def test_publish_recursive(pristine_origin, origin_path, src_path, dst_path,
sub1_pub, sub2_pub):
# we will be publishing back to origin, so to not alter testrepo
# we will first clone it
origin = install(origin_path, source=pristine_origin, recursive=True)
# prepare src
source = install(src_path, source=origin.path, recursive=True)
# we will be trying to push into this later on, need to give permissions...
origin_sub2 = Dataset(opj(origin_path, '2'))
origin_sub2.config.set('receive.denyCurrentBranch',
'updateInstead',
where='local')
## TODO this manual fixup is needed due to gh-1548 -- needs proper solution
#os.remove(opj(origin_sub2.path, '.git'))
#os.rename(opj(origin_path, '.git', 'modules', '2'), opj(origin_sub2.path, '.git'))
# create plain git at target:
target = GitRepo(dst_path, create=True)
target.checkout("TMP", ["-b"])
source.repo.add_remote("target", dst_path)
# subdatasets have no remote yet, so recursive publishing should fail:
res = publish(dataset=source,
to="target",
recursive=True,
on_failure='ignore')
assert_result_count(res, 3)
assert_result_count(res, 1, status='ok', type='dataset', path=source.path)
assert_result_count(res,
2,
status='error',
message=("Unknown target sibling '%s' for publication",
'target'))
# now, set up targets for the submodules:
sub1_target = GitRepo(sub1_pub, create=True)
sub1_target.checkout("TMP", ["-b"])
sub2_target = AnnexRepo(sub2_pub, create=True)
# we will be testing presence of the file content, so let's make it progress
sub2_target.config.set('receive.denyCurrentBranch',
'updateInstead',
where='local')
sub1 = GitRepo(opj(src_path, 'subm 1'), create=False)
sub2 = GitRepo(opj(src_path, '2'), create=False)
sub1.add_remote("target", sub1_pub)
sub2.add_remote("target", sub2_pub)
# publish recursively
with swallow_logs(new_level=logging.DEBUG) as cml:
res = publish(dataset=source, to="target", recursive=True)
assert_not_in('forced update', cml.out,
"we probably haven't merged git-annex before pushing")
# testing result list
# base dataset was already published above, notneeded again
assert_status(('ok', 'notneeded'), res) # nothing failed
assert_result_count(res, 3, type='dataset')
eq_({r['path'] for r in res}, {src_path, sub1.path, sub2.path})
eq_(list(target.get_branch_commits_(DEFAULT_BRANCH)),
list(source.repo.get_branch_commits_(DEFAULT_BRANCH)))
assert_git_annex_branch_published(source.repo, target)
eq_(list(sub1_target.get_branch_commits_(DEFAULT_BRANCH)),
list(sub1.get_branch_commits_(DEFAULT_BRANCH)))
assert_git_annex_branch_published(sub1, sub1_target)
eq_(list(sub2_target.get_branch_commits_(DEFAULT_BRANCH)),
list(sub2.get_branch_commits_(DEFAULT_BRANCH)))
assert_git_annex_branch_published(sub2, sub2_target)
# we are tracking origin but origin has different git-annex, since we
# cloned from it, so it is not aware of our git-annex
neq_(list(origin.repo.get_branch_commits_("git-annex")),
list(source.repo.get_branch_commits_("git-annex")))
# So if we first publish to it recursively, we would update
# all sub-datasets since git-annex branch would need to be pushed
res_ = publish(dataset=source, recursive=True)
assert_result_count(res_, 1, status='ok', path=source.path)
assert_result_count(res_, 1, status='ok', path=sub1.path)
assert_result_count(res_, 1, status='ok', path=sub2.path)
# and now should carry the same state for git-annex
assert_git_annex_branch_published(source.repo, origin.repo)
# test for publishing with --since. By default since no changes, nothing pushed
res_ = publish(dataset=source, recursive=True)
assert_result_count(res_, 3, status='notneeded', type='dataset')
# still nothing gets pushed, because origin is up to date
res_ = publish(dataset=source, recursive=True, since='HEAD^')
assert_result_count(res_, 3, status='notneeded', type='dataset')
# and we should not fail if we run it from within the dataset
with chpwd(source.path):
res_ = publish(recursive=True, since='HEAD^')
assert_result_count(res_, 3, status='notneeded', type='dataset')
# Let's now update one subm
with open(opj(sub2.path, "file.txt"), 'w') as f:
f.write('')
