def __init__(self, root_dir=None, scm=None, rev=None):
from dvc.state import State, StateNoop
from dvc.lock import make_lock
from dvc.scm import SCM
from dvc.cache import Cache
from dvc.data_cloud import DataCloud
from dvc.repo.metrics import Metrics
from dvc.repo.plots import Plots
from dvc.repo.params import Params
from dvc.scm.tree import WorkingTree
from dvc.utils.fs import makedirs
from dvc.stage.cache import StageCache
if scm:
# use GitTree instead of WorkingTree as default repo tree instance
tree = scm.get_tree(rev)
self.root_dir = self.find_root(root_dir, tree)
self.scm = scm
self.tree = tree
self.state = StateNoop()
else:
root_dir = self.find_root(root_dir)
self.root_dir = os.path.abspath(os.path.realpath(root_dir))
self.tree = WorkingTree(self.root_dir)
self.dvc_dir = os.path.join(self.root_dir, self.DVC_DIR)
self.config = Config(self.dvc_dir, tree=self.tree)
if not scm:
no_scm = self.config["core"].get("no_scm", False)
self.scm = SCM(self.root_dir, no_scm=no_scm)
self.tmp_dir = os.path.join(self.dvc_dir, "tmp")
self.index_dir = os.path.join(self.tmp_dir, "index")
makedirs(self.index_dir, exist_ok=True)
hardlink_lock = self.config["core"].get("hardlink_lock", False)
self.lock = make_lock(
os.path.join(self.tmp_dir, "lock"),
tmp_dir=self.tmp_dir,
hardlink_lock=hardlink_lock,
friendly=True,
)
self.cache = Cache(self)
self.cloud = DataCloud(self)
if not scm:
# NOTE: storing state and link_state in the repository itself to
# avoid any possible state corruption in 'shared cache dir'
# scenario.
self.state = State(self.cache.local)
self.stage_cache = StageCache(self)
self.metrics = Metrics(self)
self.plots = Plots(self)
self.params = Params(self)
self._ignore()
def test_nobranch(self):
tree = WorkingTree(self._root_dir)
self.assertWalkEqual(
tree.walk("."),
[
(
self._root_dir,
["data_dir"],
["bar", "тест", "code.py", "foo"],
),
(join(self._root_dir, "data_dir"), ["data_sub_dir"], ["data"]),
(
join(self._root_dir, "data_dir", "data_sub_dir"),
[],
["data_sub"],
),
],
)
self.assertWalkEqual(
tree.walk(join("data_dir", "data_sub_dir")),
[(
join(self._root_dir, "data_dir", "data_sub_dir"),
[],
["data_sub"],
)],
)
def test_nobranch(self):
tree = WorkingTree(self._root_dir)
dvcignore = DvcIgnoreFilter(self._root_dir, tree)
self.assertWalkEqual(
tree.walk(".", dvcignore=dvcignore),
[
(
self._root_dir,
["data_dir"],
["bar", "тест", "code.py", "foo"],
),
(join(self._root_dir, "data_dir"), ["data_sub_dir"], ["data"]),
(
join(self._root_dir, "data_dir", "data_sub_dir"),
[],
["data_sub"],
),
],
)
self.assertWalkEqual(
tree.walk(join("data_dir", "data_sub_dir"), dvcignore=dvcignore),
[(
join(self._root_dir, "data_dir", "data_sub_dir"),
[],
["data_sub"],
)],
)
def test_subdir(self):
tree = WorkingTree(self._root_dir)
self.assertWalkEqual(
tree.walk(join("data_dir", "data_sub_dir")),
[(
join(self._root_dir, "data_dir", "data_sub_dir"),
[],
["data_sub"],
)],
)
def test(self):
tree = WorkingTree()
self.assertWalkEqual(
tree.walk("."),
[
(".", ["data_dir"], ["code.py", "bar", "тест", "foo"]),
(join("data_dir"), ["data_sub_dir"], ["data"]),
(join("data_dir", "data_sub_dir"), [], ["data_sub"]),
],
)
def test_subdir(self):
dvcignore = DvcIgnoreFilter(self.root_dir)
tree = WorkingTree(self._root_dir)
self.assertWalkEqual(
tree.walk(join("data_dir", "data_sub_dir"), dvcignore=dvcignore),
[(
join(self._root_dir, "data_dir", "data_sub_dir"),
[],
["data_sub"],
)],
)
def __init__(self, root_dir=None):
from dvc.state import State
from dvc.lock import make_lock
from dvc.scm import SCM
from dvc.cache import Cache
from dvc.data_cloud import DataCloud
from dvc.repo.metrics import Metrics
from dvc.scm.tree import WorkingTree
from dvc.repo.tag import Tag
from dvc.utils import makedirs
root_dir = self.find_root(root_dir)
self.root_dir = os.path.abspath(os.path.realpath(root_dir))
self.dvc_dir = os.path.join(self.root_dir, self.DVC_DIR)
self.config = Config(self.dvc_dir)
self.scm = SCM(self.root_dir)
self.tree = WorkingTree(self.root_dir)
self.tmp_dir = os.path.join(self.dvc_dir, "tmp")
makedirs(self.tmp_dir, exist_ok=True)
hardlink_lock = self.config.config["core"].get("hardlink_lock", False)
self.lock = make_lock(
os.path.join(self.dvc_dir, "lock"),
tmp_dir=os.path.join(self.dvc_dir, "tmp"),
hardlink_lock=hardlink_lock,
friendly=True,
)
