/b-gtk/fix-viz

#!/usr/bin/python

# -*- coding: UTF-8 -*-

"""Directed graph production.

This module contains the code to produce an ordered directed graph of a

bzr branch, such as we display in the tree view at the top of the bzrk

window.

"""

__copyright__ = "Copyright © 2005 Canonical Ltd."

__author__    = "Scott James Remnant <scott@ubuntu.com>"

from bzrlib.errors import NoSuchRevision

from bzrlib.tsort import merge_sort

class DummyRevision(object):

    """Dummy bzr revision.

    Sometimes, especially in older bzr branches, a revision is referenced

    as the parent of another but not actually present in the branch's store.

    When this happens we use an instance of this class instead of the real

    Revision object (which we can't get).

    """

    def __init__(self, revid):

        self.revision_id = revid

        self.parent_ids = []

        self.committer = None

        self.message = self.revision_id

class RevisionProxy(object):

    """A revision proxy object.

    This will demand load the revision it represents when the committer or

    message attributes are accessed in order to populate them. It is 

    constructed with the revision id and parent_ids list and a repository

    object to request the revision from when needed.

    """

    def __init__(self, revid, parent_ids, repository):

        self.revision_id = revid

        self.parent_ids = parent_ids

        self._repository = repository

        self._revision = None

    def _get_attribute_getter(attr):

        def get_attribute(self):

            if self._revision is None:

                self._load()

            return getattr(self._revision, attr)

        return get_attribute

    committer = property(_get_attribute_getter('committer'))

    message = property(_get_attribute_getter('message'))

    properties = property(_get_attribute_getter('properties'))

    timestamp = property(_get_attribute_getter('timestamp'))

    timezone = property(_get_attribute_getter('timezone'))

    def _load(self):

        """Load the revision object."""

        self._revision = self._repository.get_revision(self.revision_id)

class DistanceMethod(object):

    def __init__(self, branch, start):

        self.branch = branch

        self.start = start

        self.revisions = {}

        self.children = {}

        self.children_of_id = {start: set()}

        self.parent_ids_of = {}

        self.colours = { start: 0 }

        self.last_colour = 0

        self.direct_parent_of = {}

        self.graph = {}

    def fill_caches(self):

        # FIXME: look at using repository.get_revision_graph_with_ghosts - RBC.

        graph = self.branch.repository.get_revision_graph_with_ghosts([self.start])

        for revid in graph.ghosts:

            self.cache_revision(DummyRevision(revid))

        for revid, parents in graph.get_ancestors().items():

            self.cache_revision(RevisionProxy(revid, parents, self.branch.repository))

    def cache_revision(self, revision):

        "Set the caches for a newly retrieved revision."""

        revid = revision.revision_id

        # Build a revision cache

        self.revisions[revid] = revision

        # Build a children dictionary

        for parent_id in revision.parent_ids:

            self.children_of_id.setdefault(parent_id, set()).add(revision)

        # Build a parents dictionnary, where redundant parents will be removed,

        # and that will be passed along tothe rest of program.

        if len(revision.parent_ids) != len(set(revision.parent_ids)):

            # fix the parent_ids list.

            parent_ids = []

            parent_ids_set = set()

            for parent_id in revision.parent_ids:

                if parent_id in parent_ids_set:

                    continue

                parent_ids.append(parent_id)

                parent_ids_set.add(parent_id)

            revision.parent_ids = parent_ids

        self.parent_ids_of[revision] = list(revision.parent_ids)

        self.graph[revid] = revision.parent_ids

    def make_children_map(self):

        revisions = self.revisions

        return dict((revisions[revid], c)

                    for (revid, c) in self.children_of_id.iteritems())

    def remove_redundant_parents(self, sorted_revids):

        children_of_id = self.children_of_id

        revisions = self.revisions

        parent_ids_of = self.parent_ids_of

        # Count the number of children of each revision, so we can release

        # memory for ancestry data as soon as it's not going to be needed

        # anymore.

        pending_count_of = {}

        for parent_id, children in children_of_id.iteritems():

            pending_count_of[parent_id] = len(children)

        # Build the ancestry dictionnary by examining older revisions first,

        # and remove revision parents that are ancestors of other parents of

        # the same revision.

        ancestor_ids_of = {}

        for revid in reversed(sorted_revids):

            revision = revisions[revid]

            parent_ids = parent_ids_of[revision]

