/brz/remove-bazaar

# Copyright (C) 2008, 2009 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., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA

"""Tests for pack repositories.

These tests are repeated for all pack-based repository formats.

"""

from cStringIO import StringIO

from stat import S_ISDIR

from bzrlib.btree_index import BTreeGraphIndex

from bzrlib.index import GraphIndex

from bzrlib import (

    bzrdir,

    errors,

    inventory,

    osutils,

    progress,

    repository,

    revision as _mod_revision,

    symbol_versioning,

    tests,

    ui,

    upgrade,

    workingtree,

    )

from bzrlib.repofmt import (

    pack_repo,

    groupcompress_repo,

    )

from bzrlib.repofmt.groupcompress_repo import RepositoryFormat2a

from bzrlib.smart import (

    client,

    server,

    )

from bzrlib.tests import (

    TestCase,

    TestCaseWithTransport,

    TestNotApplicable,

    TestSkipped,

    )

from bzrlib.transport import (

    fakenfs,

    memory,

    get_transport,

    )

from bzrlib.tests.per_repository import TestCaseWithRepository

class TestPackRepository(TestCaseWithTransport):

    """Tests to be repeated across all pack-based formats.

    The following are populated from the test scenario:

    :ivar format_name: Registered name fo the format to test.

    :ivar format_string: On-disk format marker.

    :ivar format_supports_external_lookups: Boolean.

    """

    def get_format(self):

        return bzrdir.format_registry.make_bzrdir(self.format_name)

    def test_attribute__fetch_order(self):

        """Packs do not need ordered data retrieval."""

        format = self.get_format()

        repo = self.make_repository('.', format=format)

        self.assertEqual('unordered', repo._format._fetch_order)

    def test_attribute__fetch_uses_deltas(self):

        """Packs reuse deltas."""

        format = self.get_format()

        repo = self.make_repository('.', format=format)

        if isinstance(format.repository_format, RepositoryFormat2a):

            # TODO: This is currently a workaround. CHK format repositories

            #       ignore the 'deltas' flag, but during conversions, we can't

            #       do unordered delta fetches. Remove this clause once we

            #       improve the inter-format fetching.

            self.assertEqual(False, repo._format._fetch_uses_deltas)

        else:

            self.assertEqual(True, repo._format._fetch_uses_deltas)

    def test_disk_layout(self):

        format = self.get_format()

        repo = self.make_repository('.', format=format)

        # in case of side effects of locking.

        repo.lock_write()

        repo.unlock()

        t = repo.bzrdir.get_repository_transport(None)

        self.check_format(t)

        # XXX: no locks left when unlocked at the moment

        # self.assertEqualDiff('', t.get('lock').read())

        self.check_databases(t)

    def check_format(self, t):

        self.assertEqualDiff(

            self.format_string, # from scenario

            t.get('format').read())

    def assertHasNoKndx(self, t, knit_name):

        """Assert that knit_name has no index on t."""

        self.assertFalse(t.has(knit_name + '.kndx'))

    def assertHasNoKnit(self, t, knit_name):

        """Assert that knit_name exists on t."""

        # no default content

        self.assertFalse(t.has(knit_name + '.knit'))

    def check_databases(self, t):

        """check knit content for a repository."""

        # check conversion worked

        self.assertHasNoKndx(t, 'inventory')

        self.assertHasNoKnit(t, 'inventory')

        self.assertHasNoKndx(t, 'revisions')

        self.assertHasNoKnit(t, 'revisions')

        self.assertHasNoKndx(t, 'signatures')

        self.assertHasNoKnit(t, 'signatures')

        self.assertFalse(t.has('knits'))

        # revision-indexes file-container directory

        self.assertEqual([],

            list(self.index_class(t, 'pack-names', None).iter_all_entries()))

        self.assertTrue(S_ISDIR(t.stat('packs').st_mode))

        self.assertTrue(S_ISDIR(t.stat('upload').st_mode))

        self.assertTrue(S_ISDIR(t.stat('indices').st_mode))

        self.assertTrue(S_ISDIR(t.stat('obsolete_packs').st_mode))

    def test_shared_disk_layout(self):

        format = self.get_format()

        repo = self.make_repository('.', shared=True, format=format)

