uid)

        data.append(self.tip_node)

        return b''.join(data)

def _rawdata_filepath(revlog, docket):

    """The (vfs relative) nodemap's rawdata file for a given uid"""

    if revlog.nodemap_file.endswith(b'.n.a'):

        prefix = revlog.nodemap_file[:-4]

    else:

        prefix = revlog.nodemap_file[:-2]

    return b"%s-%s.nd" % (prefix, docket.uid)

def _other_rawdata_filepath(revlog, docket):

    prefix = revlog.nodemap_file[:-2]

    pattern = re.compile(br"(^|/)%s-[0-9a-f]+\.nd$" % prefix)

    new_file_path = _rawdata_filepath(revlog, docket)

    new_file_name = revlog.opener.basename(new_file_path)

    dirpath = revlog.opener.dirname(new_file_path)

    others = []

    for f in revlog.opener.listdir(dirpath):

        if pattern.match(f) and f != new_file_name:

            others.append(f)

    return others

### Nodemap Trie

#

# This is a simple reference implementation to compute and persist a nodemap

# trie. This reference implementation is write only. The python version of this