~bzr-pqm/bzr/bzr.dev

#! /usr/bin/python

# Copyright (C) 2005 Canonical Ltd

#

# This program is free software; you can redistribute it and/or modify

# it under the terms of the GNU General Public License as published by

# the Free Software Foundation; either version 2 of the License, or

# (at your option) any later version.

#

# This program is distributed in the hope that it will be useful,

# but WITHOUT ANY WARRANTY; without even the implied warranty of

# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the

# GNU General Public License for more details.

#

# You should have received a copy of the GNU General Public License

# along with this program; if not, write to the Free Software

# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA

# Author: Martin Pool <mbp@canonical.com>

"""Weave - storage of related text file versions"""

# XXX: If we do weaves this way, will a merge still behave the same

# way if it's done in a different order?  That's a pretty desirable

# property.

# TODO: Nothing here so far assumes the lines are really \n newlines,

# rather than being split up in some other way.  We could accommodate

# binaries, perhaps by naively splitting on \n or perhaps using

# something like a rolling checksum.

# TODO: End marker for each version so we can stop reading?

# TODO: Check that no insertion occurs inside a deletion that was

# active in the version of the insertion.

# TODO: In addition to the SHA-1 check, perhaps have some code that

# checks structural constraints of the weave: ie that insertions are

# properly nested, that there is no text outside of an insertion, that

# insertions or deletions are not repeated, etc.

# TODO: Parallel-extract that passes back each line along with a

# description of which revisions include it.  Nice for checking all

# shas or calculating stats in parallel.

# TODO: Using a single _extract routine and then processing the output

# is probably inefficient.  It's simple enough that we can afford to

# have slight specializations for different ways its used: annotate,

# basis for add, get, etc.

# TODO: Probably the API should work only in names to hide the integer

# indexes from the user.

# TODO: Is there any potential performance win by having an add()

# variant that is passed a pre-cooked version of the single basis

# version?

# TODO: Reweave can possibly be made faster by remembering diffs

# where the basis and destination are unchanged.

# FIXME: Sometimes we will be given a parents list for a revision

# that includes some redundant parents (i.e. already a parent of 

# something in the list.)  We should eliminate them.  This can 

# be done fairly efficiently because the sequence numbers constrain

# the possible relationships.

# FIXME: the conflict markers should be *7* characters

from copy import copy

from cStringIO import StringIO

import os

import sha

import time

import warnings

from bzrlib import (

    progress,

    )

from bzrlib.trace import mutter

from bzrlib.errors import (WeaveError, WeaveFormatError, WeaveParentMismatch,

        RevisionAlreadyPresent,

        RevisionNotPresent,

        WeaveRevisionAlreadyPresent,

        WeaveRevisionNotPresent,

        )

import bzrlib.errors as errors

from bzrlib.osutils import sha_strings

import bzrlib.patiencediff

from bzrlib.tsort import topo_sort

from bzrlib.versionedfile import VersionedFile, InterVersionedFile

from bzrlib.weavefile import _read_weave_v5, write_weave_v5

class Weave(VersionedFile):

    """weave - versioned text file storage.

    A Weave manages versions of line-based text files, keeping track

    of the originating version for each line.

    To clients the "lines" of the file are represented as a list of strings.

    These strings  will typically have terminal newline characters, but

    this is not required.  In particular files commonly do not have a newline

    at the end of the file.

    Texts can be identified in either of two ways:

    * a nonnegative index number.

    * a version-id string.

    Typically the index number will be valid only inside this weave and

    the version-id is used to reference it in the larger world.

    The weave is represented as a list mixing edit instructions and

    literal text.  Each entry in _weave can be either a string (or

    unicode), or a tuple.  If a string, it means that the given line

    should be output in the currently active revisions.

    If a tuple, it gives a processing instruction saying in which

    revisions the enclosed lines are active.  The tuple has the form

    (instruction, version).

    The instruction can be '{' or '}' for an insertion block, and '['

    and ']' for a deletion block respectively.  The version is the

    integer version index.  There is no replace operator, only deletes

    and inserts.  For '}', the end of an insertion, there is no

    version parameter because it always closes the most recently

    opened insertion.

    Constraints/notes:

    * A later version can delete lines that were introduced by any

      number of ancestor versions; this implies that deletion

      instructions can span insertion blocks without regard to the

      insertion block's nesting.

