def test_ssh_get_connection():
manager = SSHManager()
assert manager._socket_dir is None, \
"Should be unset upon initialization. Got %s" % str(manager._socket_dir)
c1 = manager.get_connection('ssh://localhost')
assert manager._socket_dir, "Should be set after interactions with the manager"
assert_is_instance(c1, SSHConnection)
# subsequent call returns the very same instance:
ok_(manager.get_connection('ssh://localhost') is c1)
# fail on malformed URls (meaning: our fancy URL parser can't correctly
# deal with them):
#assert_raises(ValueError, manager.get_connection, 'localhost')
# we now allow those simple specifications of host to get_connection
c2 = manager.get_connection('localhost')
assert_is_instance(c2, SSHConnection)
# but should fail if it looks like something else
assert_raises(ValueError, manager.get_connection, 'localhost/')
assert_raises(ValueError, manager.get_connection, ':localhost')
# we can do what urlparse cannot
# assert_raises(ValueError, manager.get_connection, 'someone@localhost')
# next one is considered a proper url by urlparse (netloc:'',
# path='/localhost), but eventually gets turned into SSHRI(hostname='ssh',
# path='/localhost') -- which is fair IMHO -> invalid test
# assert_raises(ValueError, manager.get_connection, 'ssh:/localhost')
manager.close()
def test_ssh_get_connection():
manager = SSHManager()
assert manager._socket_dir is None, \
"Should be unset upon initialization. Got %s" % str(manager._socket_dir)
c1 = manager.get_connection('ssh://localhost')
assert manager._socket_dir, "Should be set after interactions with the manager"
assert_is_instance(c1, SSHConnection)
# subsequent call returns the very same instance:
ok_(manager.get_connection('ssh://localhost') is c1)
# fail on malformed URls (meaning: our fancy URL parser can't correctly
# deal with them):
#assert_raises(ValueError, manager.get_connection, 'localhost')
# we now allow those simple specifications of host to get_connection
c2 = manager.get_connection('localhost')
assert_is_instance(c2, SSHConnection)
# but should fail if it looks like something else
assert_raises(ValueError, manager.get_connection, 'localhost/')
assert_raises(ValueError, manager.get_connection, ':localhost')
# we can do what urlparse cannot
# assert_raises(ValueError, manager.get_connection, 'someone@localhost')
# next one is considered a proper url by urlparse (netloc:'',
# path='/localhost), but eventually gets turned into SSHRI(hostname='ssh',
# path='/localhost') -- which is fair IMHO -> invalid test
# assert_raises(ValueError, manager.get_connection, 'ssh:/localhost')
manager.close()
def test_protocol_commons(protocol_file):
for protocol_class in [
DryRunProtocol, DryRunExternalsProtocol, ExecutionTimeProtocol,
ExecutionTimeExternalsProtocol, NullProtocol
]:
protocol = protocol_class()
assert_is_instance(protocol, ProtocolInterface)
assert_equal(len(protocol), 0)
protocol.add_section(['some_command', 'some_option'],
Exception("Whatever exception"))
protocol.add_section(['another_command'], None)
assert_equal(len(protocol), 2 if protocol_class != NullProtocol else 0)
# test iterable:
assert_raises(AssertionError, assert_raises, TypeError, iter, protocol)
for section in protocol:
assert_in('command', section)
for item in range(len(protocol)):
assert_is_instance(protocol.__getitem__(item), dict)
# test __str__:
str_ = str(protocol)
# test write_to_file:
protocol.write_to_file(protocol_file)
read_str = ''
with open(protocol_file, 'r') as f:
for line in f.readlines():
read_str += line
assert_equal(str_, read_str)
def test_newthings_coming_down(originpath, destpath):
origin = GitRepo(originpath, create=True)
create_tree(originpath, {'load.dat': 'heavy'})
Dataset(originpath).save('load.dat')
ds = install(source=originpath,
path=destpath,
result_xfm='datasets',
return_type='item-or-list')
assert_is_instance(ds.repo, GitRepo)
assert_in('origin', ds.repo.get_remotes())
# turn origin into an annex
origin = AnnexRepo(originpath, create=True)
# clone doesn't know yet
assert_false(knows_annex(ds.path))
# but after an update it should
# no merge, only one sibling, no parameters should be specific enough
assert_result_count(ds.update(), 1, status='ok', type='dataset')
assert (knows_annex(ds.path))
# no branches appeared
eq_(ds.repo.get_branches(), [DEFAULT_BRANCH])
# now merge, and get an annex
assert_result_count(ds.update(merge=True),
1,
action='update',
status='ok',
type='dataset')
assert_in('git-annex', ds.repo.get_branches())
assert_is_instance(ds.repo, AnnexRepo)
# should be fully functional
testfname = opj(ds.path, 'load.dat')
assert_false(ds.repo.file_has_content(testfname))
ds.get('.')