            # ignore candidate parents which are an ancestor of another parent,

            # but never ignore the leftmost parent

            redundant_ids = []

            ignorable_parent_ids = parent_ids[1:] # never ignore leftmost

            for candidate_id in ignorable_parent_ids: 

                for parent_id in list(parent_ids):

                    if candidate_id in ancestor_ids_of[parent_id]:

                        redundant_ids.append(candidate_id)

                        parent_ids.remove(candidate_id)

                        children_of_candidate = children_of_id[candidate_id]

                        children_of_candidate.remove(revision)

                        break

            # save the set of ancestors of that revision

            ancestor_ids = set(parent_ids)

            for parent_id in parent_ids:

                ancestor_ids.update(ancestor_ids_of[parent_id])

            ancestor_ids_of[revid] = ancestor_ids

            # discard ancestry data for revisions whose children are already

            # done

            for parent_id in parent_ids + redundant_ids:

                pending_count = pending_count_of[parent_id] - 1

                pending_count_of[parent_id] = pending_count

                if pending_count == 0:

                    ancestor_ids_of[parent_id] = None

    def sort_revisions(self, sorted_revids, maxnum):

        revisions = self.revisions

        parent_ids_of = self.parent_ids_of

        children_of_id = self.children_of_id

        # Try to compact sequences of revisions on the same branch.

        distances = {}

        skipped_revids = []

        expected_id = sorted_revids[0]

        pending_ids = []

        while True:

            revid = sorted_revids.pop(0)

            if revid != expected_id:

                skipped_revids.append(revid)

                continue

            revision = revisions[revid]

            for child in children_of_id[revid]:

                # postpone if any child is missing

                if child.revision_id not in distances:

                    if expected_id not in pending_ids:

                        pending_ids.append(expected_id)

                    expected_id = pending_ids.pop(0)

                    skipped_revids.append(revid)

                    sorted_revids[:0] = skipped_revids

                    del skipped_revids[:]

                    break

            else:

                # all children are here, push!

                distances[revid] = len(distances)

                if maxnum is not None and len(distances) > maxnum:

                    # bail out early if a limit was specified

                    sorted_revids[:0] = skipped_revids

                    for revid in sorted_revids:

                        distances[revid] = len(distances)

                    break

                # all parents will need to be pushed as soon as possible

                for parent in parent_ids_of[revision]:

                    if parent not in pending_ids:

                        pending_ids.insert(0, parent)

                if not pending_ids:

                    break

                expected_id = pending_ids.pop(0)

                # if the next expected revid has already been skipped, requeue

                # the skipped ids, except those that would go right back to the

                # skipped list.

                if expected_id in skipped_revids:

                    pos = skipped_revids.index(expected_id)

                    sorted_revids[:0] = skipped_revids[pos:]

                    del skipped_revids[pos:]

        self.distances = distances

        return sorted(distances, key=distances.get)

    def choose_colour(self, revid):

        revision = self.revisions[revid]

        children_of_id = self.children_of_id

        parent_ids_of = self.parent_ids_of

        colours = self.colours

        # choose colour

        the_children = children_of_id[revid]

        if len(the_children) == 1:

            [child] = the_children

            if len(parent_ids_of[child]) == 1:

                # one-one relationship between parent and child, same

                # colour

                colours[revid] = colours[child.revision_id]

            else:

                self.choose_colour_one_child(revision, child)

        else:

            self.choose_colour_many_children(revision, the_children)

    def choose_colour_one_child(self, revision, child):

        revid = revision.revision_id

        direct_parent_of = self.direct_parent_of

        revisions = self.revisions

        # one child with multiple parents, the first parent with

        # the same committer gets the colour

        direct_parent = direct_parent_of.get(child)

        if direct_parent is None:

            # if it has not been found yet, find it now and remember

            for parent_id in self.parent_ids_of[child]:

                parent_revision = revisions[parent_id]

                if parent_revision.committer == child.committer:

                    # found the first parent with the same committer

                    direct_parent = parent_revision

                    direct_parent_of[child] = direct_parent

                    break

        if direct_parent == revision:

            self.colours[revid] = self.colours[child.revision_id]

        else:

            self.colours[revid] = self.last_colour = self.last_colour + 1

    def choose_colour_many_children(self, revision, the_children):

        """Colour revision revision."""