    * Similarly, deletions need not be properly nested with regard to

      each other, because they might have been generated by

      independent revisions.

    * Insertions are always made by inserting a new bracketed block

      into a single point in the previous weave.  This implies they

      can nest but not overlap, and the nesting must always have later

      insertions on the inside.

    * It doesn't seem very useful to have an active insertion

      inside an inactive insertion, but it might happen.

    * Therefore, all instructions are always"considered"; that

      is passed onto and off the stack.  An outer inactive block

      doesn't disable an inner block.

    * Lines are enabled if the most recent enclosing insertion is

      active and none of the enclosing deletions are active.

    * There is no point having a deletion directly inside its own

      insertion; you might as well just not write it.  And there

      should be no way to get an earlier version deleting a later

      version.

    _weave

        Text of the weave; list of control instruction tuples and strings.

    _parents

        List of parents, indexed by version number.

        It is only necessary to store the minimal set of parents for

        each version; the parent's parents are implied.

    _sha1s

        List of hex SHA-1 of each version.

    _names

        List of symbolic names for each version.  Each should be unique.

    _name_map

        For each name, the version number.

    _weave_name

        Descriptive name of this weave; typically the filename if known.

        Set by read_weave.

    """

    __slots__ = ['_weave', '_parents', '_sha1s', '_names', '_name_map',

                 '_weave_name', '_matcher']

    def __init__(self, weave_name=None, access_mode='w', matcher=None):

        super(Weave, self).__init__(access_mode)

        self._weave = []

        self._parents = []

        self._sha1s = []

        self._names = []

        self._name_map = {}

        self._weave_name = weave_name

        if matcher is None:

            self._matcher = bzrlib.patiencediff.PatienceSequenceMatcher

        else:

            self._matcher = matcher

    def __repr__(self):

        return "Weave(%r)" % self._weave_name

    def copy(self):

        """Return a deep copy of self.

        The copy can be modified without affecting the original weave."""

        other = Weave()

        other._weave = self._weave[:]

        other._parents = self._parents[:]

        other._sha1s = self._sha1s[:]

        other._names = self._names[:]

        other._name_map = self._name_map.copy()

        other._weave_name = self._weave_name

        return other

    def __eq__(self, other):

        if not isinstance(other, Weave):

            return False

        return self._parents == other._parents \

               and self._weave == other._weave \

               and self._sha1s == other._sha1s 

    def __ne__(self, other):

        return not self.__eq__(other)

    def _idx_to_name(self, version):

        return self._names[version]

    def _lookup(self, name):

        """Convert symbolic version name to index."""

        self.check_not_reserved_id(name)

        try:

            return self._name_map[name]

        except KeyError:

            raise RevisionNotPresent(name, self._weave_name)

    def versions(self):

        """See VersionedFile.versions."""

        return self._names[:]

    def has_version(self, version_id):

        """See VersionedFile.has_version."""

        return (version_id in self._name_map)

    __contains__ = has_version

    def get_parents(self, version_id):

        """See VersionedFile.get_parent."""

        return map(self._idx_to_name, self._parents[self._lookup(version_id)])

    def _check_repeated_add(self, name, parents, text, sha1):

        """Check that a duplicated add is OK.

        If it is, return the (old) index; otherwise raise an exception.

        """

        idx = self._lookup(name)

        if sorted(self._parents[idx]) != sorted(parents) \

            or sha1 != self._sha1s[idx]:

            raise RevisionAlreadyPresent(name, self._weave_name)

        return idx

    def _add_lines(self, version_id, parents, lines, parent_texts,

       left_matching_blocks, nostore_sha, random_id, check_content):

        """See VersionedFile.add_lines."""

        idx = self._add(version_id, lines, map(self._lookup, parents),

            nostore_sha=nostore_sha)

        return sha_strings(lines), sum(map(len, lines)), idx

    def _add(self, version_id, lines, parents, sha1=None, nostore_sha=None):

        """Add a single text on top of the weave.

        Returns the index number of the newly added version.

        version_id

            Symbolic name for this version.

            (Typically the revision-id of the revision that added it.)

        parents

            List or set of direct parent version numbers.

        lines

            Sequence of lines to be added in the new version.