        # we want:

        t = repo.bzrdir.get_repository_transport(None)

        self.check_format(t)

        # XXX: no locks left when unlocked at the moment

        # self.assertEqualDiff('', t.get('lock').read())

        # We should have a 'shared-storage' marker file.

        self.assertEqualDiff('', t.get('shared-storage').read())

        self.check_databases(t)

    def test_shared_no_tree_disk_layout(self):

        format = self.get_format()

        repo = self.make_repository('.', shared=True, format=format)

        repo.set_make_working_trees(False)

        # we want:

        t = repo.bzrdir.get_repository_transport(None)

        self.check_format(t)

        # XXX: no locks left when unlocked at the moment

        # self.assertEqualDiff('', t.get('lock').read())

        # We should have a 'shared-storage' marker file.

        self.assertEqualDiff('', t.get('shared-storage').read())

        # We should have a marker for the no-working-trees flag.

        self.assertEqualDiff('', t.get('no-working-trees').read())

        # The marker should go when we toggle the setting.

        repo.set_make_working_trees(True)

        self.assertFalse(t.has('no-working-trees'))

        self.check_databases(t)

    def test_adding_revision_creates_pack_indices(self):

        format = self.get_format()

        tree = self.make_branch_and_tree('.', format=format)

        trans = tree.branch.repository.bzrdir.get_repository_transport(None)

        self.assertEqual([],

            list(self.index_class(trans, 'pack-names', None).iter_all_entries()))

        tree.commit('foobarbaz')

        index = self.index_class(trans, 'pack-names', None)

        index_nodes = list(index.iter_all_entries())

        self.assertEqual(1, len(index_nodes))

        node = index_nodes[0]

        name = node[1][0]

        # the pack sizes should be listed in the index

        pack_value = node[2]

        sizes = [int(digits) for digits in pack_value.split(' ')]

        for size, suffix in zip(sizes, ['.rix', '.iix', '.tix', '.six']):

            stat = trans.stat('indices/%s%s' % (name, suffix))

            self.assertEqual(size, stat.st_size)

    def test_pulling_nothing_leads_to_no_new_names(self):

        format = self.get_format()

        tree1 = self.make_branch_and_tree('1', format=format)

        tree2 = self.make_branch_and_tree('2', format=format)

        tree1.branch.repository.fetch(tree2.branch.repository)

        trans = tree1.branch.repository.bzrdir.get_repository_transport(None)

        self.assertEqual([],

            list(self.index_class(trans, 'pack-names', None).iter_all_entries()))

    def test_commit_across_pack_shape_boundary_autopacks(self):

        format = self.get_format()

        tree = self.make_branch_and_tree('.', format=format)

        trans = tree.branch.repository.bzrdir.get_repository_transport(None)

        # This test could be a little cheaper by replacing the packs

        # attribute on the repository to allow a different pack distribution

        # and max packs policy - so we are checking the policy is honoured

        # in the test. But for now 11 commits is not a big deal in a single

        # test.

        for x in range(9):

            tree.commit('commit %s' % x)

        # there should be 9 packs:

        index = self.index_class(trans, 'pack-names', None)

        self.assertEqual(9, len(list(index.iter_all_entries())))

        # insert some files in obsolete_packs which should be removed by pack.

        trans.put_bytes('obsolete_packs/foo', '123')

        trans.put_bytes('obsolete_packs/bar', '321')

        # committing one more should coalesce to 1 of 10.

        tree.commit('commit triggering pack')

        index = self.index_class(trans, 'pack-names', None)

        self.assertEqual(1, len(list(index.iter_all_entries())))

        # packing should not damage data

        tree = tree.bzrdir.open_workingtree()

        check_result = tree.branch.repository.check(

            [tree.branch.last_revision()])

        nb_files = 5 # .pack, .rix, .iix, .tix, .six

        if tree.branch.repository._format.supports_chks:

            nb_files += 1 # .cix

        # We should have 10 x nb_files files in the obsolete_packs directory.

        obsolete_files = list(trans.list_dir('obsolete_packs'))

        self.assertFalse('foo' in obsolete_files)

        self.assertFalse('bar' in obsolete_files)

        self.assertEqual(10 * nb_files, len(obsolete_files))