ok_file_has_content(opj(ds.path, 'load.dat'), 'heavy')
# check that a new tag comes down
origin.tag('first!')
assert_result_count(ds.update(), 1, status='ok', type='dataset')
eq_(ds.repo.get_tags(output='name')[0], 'first!')
# and now we destroy the remote annex
origin.call_git(['config', '--remove-section', 'annex'])
rmtree(opj(origin.path, '.git', 'annex'), chmod_files=True)
origin.call_git(['branch', '-D', 'git-annex'])
origin = GitRepo(originpath)
assert_false(knows_annex(originpath))
# and update the local clone
# for now this should simply not fail (see gh-793), later might be enhanced to a
# graceful downgrade
before_branches = ds.repo.get_branches()
assert_result_count(ds.update(), 1, status='ok', type='dataset')
eq_(before_branches, ds.repo.get_branches())
# annex branch got pruned
eq_(['origin/HEAD', 'origin/' + DEFAULT_BRANCH],
ds.repo.get_remote_branches())
# check that a new tag comes down even if repo types mismatch
origin.tag('second!')
assert_result_count(ds.update(), 1, status='ok', type='dataset')
eq_(ds.repo.get_tags(output='name')[-1], 'second!')
def test_install_dataset_from_instance(src, dst):
origin = Dataset(src)
clone = install(source=origin, path=dst)
assert_is_instance(clone, Dataset)
ok_startswith(clone.path, dst)
ok_(clone.is_installed())
ok_(GitRepo.is_valid_repo(clone.path))
ok_clean_git(clone.path, annex=None)
assert_in('INFO.txt', clone.repo.get_indexed_files())
def test_install_dataset_from_instance(src, dst):
origin = Dataset(src)
clone = install(source=origin, path=dst)
assert_is_instance(clone, Dataset)
ok_startswith(clone.path, dst)
ok_(clone.is_installed())
ok_(GitRepo.is_valid_repo(clone.path))
ok_clean_git(clone.path, annex=None)
assert_in('INFO.txt', clone.repo.get_indexed_files())
def test_ssh_get_connection():
manager = SSHManager()
c1 = manager.get_connection('ssh://localhost')
assert_is_instance(c1, SSHConnection)
# subsequent call returns the very same instance:
ok_(manager.get_connection('ssh://localhost') is c1)
# fail on malformed URls (meaning: out fancy URL parser can't correctly
# deal with them):
assert_raises(ValueError, manager.get_connection, 'localhost')
def test_ssh_get_connection():
manager = SSHManager()
c1 = manager.get_connection('ssh://localhost')
assert_is_instance(c1, SSHConnection)
# subsequent call returns the very same instance:
ok_(manager.get_connection('ssh://localhost') is c1)
# fail on malformed URls (meaning: out fancy URL parser can't correctly
# deal with them):
assert_raises(ValueError, manager.get_connection, 'localhost')
def decorated_test2(ds):
# we get a Dataset instance
assert_is_instance(ds, Dataset)
# it's a clone in a temp. location, not within the cache
assert_not_in(cache_dir, ds.pathobj.parents)
assert_result_count(ds.siblings(), 1, type="sibling",
name="origin",
url=str(cache_dir / name_in_cache))
here = ds.config.get("annex.uuid")
origin = ds.config.get("remote.origin.annex-uuid")
where = ds.repo.whereis(str(annexed_file))
assert_in(here, where)
assert_in(origin, where)
return ds.pathobj, ds.repo.pathobj