# add to subdataset, does not alter super dataset!
# MIH: use `to_git` because original test author used
# and explicit `GitRepo.add` -- keeping this for now
Dataset(sub2.path).save('file.txt', to_git=True)
# Let's now update one subm
create_tree(sub2.path, {'file.dat': 'content'})
# add to subdataset, without reflecting the change in its super(s)
Dataset(sub2.path).save('file.dat')
# note: will publish to origin here since that is what it tracks
res_ = publish(dataset=source, recursive=True, on_failure='ignore')
## only updates published, i.e. just the subdataset, super wasn't altered
## nothing copied!
assert_status(('ok', 'notneeded'), res_)
assert_result_count(res_, 1, status='ok', path=sub2.path, type='dataset')
assert_result_count(res_, 0, path=opj(sub2.path, 'file.dat'), type='file')
# since published to origin -- destination should not get that file
nok_(lexists(opj(sub2_target.path, 'file.dat')))
res_ = publish(dataset=source, to='target', recursive=True)
assert_status(('ok', 'notneeded'), res_)
assert_result_count(res_, 1, status='ok', path=sub2.path, type='dataset')
assert_result_count(res_, 0, path=opj(sub2.path, 'file.dat'), type='file')
# Note: with updateInstead only in target2 and not saving change in
# super-dataset we would have made remote dataset, if we had entire
# hierarchy, to be somewhat inconsistent.
# But here, since target datasets are independent -- it is ok
# and the file itself was transferred
ok_(lexists(opj(sub2_target.path, 'file.dat')))
nok_(sub2_target.file_has_content('file.dat'))
## but now we can redo publish recursively, with explicitly requested data transfer
res_ = publish(dataset=source,
to='target',
recursive=True,
transfer_data='all')
ok_(sub2_target.file_has_content('file.dat'))
assert_result_count(res_, 1, status='ok', path=opj(sub2.path, 'file.dat'))
# Let's save those present changes and publish while implying "since last
# merge point"
source.save(message="Changes in subm2")
# and test if it could deduce the remote/branch to push to
source.config.set('branch.{}.remote'.format(DEFAULT_BRANCH),
'target',
where='local')
with chpwd(source.path):
res_ = publish(since='^', recursive=True)
# TODO: somehow test that there were no even attempt to diff within "subm 1"
# since if `--since=''` worked correctly, nothing has changed there and it
# should have not been even touched
assert_status(('ok', 'notneeded'), res_)
assert_result_count(res_, 1, status='ok', path=source.path, type='dataset')
# Don't fail when a string is passed as `dataset` and since="".
assert_status("notneeded", publish(since='^', dataset=source.path))
def test_publish_with_data(origin, src_path, dst_path, sub1_pub, sub2_pub,
dst_clone_path):
# prepare src
source = install(src_path, source=origin, recursive=True)
source.repo.get('test-annex.dat')
# create plain git at target:
target = AnnexRepo(dst_path, create=True)
target.checkout("TMP", ["-b"])
source.repo.add_remote("target", dst_path)
# now, set up targets for the submodules:
# the need to be annexes, because we want to be able to copy data to them
# further down
sub1_target = AnnexRepo(sub1_pub, create=True)
sub1_target.checkout("TMP", ["-b"])
sub2_target = AnnexRepo(sub2_pub, create=True)
sub2_target.checkout("TMP", ["-b"])
sub1 = GitRepo(opj(src_path, 'subm 1'), create=False)
sub2 = GitRepo(opj(src_path, '2'), create=False)
sub1.add_remote("target", sub1_pub)
sub2.add_remote("target", sub2_pub)
res = publish(dataset=source,
to="target",
path=['test-annex.dat'],
result_xfm='paths')
# first it would publish data and then push
# TODO order is not fixed (yet)
#eq_(res, [opj(source.path, 'test-annex.dat'), source.path])
eq_(set(res), set([opj(source.path, 'test-annex.dat'), source.path]))
# XXX master was not checked out in dst!
eq_(list(target.get_branch_commits_(DEFAULT_BRANCH)),
list(source.repo.get_branch_commits_(DEFAULT_BRANCH)))
assert_git_annex_branch_published(source.repo, target)
# we need compare target/<default branch>:
target.checkout(DEFAULT_BRANCH)
ok_(target.file_has_content('test-annex.dat'))
# make sure that whatever we published is actually consumable
dst_clone = install(dst_clone_path,
source=dst_path,
result_xfm='datasets',
return_type='item-or-list')
nok_(dst_clone.repo.file_has_content('test-annex.dat'))
res = dst_clone.get('test-annex.dat')
ok_(dst_clone.repo.file_has_content('test-annex.dat'))
res = publish(dataset=source, to="target", path=['.'])