        # XXX: Todo check packs obsoleted correctly - old packs and indices

        # in the obsolete_packs directory.

        large_pack_name = list(index.iter_all_entries())[0][1][0]

        # finally, committing again should not touch the large pack.

        tree.commit('commit not triggering pack')

        index = self.index_class(trans, 'pack-names', None)

        self.assertEqual(2, len(list(index.iter_all_entries())))

        pack_names = [node[1][0] for node in index.iter_all_entries()]

        self.assertTrue(large_pack_name in pack_names)

    def test_commit_write_group_returns_new_pack_names(self):

        # This test doesn't need real disk.

        self.vfs_transport_factory = tests.MemoryServer

        format = self.get_format()

        repo = self.make_repository('foo', format=format)

        repo.lock_write()

        try:

            # All current pack repository styles autopack at 10 revisions; and

            # autopack as well as regular commit write group needs to return

            # the new pack name. Looping is a little ugly, but we don't have a

            # clean way to test both the autopack logic and the normal code

            # path without doing this loop.

            for pos in range(10):

                revid = str(pos)

                repo.start_write_group()

                try:

                    inv = inventory.Inventory(revision_id=revid)

                    inv.root.revision = revid

                    repo.texts.add_lines((inv.root.file_id, revid), [], [])

                    rev = _mod_revision.Revision(timestamp=0, timezone=None,

                        committer="Foo Bar <foo@example.com>", message="Message",

                        revision_id=revid)

                    rev.parent_ids = ()

                    repo.add_revision(revid, rev, inv=inv)

                except:

                    repo.abort_write_group()

                    raise

                else:

                    old_names = repo._pack_collection._names.keys()

                    result = repo.commit_write_group()

                    cur_names = repo._pack_collection._names.keys()

                    new_names = list(set(cur_names) - set(old_names))

                    self.assertEqual(new_names, result)

        finally:

            repo.unlock()

    def test_fail_obsolete_deletion(self):

        # failing to delete obsolete packs is not fatal

        format = self.get_format()

        server = fakenfs.FakeNFSServer()

        self.start_server(server)

        transport = get_transport(server.get_url())

        bzrdir = self.get_format().initialize_on_transport(transport)

        repo = bzrdir.create_repository()

        repo_transport = bzrdir.get_repository_transport(None)

        self.assertTrue(repo_transport.has('obsolete_packs'))

        # these files are in use by another client and typically can't be deleted

        repo_transport.put_bytes('obsolete_packs/.nfsblahblah', 'contents')

        repo._pack_collection._clear_obsolete_packs()

        self.assertTrue(repo_transport.has('obsolete_packs/.nfsblahblah'))

    def test_pack_after_two_commits_packs_everything(self):

        format = self.get_format()

        tree = self.make_branch_and_tree('.', format=format)

        trans = tree.branch.repository.bzrdir.get_repository_transport(None)

        tree.commit('start')

        tree.commit('more work')

        tree.branch.repository.pack()

        # there should be 1 pack:

        index = self.index_class(trans, 'pack-names', None)

        self.assertEqual(1, len(list(index.iter_all_entries())))

        self.assertEqual(2, len(tree.branch.repository.all_revision_ids()))

    def test_pack_preserves_all_inventories(self):

        # This is related to bug:

        #   https://bugs.launchpad.net/bzr/+bug/412198

        # Stacked repositories need to keep the inventory for parents, even

        # after a pack operation. However, it is harder to test that, then just

        # test that all inventory texts are preserved.

        format = self.get_format()

        builder = self.make_branch_builder('source', format=format)

        builder.start_series()

        builder.build_snapshot('A-id', None, [

            ('add', ('', 'root-id', 'directory', None))])

        builder.build_snapshot('B-id', None, [

            ('add', ('file', 'file-id', 'file', 'B content\n'))])

        builder.build_snapshot('C-id', None, [

            ('modify', ('file-id', 'C content\n'))])

        builder.finish_series()

        b = builder.get_branch()

        b.lock_read()

        self.addCleanup(b.unlock)

        repo = self.make_repository('repo', shared=True, format=format)

        repo.lock_write()

        self.addCleanup(repo.unlock)

        repo.fetch(b.repository, revision_id='B-id')

        inv = b.repository.iter_inventories(['C-id']).next()

        repo.start_write_group()

        repo.add_inventory('C-id', inv, ['B-id'])

        repo.commit_write_group()

        self.assertEqual([('A-id',), ('B-id',), ('C-id',)],

                         sorted(repo.inventories.keys()))

        repo.pack()

        self.assertEqual([('A-id',), ('B-id',), ('C-id',)],

                         sorted(repo.inventories.keys()))