# NOTE: storing state and link_state in the repository itself to avoid
# any possible state corruption in 'shared cache dir' scenario.
self.state = State(self, self.config.config)
core = self.config.config[Config.SECTION_CORE]
level = core.get(Config.SECTION_CORE_LOGLEVEL)
if level:
logger.setLevel(level.upper())
self.cache = Cache(self)
self.cloud = DataCloud(self)
self.metrics = Metrics(self)
self.tag = Tag(self)
self._ignore()
def __init__(self, root_dir=None):
from dvc.state import State
from dvc.lock import make_lock
from dvc.scm import SCM
from dvc.cache import Cache
from dvc.data_cloud import DataCloud
from dvc.repo.metrics import Metrics
from dvc.repo.params import Params
from dvc.scm.tree import WorkingTree
from dvc.utils.fs import makedirs
from dvc.stage.cache import StageCache
root_dir = self.find_root(root_dir)
self.root_dir = os.path.abspath(os.path.realpath(root_dir))
self.dvc_dir = os.path.join(self.root_dir, self.DVC_DIR)
self.config = Config(self.dvc_dir)
no_scm = self.config["core"].get("no_scm", False)
self.scm = SCM(self.root_dir, no_scm=no_scm)
self.tree = WorkingTree(self.root_dir)
self.tmp_dir = os.path.join(self.dvc_dir, "tmp")
self.index_dir = os.path.join(self.tmp_dir, "index")
makedirs(self.index_dir, exist_ok=True)
hardlink_lock = self.config["core"].get("hardlink_lock", False)
self.lock = make_lock(
os.path.join(self.tmp_dir, "lock"),
tmp_dir=self.tmp_dir,
hardlink_lock=hardlink_lock,
friendly=True,
)
# NOTE: storing state and link_state in the repository itself to avoid
# any possible state corruption in 'shared cache dir' scenario.
self.state = State(self)
self.cache = Cache(self)
self.cloud = DataCloud(self)
self.stage_cache = StageCache(self.cache.local.cache_dir)
self.metrics = Metrics(self)
self.params = Params(self)
self._ignore()
def work_tree(self):