def decorated_test3(ds):
# we get a Dataset instance
assert_is_instance(ds, Dataset)
# it's a clone in a temp. location, not within the cache
assert_not_in(cache_dir, ds.pathobj.parents)
assert_result_count(ds.siblings(), 1, type="sibling",
name="origin",
url=str(cache_dir / name_in_cache))
# origin is the same cached dataset, that got this content in
# decorated_test2 before. Should still be there. But "here" we
# didn't request it
here = ds.config.get("annex.uuid")
origin = ds.config.get("remote.origin.annex-uuid")
where = ds.repo.whereis(str(annexed_file))
assert_not_in(here, where)
assert_in(origin, where)
return ds.pathobj, ds.repo.pathobj
def test_ssh_get_connection():
manager = SSHManager()
c1 = manager.get_connection('ssh://localhost')
assert_is_instance(c1, SSHConnection)
# subsequent call returns the very same instance:
ok_(manager.get_connection('ssh://localhost') is c1)
# fail on malformed URls (meaning: our fancy URL parser can't correctly
# deal with them):
assert_raises(ValueError, manager.get_connection, 'localhost')
# we can do what urlparse cannot
# assert_raises(ValueError, manager.get_connection, 'someone@localhost')
# next one is considered a proper url by urlparse (netloc:'',
# path='/localhost), but eventually gets turned into SSHRI(hostname='ssh',
# path='/localhost') -- which is fair IMHO -> invalid test
# assert_raises(ValueError, manager.get_connection, 'ssh:/localhost')
manager.close()
def decorated_test4(ds):
# we get a Dataset instance
assert_is_instance(ds, Dataset)
# it's a clone in a temp. location, not within the cache
assert_not_in(cache_dir, ds.pathobj.parents)
assert_result_count(ds.siblings(), 1, type="sibling",
name=DEFAULT_REMOTE,
url=str(cache_dir / name_in_cache))
# origin is the same cached dataset, that got this content in
# decorated_test2 before. Should still be there. But "here" we
# didn't request it
here = ds.config.get("annex.uuid")
origin = ds.config.get(f"remote.{DEFAULT_REMOTE}.annex-uuid")
where = ds.repo.whereis(str(annexed_file))
assert_not_in(here, where)
assert_in(origin, where)
assert_equal(ds.repo.get_hexsha(),
"541cf855d13c2a338ff2803d4488daf0035e568f")
return ds.pathobj, ds.repo.pathobj
def test_find_containers(path):
ds = Dataset(path).create(force=True)
ds.save(path=[op.join('sub', 'i.img')], message="dummy container")
ds.containers_add("i", image=op.join('sub', 'i.img'))
ok_clean_git(path)
# find the only one
res = find_container(ds)
assert_is_instance(res, dict)
assert_result_count([res],
1,
status="ok",
path=op.join(ds.path, "sub", "i.img"))
# find by name
res = find_container(ds, "i")
assert_is_instance(res, dict)
assert_result_count([res],
1,
status="ok",
path=op.join(ds.path, "sub", "i.img"))
# find by path
res = find_container(ds, op.join("sub", "i.img"))
assert_is_instance(res, dict)
assert_result_count([res],
1,
status="ok",
path=op.join(ds.path, "sub", "i.img"))
# don't find another thing
assert_raises(ValueError, find_container, ds, "nothere")
def test_get_repo_instance_annex(path):