        # Content should be preserved as well

        self.assertEqual(inv, repo.iter_inventories(['C-id']).next())

    def test_pack_layout(self):

        # Test that the ordering of revisions in pack repositories is

        # tip->ancestor

        format = self.get_format()

        tree = self.make_branch_and_tree('.', format=format)

        trans = tree.branch.repository.bzrdir.get_repository_transport(None)

        tree.commit('start', rev_id='1')

        tree.commit('more work', rev_id='2')

        tree.branch.repository.pack()

        tree.lock_read()

        self.addCleanup(tree.unlock)

        pack = tree.branch.repository._pack_collection.get_pack_by_name(

            tree.branch.repository._pack_collection.names()[0])

        # revision access tends to be tip->ancestor, so ordering that way on

        # disk is a good idea.

        for _1, key, val, refs in pack.revision_index.iter_all_entries():

            if type(format.repository_format) is RepositoryFormat2a:

                # group_start, group_len, internal_start, internal_len

                pos = map(int, val.split())

            else:

                # eol_flag, start, len

                pos = int(val[1:].split()[0])

            if key == ('1',):

                pos_1 = pos

            else:

                pos_2 = pos

        self.assertTrue(pos_2 < pos_1, 'rev 1 came before rev 2 %s > %s'

                                       % (pos_1, pos_2))

    def test_pack_repositories_support_multiple_write_locks(self):

        format = self.get_format()

        self.make_repository('.', shared=True, format=format)

        r1 = repository.Repository.open('.')

        r2 = repository.Repository.open('.')

        r1.lock_write()

        self.addCleanup(r1.unlock)

        r2.lock_write()

        r2.unlock()

    def _add_text(self, repo, fileid):

        """Add a text to the repository within a write group."""

        repo.texts.add_lines((fileid, 'samplerev+'+fileid), [],

            ['smaplerev+'+fileid])

    def test_concurrent_writers_merge_new_packs(self):

        format = self.get_format()

        self.make_repository('.', shared=True, format=format)

        r1 = repository.Repository.open('.')

        r2 = repository.Repository.open('.')

        r1.lock_write()

        try:

            # access enough data to load the names list

            list(r1.all_revision_ids())

            r2.lock_write()

            try:

                # access enough data to load the names list

                list(r2.all_revision_ids())

                r1.start_write_group()

                try:

                    r2.start_write_group()

                    try:

                        self._add_text(r1, 'fileidr1')

                        self._add_text(r2, 'fileidr2')

                    except:

                        r2.abort_write_group()

                        raise

                except:

                    r1.abort_write_group()

                    raise

                # both r1 and r2 have open write groups with data in them

                # created while the other's write group was open.

                # Commit both which requires a merge to the pack-names.

                try:

                    r1.commit_write_group()

                except:

                    r1.abort_write_group()

                    r2.abort_write_group()

                    raise

                r2.commit_write_group()

                # tell r1 to reload from disk

                r1._pack_collection.reset()

                # Now both repositories should know about both names

                r1._pack_collection.ensure_loaded()

                r2._pack_collection.ensure_loaded()

                self.assertEqual(r1._pack_collection.names(), r2._pack_collection.names())

                self.assertEqual(2, len(r1._pack_collection.names()))

            finally:

                r2.unlock()

        finally:

            r1.unlock()

    def test_concurrent_writer_second_preserves_dropping_a_pack(self):

        format = self.get_format()

        self.make_repository('.', shared=True, format=format)

        r1 = repository.Repository.open('.')

        r2 = repository.Repository.open('.')