# there is nothing to publish on 2nd attempt
#eq_(res, ([source, 'test-annex.dat'], []))
assert_result_count(res, 1, status='notneeded')
import glob
res = publish(dataset=source,
to="target",
path=glob.glob1(source.path, '*'))
# Note: This leads to recursive publishing, since expansion of '*'
# contains the submodules themselves in this setup
# only the subdatasets, targets are plain git repos, hence
# no file content is pushed, all content in super was pushed
# before
assert_result_count(res, 3)
assert_result_count(res, 1, status='ok', path=sub1.path)
assert_result_count(res, 1, status='ok', path=sub2.path)
assert_result_count(res, 1, status='notneeded', path=source.path)
# if we publish again -- nothing to be published
res = source.publish(to="target")
assert_result_count(res, 1, status='notneeded', path=source.path)
# if we drop a file and publish again -- dataset should be published
# since git-annex branch was updated
source.drop('test-annex.dat')
res = source.publish(to="target")
assert_result_count(res, 1, status='ok', path=source.path)
# and empty again if we try again
res = source.publish(to="target")
assert_result_count(res, 1, status='notneeded', path=source.path)
# data integrity check looks identical from all perspectives
# minus "note" statements from git-annex
eq_(filter_fsck_error_msg(source.repo.fsck()),
filter_fsck_error_msg(source.repo.fsck(remote='target')))
eq_(filter_fsck_error_msg(target.fsck()),
filter_fsck_error_msg(source.repo.fsck(remote='target')))
def test_publish_with_data(origin, src_path, dst_path, sub1_pub, sub2_pub,
dst_clone_path):
# prepare src
source = install(src_path, source=origin, recursive=True)
source.repo.get('test-annex.dat')
# create plain git at target:
target = AnnexRepo(dst_path, create=True)
target.checkout("TMP", ["-b"])
source.repo.add_remote("target", dst_path)
# now, set up targets for the submodules:
# the need to be annexes, because we want to be able to copy data to them
# further down
sub1_target = AnnexRepo(sub1_pub, create=True)
sub1_target.checkout("TMP", ["-b"])
sub2_target = AnnexRepo(sub2_pub, create=True)
sub2_target.checkout("TMP", ["-b"])
sub1 = GitRepo(opj(src_path, 'subm 1'), create=False)
sub2 = GitRepo(opj(src_path, '2'), create=False)
sub1.add_remote("target", sub1_pub)
sub2.add_remote("target", sub2_pub)
res = publish(dataset=source,
to="target",
path=['test-annex.dat'],
result_xfm='paths')
# first it would publish data and then push
# TODO order is not fixed (yet)
#eq_(res, [opj(source.path, 'test-annex.dat'), source.path])
eq_(set(res), set([opj(source.path, 'test-annex.dat'), source.path]))
# XXX master was not checked out in dst!
eq_(list(target.get_branch_commits_("master")),
list(source.repo.get_branch_commits_("master")))
# TODO: last commit in git-annex branch differs. Probably fine,
# but figure out, when exactly to expect this for proper testing:
# yoh: they differ because local annex records information about now
# file being available in that remote, and remote one does it via a call in
# the hook I guess. So they both get the same information but in two
# different commits. I do not observe such behavior of remote having git-annex
# automagically updated in older clones
# which do not have post-receive hook on remote side
eq_(
list(target.get_branch_commits_("git-annex"))[1:],
list(source.repo.get_branch_commits_("git-annex"))[1:])
# we need compare target/master:
target.checkout("master")
ok_(target.file_has_content('test-annex.dat'))
# make sure that whatever we published is actually consumable
dst_clone = install(dst_clone_path,
source=dst_path,
result_xfm='datasets',
return_type='item-or-list')
nok_(dst_clone.repo.file_has_content('test-annex.dat'))
res = dst_clone.get('test-annex.dat')
ok_(dst_clone.repo.file_has_content('test-annex.dat'))
res = publish(dataset=source, to="target", path=['.'])
# there is nothing to publish on 2nd attempt
#eq_(res, ([source, 'test-annex.dat'], []))
assert_result_count(res, 1, status='notneeded')
import glob
res = publish(dataset=source,
to="target",
path=glob.glob1(source.path, '*'))
# Note: This leads to recursive publishing, since expansion of '*'
# contains the submodules themselves in this setup
# only the subdatasets, targets are plain git repos, hence
# no file content is pushed, all content in super was pushed
# before
assert_result_count(res, 3)
assert_result_count(res, 1, status='ok', path=sub1.path)
assert_result_count(res, 1, status='ok', path=sub2.path)
assert_result_count(res, 1, status='notneeded', path=source.path)
# if we publish again -- nothing to be published
res = source.publish(to="target")
assert_result_count(res, 1, status='notneeded', path=source.path)
# if we drop a file and publish again -- dataset should be published
# since git-annex branch was updated
source.drop('test-annex.dat')
res = source.publish(to="target")
assert_result_count(res, 1, status='ok', path=source.path)
# and empty again if we try again
res = source.publish(to="target")
assert_result_count(res, 1, status='notneeded', path=source.path)
# data integrity check looks identical from all perspectives
# minus "note" statements from git-annex
eq_(filter_fsck_error_msg(source.repo.fsck()),
filter_fsck_error_msg(source.repo.fsck(remote='target')))
eq_(filter_fsck_error_msg(target.fsck()),
filter_fsck_error_msg(source.repo.fsck(remote='target')))