# get instance from path
repo = get_repo_instance(path, AnnexRepo)
assert_is_instance(repo, AnnexRepo)
eq_(realpath(repo.path), realpath(path))
old_pwd = getpwd()
# get instance from current dir
chpwd(path)
repo = get_repo_instance()
assert_is_instance(repo, AnnexRepo)
eq_(realpath(repo.path), realpath(path))
# get instance from current subdir
new_subdir = opj(path, "subdir")
mkdir(new_subdir)
chpwd(new_subdir)
eq_(new_subdir, getpwd())
repo = get_repo_instance()
assert_is_instance(repo, AnnexRepo)
eq_(realpath(repo.path), realpath(path))
chpwd(old_pwd)
def test_get_repo_instance_git(path):
real_path = Path(path).resolve()
# get instance from path
repo = get_repo_instance(path, GitRepo)
assert_is_instance(repo, GitRepo)
eq_(repo.pathobj, real_path)
old_pwd = getpwd()
# get instance from current dir
chpwd(path)
repo = get_repo_instance()
assert_is_instance(repo, GitRepo)
eq_(repo.pathobj, real_path)
# get instance from current subdir
new_subdir = opj(path, "subdir")
mkdir(new_subdir)
chpwd(new_subdir)
eq_(new_subdir, getpwd())
repo = get_repo_instance()
assert_is_instance(repo, GitRepo)
eq_(repo.pathobj, real_path)
chpwd(old_pwd)
def test_GitRepo_instance_from_not_existing(path, path2):
# 1. create=False and path doesn't exist:
repo = GitRepo(path)
assert_false(op.exists(path))
# 2. create=False, path exists, but no git repo:
os.mkdir(path)
ok_(op.exists(path))
repo = GitRepo(path)
assert_false(op.exists(op.join(path, '.git')))
# 3. create=True, path doesn't exist:
gr = GitRepo(path2).init()
assert_is_instance(gr, GitRepo, "GitRepo was not created.")
ok_(op.exists(op.join(path2, '.git')))
# re-enable from core GitRepo has a status() method
#assert_repo_status(path2, annex=False)
# 4. create=True, path exists, but no git repo:
gr = GitRepo(path).init()
assert_is_instance(gr, GitRepo, "GitRepo was not created.")
ok_(op.exists(op.join(path, '.git')))
def test_Dataset_flyweight(path1, path2):
ds1 = Dataset(path1)
assert_is_instance(ds1, Dataset)
# instantiate again:
ds2 = Dataset(path1)
assert_is_instance(ds2, Dataset)
# the very same object:
ok_(ds1 is ds2)
# reference the same via relative path:
with chpwd(path1):
ds3 = Dataset(relpath(path1, start=path2))
ok_(ds1 == ds3)
ok_(ds1 is ds3)
# on windows as symlink is not what you think it is
if not on_windows:
# reference the same via symlink:
with chpwd(path2):
os.symlink(path1, 'linked')
ds3 = Dataset('linked')
ok_(ds3 == ds1)
ok_(ds3 is not ds1)
def test_GitRepo_flyweight(path1, path2):
import gc
repo1 = GitRepo(path1).init()
assert_is_instance(repo1, GitRepo)
# Due to issue 4862, we currently still require gc.collect() under unclear
# circumstances to get rid of an exception traceback when creating in an
# existing directory. That traceback references the respective function
# frames which in turn reference the repo instance (they are methods).
# Doesn't happen on all systems, though. Eventually we need to figure that
# out.
# However, still test for the refcount after gc.collect() to ensure we don't
# introduce new circular references and make the issue worse!
gc.collect()