        # add a pack to drop

        r1.lock_write()

        try:

            r1.start_write_group()

            try:

                self._add_text(r1, 'fileidr1')

            except:

                r1.abort_write_group()

                raise

            else:

                r1.commit_write_group()

            r1._pack_collection.ensure_loaded()

            name_to_drop = r1._pack_collection.all_packs()[0].name

        finally:

            r1.unlock()

        r1.lock_write()

        try:

            # access enough data to load the names list

            list(r1.all_revision_ids())

            r2.lock_write()

            try:

                # access enough data to load the names list

                list(r2.all_revision_ids())

                r1._pack_collection.ensure_loaded()

                try:

                    r2.start_write_group()

                    try:

                        # in r1, drop the pack

                        r1._pack_collection._remove_pack_from_memory(

                            r1._pack_collection.get_pack_by_name(name_to_drop))

                        # in r2, add a pack

                        self._add_text(r2, 'fileidr2')

                    except:

                        r2.abort_write_group()

                        raise

                except:

                    r1._pack_collection.reset()

                    raise

                # r1 has a changed names list, and r2 an open write groups with

                # changes.

                # save r1, and then commit the r2 write group, which requires a

                # merge to the pack-names, which should not reinstate

                # name_to_drop

                try:

                    r1._pack_collection._save_pack_names()

                    r1._pack_collection.reset()

                except:

                    r2.abort_write_group()

                    raise

                try:

                    r2.commit_write_group()

                except:

                    r2.abort_write_group()

                    raise

                # Now both repositories should now about just one name.

                r1._pack_collection.ensure_loaded()

                r2._pack_collection.ensure_loaded()

                self.assertEqual(r1._pack_collection.names(), r2._pack_collection.names())

                self.assertEqual(1, len(r1._pack_collection.names()))

                self.assertFalse(name_to_drop in r1._pack_collection.names())

            finally:

                r2.unlock()

        finally:

            r1.unlock()

    def test_concurrent_pack_triggers_reload(self):

        # create 2 packs, which we will then collapse

        tree = self.make_branch_and_tree('tree')

        tree.lock_write()

        try:

            rev1 = tree.commit('one')

            rev2 = tree.commit('two')

            r2 = repository.Repository.open('tree')

            r2.lock_read()

            try:

                # Now r2 has read the pack-names file, but will need to reload

                # it after r1 has repacked

                tree.branch.repository.pack()

                self.assertEqual({rev2:(rev1,)}, r2.get_parent_map([rev2]))

            finally:

                r2.unlock()

        finally:

            tree.unlock()

    def test_concurrent_pack_during_get_record_reloads(self):

        tree = self.make_branch_and_tree('tree')

        tree.lock_write()

        try:

            rev1 = tree.commit('one')

            rev2 = tree.commit('two')

            keys = [(rev1,), (rev2,)]

            r2 = repository.Repository.open('tree')

            r2.lock_read()

            try:

                # At this point, we will start grabbing a record stream, and

                # trigger a repack mid-way

                packed = False

                result = {}

                record_stream = r2.revisions.get_record_stream(keys,

                                    'unordered', False)

                for record in record_stream:

                    result[record.key] = record

                    if not packed:

                        tree.branch.repository.pack()

                        packed = True

                # The first record will be found in the original location, but

                # after the pack, we have to reload to find the next record

                self.assertEqual(sorted(keys), sorted(result.keys()))

            finally:

                r2.unlock()

        finally:

            tree.unlock()

    def test_concurrent_pack_during_autopack(self):

        tree = self.make_branch_and_tree('tree')

        tree.lock_write()

        try:

            for i in xrange(9):

                tree.commit('rev %d' % (i,))

            r2 = repository.Repository.open('tree')

            r2.lock_write()

            try:

                # Monkey patch so that pack occurs while the other repo is

                # autopacking. This is slightly bad, but all current pack

                # repository implementations have a _pack_collection, and we

                # test that it gets triggered. So if a future format changes

                # things, the test will fail rather than succeed accidentally.

                autopack_count = [0]

                r1 = tree.branch.repository

                orig = r1._pack_collection.pack_distribution

                def trigger_during_auto(*args, **kwargs):

                    ret = orig(*args, **kwargs)

                    if not autopack_count[0]:

                        r2.pack()

                    autopack_count[0] += 1

                    return ret

                r1._pack_collection.pack_distribution = trigger_during_auto

                tree.commit('autopack-rev')

                # This triggers 2 autopacks. The first one causes r2.pack() to

                # fire, but r2 doesn't see the new pack file yet. The

                # autopack restarts and sees there are 2 files and there

                # should be only 1 for 10 commits. So it goes ahead and

                # finishes autopacking.

                self.assertEqual([2], autopack_count)

            finally:

                r2.unlock()

        finally:

            tree.unlock()

    def test_lock_write_does_not_physically_lock(self):

        repo = self.make_repository('.', format=self.get_format())

        repo.lock_write()

        self.addCleanup(repo.unlock)

        self.assertFalse(repo.get_physical_lock_status())

    def prepare_for_break_lock(self):

        # Setup the global ui factory state so that a break-lock method call

        # will find usable input in the input stream.

        ui.ui_factory = ui.CannedInputUIFactory([True])

    def test_break_lock_breaks_physical_lock(self):

        repo = self.make_repository('.', format=self.get_format())

        repo._pack_collection.lock_names()

        repo.control_files.leave_in_place()

        repo.unlock()

        repo2 = repository.Repository.open('.')

        self.assertTrue(repo.get_physical_lock_status())

        self.prepare_for_break_lock()

        repo2.break_lock()

        self.assertFalse(repo.get_physical_lock_status())

    def test_broken_physical_locks_error_on__unlock_names_lock(self):

        repo = self.make_repository('.', format=self.get_format())

        repo._pack_collection.lock_names()

        self.assertTrue(repo.get_physical_lock_status())

        repo2 = repository.Repository.open('.')

        self.prepare_for_break_lock()

        repo2.break_lock()

        self.assertRaises(errors.LockBroken, repo._pack_collection._unlock_names)

    def test_fetch_without_find_ghosts_ignores_ghosts(self):

        # we want two repositories at this point:

        # one with a revision that is a ghost in the other

        # repository.

        # 'ghost' is present in has_ghost, 'ghost' is absent in 'missing_ghost'.

        # 'references' is present in both repositories, and 'tip' is present

        # just in has_ghost.

        # has_ghost       missing_ghost

        #------------------------------

        # 'ghost'             -

        # 'references'    'references'

        # 'tip'               -

        # In this test we fetch 'tip' which should not fetch 'ghost'

        has_ghost = self.make_repository('has_ghost', format=self.get_format())

        missing_ghost = self.make_repository('missing_ghost',

            format=self.get_format())

        def add_commit(repo, revision_id, parent_ids):

            repo.lock_write()

            repo.start_write_group()

            inv = inventory.Inventory(revision_id=revision_id)

            inv.root.revision = revision_id

            root_id = inv.root.file_id

            sha1 = repo.add_inventory(revision_id, inv, [])

            repo.texts.add_lines((root_id, revision_id), [], [])

            rev = _mod_revision.Revision(timestamp=0,

                                         timezone=None,

                                         committer="Foo Bar <foo@example.com>",

                                         message="Message",

                                         inventory_sha1=sha1,

                                         revision_id=revision_id)

            rev.parent_ids = parent_ids

            repo.add_revision(revision_id, rev)

            repo.commit_write_group()

            repo.unlock()

        add_commit(has_ghost, 'ghost', [])

        add_commit(has_ghost, 'references', ['ghost'])

        add_commit(missing_ghost, 'references', ['ghost'])

        add_commit(has_ghost, 'tip', ['references'])

        missing_ghost.fetch(has_ghost, 'tip')

        # missing ghost now has tip and not ghost.

        rev = missing_ghost.get_revision('tip')

        inv = missing_ghost.get_inventory('tip')

        self.assertRaises(errors.NoSuchRevision,

            missing_ghost.get_revision, 'ghost')

        self.assertRaises(errors.NoSuchRevision,

            missing_ghost.get_inventory, 'ghost')

    def make_write_ready_repo(self):

        format = self.get_format()

        if isinstance(format.repository_format, RepositoryFormat2a):