        :param nostore_sha: See VersionedFile.add_lines.

        """

        assert isinstance(version_id, basestring)

        self._check_lines_not_unicode(lines)

        self._check_lines_are_lines(lines)

        if not sha1:

            sha1 = sha_strings(lines)

        if sha1 == nostore_sha:

            raise errors.ExistingContent

        if version_id in self._name_map:

            return self._check_repeated_add(version_id, parents, lines, sha1)

        self._check_versions(parents)

        ## self._check_lines(lines)

        new_version = len(self._parents)

        # if we abort after here the (in-memory) weave will be corrupt because only

        # some fields are updated

        # XXX: FIXME implement a succeed-or-fail of the rest of this routine.

        #      - Robert Collins 20060226

        self._parents.append(parents[:])

        self._sha1s.append(sha1)

        self._names.append(version_id)

        self._name_map[version_id] = new_version

        if not parents:

            # special case; adding with no parents revision; can do

            # this more quickly by just appending unconditionally.

            # even more specially, if we're adding an empty text we

            # need do nothing at all.

            if lines:

                self._weave.append(('{', new_version))

                self._weave.extend(lines)

                self._weave.append(('}', None))

            return new_version

        if len(parents) == 1:

            pv = list(parents)[0]

            if sha1 == self._sha1s[pv]:

                # special case: same as the single parent

                return new_version

        ancestors = self._inclusions(parents)

        l = self._weave

        # basis a list of (origin, lineno, line)

        basis_lineno = []

        basis_lines = []

        for origin, lineno, line in self._extract(ancestors):

            basis_lineno.append(lineno)

            basis_lines.append(line)

        # another small special case: a merge, producing the same text

        # as auto-merge

        if lines == basis_lines:

            return new_version

        # add a sentinel, because we can also match against the final line

        basis_lineno.append(len(self._weave))

        # XXX: which line of the weave should we really consider

        # matches the end of the file?  the current code says it's the

        # last line of the weave?

        #print 'basis_lines:', basis_lines

        #print 'new_lines:  ', lines

        s = self._matcher(None, basis_lines, lines)

        # offset gives the number of lines that have been inserted

        # into the weave up to the current point; if the original edit instruction

        # says to change line A then we actually change (A+offset)

        offset = 0

        for tag, i1, i2, j1, j2 in s.get_opcodes():

            # i1,i2 are given in offsets within basis_lines; we need to map them

            # back to offsets within the entire weave

            #print 'raw match', tag, i1, i2, j1, j2

            if tag == 'equal':

                continue

            i1 = basis_lineno[i1]

            i2 = basis_lineno[i2]

            assert 0 <= j1 <= j2 <= len(lines)

            #print tag, i1, i2, j1, j2

            # the deletion and insertion are handled separately.

            # first delete the region.

            if i1 != i2:

                self._weave.insert(i1+offset, ('[', new_version))

                self._weave.insert(i2+offset+1, (']', new_version))

                offset += 2

            if j1 != j2:

                # there may have been a deletion spanning up to

                # i2; we want to insert after this region to make sure

                # we don't destroy ourselves

                i = i2 + offset

                self._weave[i:i] = ([('{', new_version)] 

                                    + lines[j1:j2] 

                                    + [('}', None)])

                offset += 2 + (j2 - j1)

        return new_version

    def _clone_text(self, new_version_id, old_version_id, parents):

        """See VersionedFile.clone_text."""

        old_lines = self.get_text(old_version_id)

        self.add_lines(new_version_id, parents, old_lines)

    def _inclusions(self, versions):

        """Return set of all ancestors of given version(s)."""

        if not len(versions):

            return []

        i = set(versions)

        for v in xrange(max(versions), 0, -1):

            if v in i:

                # include all its parents

                i.update(self._parents[v])

        return i

        ## except IndexError:

        ##     raise ValueError("version %d not present in weave" % v)

    def get_ancestry(self, version_ids, topo_sorted=True):

        """See VersionedFile.get_ancestry."""

        if isinstance(version_ids, basestring):

            version_ids = [version_ids]

        i = self._inclusions([self._lookup(v) for v in version_ids])

        return [self._idx_to_name(v) for v in i]

    def _check_lines(self, text):

        if not isinstance(text, list):

            raise ValueError("text should be a list, not %s" % type(text))

        for l in text:

            if not isinstance(l, basestring):

                raise ValueError("text line should be a string or unicode, not %s"

                                 % type(l))

    def _check_versions(self, indexes):

        """Check everything in the sequence of indexes is valid"""

        for i in indexes:

            try:

                self._parents[i]

            except IndexError:

                raise IndexError("invalid version number %r" % i)

    def _compatible_parents(self, my_parents, other_parents):

        """During join check that other_parents are joinable with my_parents.