        revid = revision.revision_id

        direct_parent_of = self.direct_parent_of

        # multiple children, get the colour of the last displayed child

        # with the same committer which does not already have its colour

        # taken

        available = {}

        for child in the_children:

            if child.committer != revision.committer:

                continue

            direct_parent = direct_parent_of.get(child)

            if direct_parent == revision:

                self.colours[revid] = self.colours[child.revision_id]

                break

            # FIXME: Colouring based on whats been displayed MUST be done with 

            # knowledge of the revisions being output.

            # until the refactoring to fold graph() into this more compactly is

            # done, I've disabled this reuse. RBC 20060403

            # if direct_parent is None:

            #     available[child] = distances[child.revision_id] 

            #   .. it will be something like available[child] =  \

            #  revs[child.revision_id][0] - which is the sequence number

        else:

            if available:

                sorted_children = sorted(available, key=available.get)

                child = sorted_children[-1]

                direct_parent_of[child] = revision

                self.colours[revid] = self.colours[child.revision_id]

            else:

                # no candidate children is available, pick the next

                # colour

                self.colours[revid] = self.last_colour = self.last_colour + 1

def distances(branch, start, robust, maxnum):

    """Sort the revisions.

    Traverses the branch revision tree starting at start and produces an

    ordered list of revisions such that a revision always comes after

    any revision it is the parent of.

    Returns a tuple of (revids, revisions, colours, children)

    """

    distance = DistanceMethod(branch, start)

    distance.fill_caches()

    if robust:

        print 'robust filtering'

        distance.remove_redundant_parents(self.graph.keys())

    distance.merge_sorted = merge_sort(distance.graph, distance.start)

    children = distance.make_children_map()

    for seq, revid, merge_depth, end_of_merge in distance.merge_sorted:

        distance.choose_colour(revid)

    if maxnum is not None:

        print 'FIXME: maxnum disabled.'

    revisions = distance.revisions

    colours = distance.colours

    parent_ids_of = distance.parent_ids_of

    return (revisions, colours, children, parent_ids_of, distance.merge_sorted)

def graph(revisions, colours, merge_sorted):

    """Produce a directed graph of a bzr branch.

    For each revision it then yields a tuple of (revision, node, lines).

    If the revision is only referenced in the branch and not present in the

    store, revision will be a DummyRevision object, otherwise it is the bzr

    Revision object with the meta-data for the revision.

    Node is a tuple of (column, colour) with column being a zero-indexed

    column number of the graph that this revision represents and colour

    being a zero-indexed colour (which doesn't specify any actual colour

    in particular) to draw the node in.

    Lines is a list of tuples which represent lines you should draw away

    from the revision, if you also need to draw lines into the revision

    you should use the lines list from the previous iteration.  Each

    typle in the list is in the form (start, end, colour) with start and

    end being zero-indexed column numbers and colour as in node.

    It's up to you how to actually draw the nodes and lines (straight,

    curved, kinked, etc.) and to pick the actual colours for each index.

    """

    if not len(merge_sorted):

        return

    # split merge_sorted into a map:

    revs = {}

    # FIXME: get a hint on this from the merge_sorted data rather than

    # calculating it ourselves

    # mapping from rev_id to the sequence number of the next lowest rev

    next_lower_rev = {}

    # mapping from rev_id to next-in-branch-revid - may be None for end

    # of branch

    next_branch_revid = {}

    # the stack we are in in the sorted data for determining which 

    # next_lower_rev to set. It is a stack which has one list at each

    # depth - the ids at that depth that need the same id allocated.

    current_stack = [[]]

    for seq, revid, indent, end_merge in merge_sorted:

        revs[revid] = (seq, indent, end_merge)

        if indent == len(current_stack):

            # new merge group starts

            current_stack.append([revid])

        elif indent == len(current_stack) - 1:

            # part of the current merge group

            current_stack[-1].append(revid)

        else:

            # end of a merge group

            while current_stack[-1]:

                stack_rev_id = current_stack[-1].pop()

                # record the next lower rev for this rev:

                next_lower_rev[stack_rev_id] = seq

                # if this followed a non-end-merge rev in this group note that

                if len(current_stack[-1]):

                    if not revs[current_stack[-1][-1]][2]:

                        next_branch_revid[current_stack[-1][-1]] = stack_rev_id

            current_stack.pop()

            # append to the now-current merge group

            current_stack[-1].append(revid)

    # assign a value to all the depth 0 revisions

    while current_stack[-1]:

        stack_rev_id = current_stack[-1].pop()

        # record the next lower rev for this rev:

        next_lower_rev[stack_rev_id] = len(merge_sorted)

        # if this followed a non-end-merge rev in this group note that

        if len(current_stack[-1]):

            if not revs[current_stack[-1][-1]][2]:

                next_branch_revid[current_stack[-1][-1]] = stack_rev_id

    # a list of the current revisions we are drawing lines TO indicating

    # the sequence of their lines on the screen.

    # i.e. [A, B, C] means that the line to A, to B, and to C are in

    # (respectively), 0, 1, 2 on the screen.

    hanging = [merge_sorted[0][1]]

    for seq, revid, indent, end_merge in merge_sorted:

        # a list of the lines to draw: their position in the

        # previous row, their position in this row, and the colour

        # (which is the colour they are routing to).

        lines = []

        new_hanging = []

        for h_idx, hang in enumerate(hanging):

            # one of these will be the current lines node:

            # we are drawing a line. h_idx 

            if hang == revid:

                # we have found the current lines node

                node = (h_idx, colours[revid])

                # note that we might have done the main parent

                drawn_parents = set()

                def draw_line(from_idx, to_idx, revision_id):

                    try:

                        n_idx = new_hanging.index(revision_id)

                    except ValueError:

                        # force this to be vertical at the place this rev was

                        # drawn.

                        new_hanging.insert(to_idx, revision_id)

                        n_idx = to_idx

                    lines.append((from_idx, n_idx, colours[revision_id]))

                # we want to draw a line to the next commit on 'this' branch

                if not end_merge:

                    # drop this line first.

                    parent_id = next_branch_revid[revid]

                    draw_line(h_idx, h_idx, parent_id)

                    # we have drawn this parent

                    drawn_parents.add(parent_id)

                else:

                    # this is the last revision in a 'merge', show where it came from

                    if len(revisions[revid].parent_ids) > 1:

                        # having > 1

                        # parents means this commit was a merge, and being

                        # the end point of a merge group means that all

                        # the parent revisions were merged into branches

                        # to the left of this before this was committed

                        # - so we want to show this as a new branch from

                        # those revisions.

                        # to do this, we show the parent with the lowest

                        # sequence number, which is the one that this

                        # branch 'spawned from', and no others.

                        # If this sounds like a problem, remember that:

                        # if the parent was not already in our mainline

                        # it would show up as a merge into this making

                        # this not the end of a merge-line.

                        lowest = len(merge_sorted)

                        for parent_id in revisions[revid].parent_ids:

                            if revs[parent_id][0] < lowest:

                                lowest = revs[parent_id][0]

                        assert lowest != len(merge_sorted)

                        draw_line(h_idx, len(new_hanging), merge_sorted[lowest][1])

                        drawn_parents.add(merge_sorted[lowest][1])

                    elif len(revisions[revid].parent_ids) == 1:

                        # only one parent, must show this link to be useful.

                        parent_id = revisions[revid].parent_ids[0]

                        draw_line(h_idx, len(new_hanging), parent_id)

                        drawn_parents.add(parent_id)

                # what do we want to draw lines to from here:

                # each parent IF its relevant.

                #

                # Now we need to hang its parents, we put them at the point

                # the old column was so anything to the right of this has

                # to move outwards to make room.  We also try and collapse

                # hangs to keep the graph small.

                # RBC: we do not draw lines to parents that were already merged

                # unless its the last revision in a merge group.

                for parent_id in revisions[revid].parent_ids:

                    if parent_id in drawn_parents:

                        continue

                    parent_seq = revs[parent_id][0]

                    parent_depth = revs[parent_id][1]

                    if parent_depth == indent + 1:

                        # the parent was a merge into this branch

                        # determine if it was already merged into the mainline

                        # via a different merge:

                        # if all revisions between us and parent_seq have a 

                        # indent greater than there are no revisions with a lower indent than

                        # us.

                        # we do not use 'parent_depth < indent' because that would allow

                        # un-uniqueified merges to show up, and merge_sorted should take

                        # care of that for us (but does not trim the values)

                        if parent_seq < next_lower_rev[revid]:

                            draw_line(h_idx, len(new_hanging), parent_id)

                    elif parent_depth == indent and parent_seq == seq + 1:

                        # part of this branch

                        draw_line(h_idx, len(new_hanging), parent_id)

            else:

                # draw a line from the previous position of this line to the 

                # new position.

                # h_idx is the old position.

                # new_indent is the new position. 

                draw_line(h_idx, len(new_hanging), hang)

        # we've calculated the row, assign new_hanging to hanging to setup for

        # the next row

        hanging = new_hanging

        yield (revisions[revid], node, lines)

def same_branch(a, b):

    """Return whether we think revisions a and b are on the same branch."""

    if len(a.parent_ids) == 1:

        # Defacto same branch if only parent

        return True

    elif a.committer == b.committer:

        # Same committer so may as well be

        return True

    else:

        return False