# As long as we don't reintroduce any circular references or produce
# garbage during instantiation that isn't picked up immediately, `repo1`
# should be the only counted reference to this instance.
# Note, that sys.getrefcount reports its own argument and therefore one
# reference too much.
assert_equal(1, sys.getrefcount(repo1) - 1)
# instantiate again:
repo2 = GitRepo(path1).init()
assert_is_instance(repo2, GitRepo)
# the very same object:
ok_(repo1 is repo2)
# reference the same in a different way:
with chpwd(path1):
repo3 = GitRepo(op.relpath(path1, start=path2))
# it's the same object:
ok_(repo1 is repo3)
# and realpath attribute is the same, so they are still equal:
ok_(repo1 == repo3)
orig_id = id(repo1)
# Be sure we have exactly one object in memory:
assert_equal(
1,
len([
o for o in gc.get_objects()
if isinstance(o, GitRepo) and o.pathobj == Path(path1)
]))
# deleting one reference doesn't change anything - we still get the same
# thing:
gc.collect() # TODO: see first comment above
del repo1
ok_(repo2 is not None)
ok_(repo2 is repo3)
ok_(repo2 == repo3)
# re-requesting still delivers the same thing:
repo1 = GitRepo(path1)
assert_equal(orig_id, id(repo1))
# killing all references should result in the instance being gc'd and
# re-request yields a new object:
del repo1
del repo2
# Killing last reference will lead to garbage collection which will call
# GitRepo's finalizer:
with swallow_logs(new_level=1) as cml:
del repo3
gc.collect() # TODO: see first comment above
cml.assert_logged(msg="Finalizer called on: GitRepo(%s)" % path1,
level="Level 1",
regex=False)
# Flyweight is gone:
assert_not_in(path1, GitRepo._unique_instances.keys())
# gc doesn't know any instance anymore:
assert_equal([], [
o for o in gc.get_objects()
if isinstance(o, GitRepo) and o.pathobj == Path(path1)
])
# new object is created on re-request:
repo1 = GitRepo(path1)
assert_equal(
1,
len([
o for o in gc.get_objects()
if isinstance(o, GitRepo) and o.pathobj == Path(path1)
]))
def test_get_cached_dataset(cache_dir):
# patch DATALAD_TESTS_CACHE to not use the actual cache with
# the test testing that very cache.
cache_dir = Path(cache_dir)
# store file-based values for testrepo-minimalds for readability:
annexed_file = opj('inannex', 'animated.gif')
annexed_file_key = "MD5E-s144625--4c458c62b7ac8ec8e19c8ff14b2e34ad.gif"
with patch(CACHE_PATCH_STR, new=cache_dir):
# tuples to test (url, version, keys, class):
test_cases = [
# a simple testrepo
("https://github.com/datalad/testrepo--minimalds",
"541cf855d13c2a338ff2803d4488daf0035e568f", None, AnnexRepo),
# Same repo, but request paths to be present. This should work
# with a subsequent call, although the first one did not already
# request any:
("https://github.com/datalad/testrepo--minimalds",
"9dd8b56cc706ab56185f2ceb75fbe9de9b606724", annexed_file_key,
AnnexRepo),
# Same repo again, but invalid version
(
"https://github.com/datalad/testrepo--minimalds",
"nonexistent",
"irrelevantkey", # invalid version; don't even try to get the key
AnnexRepo),
# same thing with different name should be treated as a new thing:
("https://github.com/datalad/testrepo--minimalds", "git-annex",
None, AnnexRepo),
# try a plain git repo to make sure we can deal with that:
# Note, that we first need a test case w/o a `key` parameter to not
# blow up the test when Clone is patched, resulting in a MagicMock
# instead of a Dataset instance within get_cached_dataset. In the
# second case it's already cached then, so the patched Clone is
# never executed.
("https://github.com/datalad/datalad.org", None, None, GitRepo),
(
"https://github.com/datalad/datalad.org",
"gh-pages",
"ignored-key", # it's a git repo; don't even try to get a key
GitRepo),
]
for url, version, keys, cls in test_cases:
target = cache_dir / url2filename(url)
# assuming it doesn't exist yet - patched cache dir!
in_cache_before = target.exists()
with patch(CLONE_PATCH_STR) as exec_clone:
try:
ds = get_cached_dataset(url, version, keys)
invalid_version = False
except AssertionError:
# should happen only if `version` wasn't found. Implies
# that the dataset exists in cache (although not returned
# due to exception)
assert_true(version)
assert_false(Dataset(target).repo.commit_exists(version))
# mark for later assertions (most of them should still hold
# true)
invalid_version = True
assert_equal(exec_clone.call_count, 0 if in_cache_before else 1)