        Joinable is defined as 'is a subset of' - supersets may require 

        regeneration of diffs, but subsets do not.

        """

        return len(other_parents.difference(my_parents)) == 0

    def annotate_iter(self, version_id):

        """Yield list of (version-id, line) pairs for the specified version.

        The index indicates when the line originated in the weave."""

        incls = [self._lookup(version_id)]

        for origin, lineno, text in self._extract(incls):

            yield self._idx_to_name(origin), text

    def iter_lines_added_or_present_in_versions(self, version_ids=None,

                                                pb=None):

        """See VersionedFile.iter_lines_added_or_present_in_versions()."""

        if version_ids is None:

            version_ids = self.versions()

        version_ids = set(version_ids)

        for lineno, inserted, deletes, line in self._walk_internal(version_ids):

            # if inserted not in version_ids then it was inserted before the

            # versions we care about, but because weaves cannot represent ghosts

            # properly, we do not filter down to that

            # if inserted not in version_ids: continue

            if line[-1] != '\n':

                yield line + '\n', inserted

            else:

                yield line, inserted

    def _walk_internal(self, version_ids=None):

        """Helper method for weave actions."""

        istack = []

        dset = set()

        lineno = 0         # line of weave, 0-based

        for l in self._weave:

            if l.__class__ == tuple:

                c, v = l

                isactive = None

                if c == '{':

                    istack.append(self._names[v])

                elif c == '}':

                    istack.pop()

                elif c == '[':

                    assert self._names[v] not in dset

                    dset.add(self._names[v])

                elif c == ']':

                    dset.remove(self._names[v])

                else:

                    raise WeaveFormatError('unexpected instruction %r' % v)

            else:

                assert l.__class__ in (str, unicode)

                assert istack

                yield lineno, istack[-1], frozenset(dset), l

            lineno += 1

        if istack:

            raise WeaveFormatError("unclosed insertion blocks "

                    "at end of weave: %s" % istack)

        if dset:

            raise WeaveFormatError("unclosed deletion blocks at end of weave: %s"

                                   % dset)

    def plan_merge(self, ver_a, ver_b):

        """Return pseudo-annotation indicating how the two versions merge.

        This is computed between versions a and b and their common

        base.

        Weave lines present in none of them are skipped entirely.

        """

        inc_a = set(self.get_ancestry([ver_a]))

        inc_b = set(self.get_ancestry([ver_b]))

        inc_c = inc_a & inc_b

        for lineno, insert, deleteset, line in self._walk_internal([ver_a, ver_b]):

            if deleteset & inc_c:

                # killed in parent; can't be in either a or b

                # not relevant to our work

                yield 'killed-base', line

            elif insert in inc_c:

                # was inserted in base

                killed_a = bool(deleteset & inc_a)

                killed_b = bool(deleteset & inc_b)

                if killed_a and killed_b:

                    yield 'killed-both', line

                elif killed_a:

                    yield 'killed-a', line

                elif killed_b:

                    yield 'killed-b', line

                else:

                    yield 'unchanged', line

            elif insert in inc_a:

                if deleteset & inc_a:

                    yield 'ghost-a', line

                else:

                    # new in A; not in B

                    yield 'new-a', line

            elif insert in inc_b:

                if deleteset & inc_b:

                    yield 'ghost-b', line

                else:

                    yield 'new-b', line

            else:

                # not in either revision

                yield 'irrelevant', line

        yield 'unchanged', ''           # terminator

    def _extract(self, versions):

        """Yield annotation of lines in included set.

        Yields a sequence of tuples (origin, lineno, text), where

        origin is the origin version, lineno the index in the weave,

        and text the text of the line.

        The set typically but not necessarily corresponds to a version.