# Patch prevents actual execution. Now do it for real. Note, that
# this might be necessary for content retrieval even if dataset was
# in cache before.
try:
ds = get_cached_dataset(url, version, keys)
except AssertionError:
# see previous call
assert_true(invalid_version)
assert_is_instance(ds, Dataset)
assert_true(ds.is_installed())
assert_equal(target, ds.pathobj)
assert_is_instance(ds.repo, cls)
if keys and not invalid_version and \
AnnexRepo.is_valid_repo(ds.path):
# Note: it's not supposed to get that content if passed
# `version` wasn't available. get_cached_dataset would then
# raise before and not download anything only to raise
# afterwards.
here = ds.config.get("annex.uuid")
where = ds.repo.whereis(ensure_list(keys), key=True)
assert_true(all(here in remotes for remotes in where))
# version check. Note, that all `get_cached_dataset` is supposed to
# do, is verifying, that specified version exists - NOT check it
# out"
if version and not invalid_version:
assert_true(ds.repo.commit_exists(version))
# re-execution
with patch(CLONE_PATCH_STR) as exec_clone:
try:
ds2 = get_cached_dataset(url, version, keys)
except AssertionError:
assert_true(invalid_version)
exec_clone.assert_not_called()
# returns the same Dataset as before:
assert_is(ds, ds2)
def test_Dataset_flyweight(path1, path2):
import gc
import sys
ds1 = Dataset(path1)
assert_is_instance(ds1, Dataset)
# Don't create circular references or anything similar
assert_equal(1, sys.getrefcount(ds1) - 1)
ds1.create()
# Due to issue 4862, we currently still require gc.collect() under unclear
# circumstances to get rid of an exception traceback when creating in an
# existing directory. That traceback references the respective function
# frames which in turn reference the repo instance (they are methods).
# Doesn't happen on all systems, though. Eventually we need to figure that
# out.
# However, still test for the refcount after gc.collect() to ensure we don't
# introduce new circular references and make the issue worse!
gc.collect()
# refcount still fine after repo creation:
assert_equal(1, sys.getrefcount(ds1) - 1)
# instantiate again:
ds2 = Dataset(path1)
assert_is_instance(ds2, Dataset)
# the very same object:
ok_(ds1 is ds2)
# reference the same via relative path:
with chpwd(path1):
ds3 = Dataset(relpath(path1, start=path2))
ok_(ds1 == ds3)
ok_(ds1 is ds3)
# gc knows one such object only:
eq_(
1,
len([
o for o in gc.get_objects()
if isinstance(o, Dataset) and o.path == path1
]))
# on windows a symlink is not what you think it is
if not on_windows:
# reference the same via symlink:
with chpwd(path2):
os.symlink(path1, 'linked')
ds4 = Dataset('linked')
ds4_id = id(ds4)
ok_(ds4 == ds1)
ok_(ds4 is not ds1)
# underlying repo, however, IS the same:
ok_(ds4.repo is ds1.repo)
# deleting one reference has no effect on the other:
del ds1
gc.collect() # TODO: see first comment above
ok_(ds2 is not None)
ok_(ds2.repo is ds3.repo)
if not on_windows:
ok_(ds2.repo is ds4.repo)
# deleting remaining references should lead to garbage collection
del ds2
with swallow_logs(new_level=1) as cml:
del ds3
gc.collect() # TODO: see first comment above
# flyweight vanished:
assert_not_in(path1, Dataset._unique_instances.keys())
# no such instance known to gc anymore:
eq_([], [
o for o in gc.get_objects()
if isinstance(o, Dataset) and o.path == path1
])
# underlying repo should only be cleaned up, if ds3 was the last
# reference to it. Otherwise the repo instance should live on
# (via symlinked ds4):
finalizer_log = "Finalizer called on: AnnexRepo(%s)" % path1
if on_windows:
cml.assert_logged(msg=finalizer_log, level="Level 1", regex=False)
else:
assert_not_in(finalizer_log, cml.out)
# symlinked is still there:
ok_(ds4 is not None)
eq_(ds4_id, id(ds4))
def test_GitRepo_instance_from_existing(path):
GitRepo(path).init()
gr = GitRepo(path)
assert_is_instance(gr, GitRepo, "GitRepo was not created.")
ok_(op.exists(op.join(path, '.git')))