        """

        for i in versions:

            if not isinstance(i, int):

                raise ValueError(i)

        included = self._inclusions(versions)

        istack = []

        iset = set()

        dset = set()

        lineno = 0         # line of weave, 0-based

        isactive = None

        result = []

        WFE = WeaveFormatError

        # wow. 

        #  449       0   4474.6820   2356.5590   bzrlib.weave:556(_extract)

        #  +285282   0   1676.8040   1676.8040   +<isinstance>

        # 1.6 seconds in 'isinstance'.

        # changing the first isinstance:

        #  449       0   2814.2660   1577.1760   bzrlib.weave:556(_extract)

        #  +140414   0    762.8050    762.8050   +<isinstance>

        # note that the inline time actually dropped (less function calls)

        # and total processing time was halved.

        # we're still spending ~1/4 of the method in isinstance though.

        # so lets hard code the acceptable string classes we expect:

        #  449       0   1202.9420    786.2930   bzrlib.weave:556(_extract)

        # +71352     0    377.5560    377.5560   +<method 'append' of 'list' 

        #                                          objects>

        # yay, down to ~1/4 the initial extract time, and our inline time

        # has shrunk again, with isinstance no longer dominating.

        # tweaking the stack inclusion test to use a set gives:

        #  449       0   1122.8030    713.0080   bzrlib.weave:556(_extract)

        # +71352     0    354.9980    354.9980   +<method 'append' of 'list' 

        #                                          objects>

        # - a 5% win, or possibly just noise. However with large istacks that

        # 'in' test could dominate, so I'm leaving this change in place -

        # when its fast enough to consider profiling big datasets we can review.

        for l in self._weave:

            if l.__class__ == tuple:

                c, v = l

                isactive = None

                if c == '{':

                    assert v not in iset

                    istack.append(v)

                    iset.add(v)

                elif c == '}':

                    iset.remove(istack.pop())

                elif c == '[':

                    if v in included:

                        assert v not in dset

                        dset.add(v)

                else:

                    assert c == ']'

                    if v in included:

                        assert v in dset

                        dset.remove(v)

            else:

                assert l.__class__ in (str, unicode)

                if isactive is None:

                    isactive = (not dset) and istack and (istack[-1] in included)

                if isactive:

                    result.append((istack[-1], lineno, l))

            lineno += 1

        if istack:

            raise WeaveFormatError("unclosed insertion blocks "

                    "at end of weave: %s" % istack)

        if dset:

            raise WeaveFormatError("unclosed deletion blocks at end of weave: %s"

                                   % dset)

        return result

    def _maybe_lookup(self, name_or_index):

        """Convert possible symbolic name to index, or pass through indexes.

        NOT FOR PUBLIC USE.

        """

        if isinstance(name_or_index, (int, long)):

            return name_or_index

        else:

            return self._lookup(name_or_index)

    def get_lines(self, version_id):

        """See VersionedFile.get_lines()."""

        int_index = self._maybe_lookup(version_id)

        result = [line for (origin, lineno, line) in self._extract([int_index])]

        expected_sha1 = self._sha1s[int_index]

        measured_sha1 = sha_strings(result)

        if measured_sha1 != expected_sha1:

            raise errors.WeaveInvalidChecksum(

                    'file %s, revision %s, expected: %s, measured %s' 

                    % (self._weave_name, version_id,

                       expected_sha1, measured_sha1))

        return result

    def get_sha1(self, version_id):

        """See VersionedFile.get_sha1()."""

        return self._sha1s[self._lookup(version_id)]

    def get_sha1s(self, version_ids):

        """See VersionedFile.get_sha1s()."""

        return [self._sha1s[self._lookup(v)] for v in version_ids]

    def num_versions(self):

        """How many versions are in this weave?"""

        l = len(self._parents)

        assert l == len(self._sha1s)

        return l

    __len__ = num_versions

    def check(self, progress_bar=None):

        # TODO evaluate performance hit of using string sets in this routine.

        # TODO: check no circular inclusions

        # TODO: create a nested progress bar

        for version in range(self.num_versions()):

            inclusions = list(self._parents[version])

            if inclusions:

                inclusions.sort()

                if inclusions[-1] >= version:

                    raise WeaveFormatError("invalid included version %d for index %d"

                                           % (inclusions[-1], version))

        # try extracting all versions; parallel extraction is used