# When using repo.brancher, repo.tree may change to/from WorkingTree to
# GitTree arbitarily. When repo.tree is GitTree, local cache needs to
# use its own WorkingTree instance.
if self.repo:
return WorkingTree(self.repo.root_dir)
return None
def __init__(self, repo):
self.repo = repo
self.root_dir = repo.root_dir
self.tree = WorkingTree(self.root_dir)
state_config = repo.config.get("state", {})
self.row_limit = state_config.get("row_limit", self.STATE_ROW_LIMIT)
self.row_cleanup_quota = state_config.get(
"row_cleanup_quota", self.STATE_ROW_CLEANUP_QUOTA
)
if not repo.tmp_dir:
self.state_file = None
return
self.state_file = os.path.join(repo.tmp_dir, self.STATE_FILE)
# https://www.sqlite.org/tempfiles.html
self.temp_files = [
self.state_file + "-journal",
self.state_file + "-wal",
]
self.database = None
self.cursor = None
self.inserts = 0
def test_path_object_and_str_are_valid_types_get_mtime_and_size(
path, repo_dir):
tree = CleanTree(WorkingTree(repo_dir.root_dir))
time, size = get_mtime_and_size(path, tree)
object_time, object_size = get_mtime_and_size(PathInfo(path), tree)
assert time == object_time
assert size == object_size
def test_ignore_collecting_dvcignores(repo_dir, dname):
top_ignore_file = os.path.join(repo_dir.root_dir, os.path.dirname(dname),
DvcIgnore.DVCIGNORE_FILE)
repo_dir.create(top_ignore_file, os.path.basename(dname))
ignore_file = os.path.join(repo_dir.root_dir, dname,
DvcIgnore.DVCIGNORE_FILE)
repo_dir.create(ignore_file, repo_dir.FOO)
assert DvcIgnoreFilter(repo_dir.root_dir,
WorkingTree(repo_dir.root_dir)).ignores == {
DvcIgnoreDirs([".git", ".hg", ".dvc"]),
DvcIgnorePatterns(
top_ignore_file,
WorkingTree(repo_dir.root_dir)),
}
def _check(self, branch, target, with_deps, expected):
if branch:
self.dvc.tree = GitTree(self.dvc.scm.git, branch)
else:
self.dvc.tree = WorkingTree()
result = self.dvc.collect(target + ".dvc", with_deps=with_deps)
self.assertEqual([[str(j) for j in i.outs] for i in result], expected)
return result
def test_path_object_and_str_are_valid_types_get_mtime_and_size(
path, repo_dir):
dvcignore = DvcIgnoreFilter(repo_dir.root_dir,
WorkingTree(repo_dir.root_dir))
time, size = get_mtime_and_size(path, dvcignore)
object_time, object_size = get_mtime_and_size(PathInfo(path), dvcignore)
assert time == object_time
assert size == object_size
def __init__(self, root_dir=None):
from dvc.config import Config
from dvc.state import State
from dvc.lock import Lock
from dvc.scm import SCM
from dvc.cache import Cache
from dvc.data_cloud import DataCloud
from dvc.updater import Updater
from dvc.repo.metrics import Metrics
from dvc.scm.tree import WorkingTree
from dvc.repo.tag import Tag
from dvc.repo.pkg import Pkg
root_dir = self.find_root(root_dir)
self.root_dir = os.path.abspath(os.path.realpath(root_dir))
self.dvc_dir = os.path.join(self.root_dir, self.DVC_DIR)
self.config = Config(self.dvc_dir)
self.tree = WorkingTree(self.root_dir)
self.scm = SCM(self.root_dir, repo=self)
self.lock = Lock(self.dvc_dir)
# NOTE: storing state and link_state in the repository itself to avoid
# any possible state corruption in 'shared cache dir' scenario.
self.state = State(self, self.config.config)
core = self.config.config[Config.SECTION_CORE]
level = core.get(Config.SECTION_CORE_LOGLEVEL)
if level:
logger.setLevel(level.upper())
self.cache = Cache(self)
self.cloud = DataCloud(self, config=self.config.config)
self.updater = Updater(self.dvc_dir)
self.metrics = Metrics(self)
self.tag = Tag(self)
self.pkg = Pkg(self)
self._ignore()
self.updater.check()
def test(self):
dvcignore = DvcIgnoreFilter(self.root_dir)
tree = WorkingTree(self._root_dir)
self.assertWalkEqual(
tree.walk(self._root_dir, dvcignore=dvcignore),
[
(
self._root_dir,
["data_dir"],
["code.py", "bar", "тест", "foo"],
),
(join(self._root_dir, "data_dir"), ["data_sub_dir"], ["data"]),
(
join(self._root_dir, "data_dir", "data_sub_dir"),
[],
["data_sub"],
),
],
)
def test_ignore_collecting_dvcignores(tmp_dir, dvc, dname):
tmp_dir.gen({"dir": {"subdir": {}}})
top_ignore_file = (tmp_dir / dname).with_name(DvcIgnore.DVCIGNORE_FILE)
top_ignore_file.write_text(os.path.basename(dname))
ignore_file = tmp_dir / dname / DvcIgnore.DVCIGNORE_FILE
ignore_file.write_text("foo")
assert dvc.tree.dvcignore.ignores == {
DvcIgnoreDirs([".git", ".hg", ".dvc"]),
DvcIgnorePatterns(fspath(top_ignore_file), WorkingTree(dvc.root_dir)),
}
def test(self):
tree = CleanTree(WorkingTree(self.root_dir))
file_time, file_size = get_mtime_and_size(self.DATA, tree)
dir_time, dir_size = get_mtime_and_size(self.DATA_DIR, tree)
actual_file_size = os.path.getsize(self.DATA)
actual_dir_size = os.path.getsize(self.DATA) + os.path.getsize(
self.DATA_SUB)
self.assertIs(type(file_time), str)
self.assertIs(type(file_size), str)
self.assertEqual(file_size, str(actual_file_size))
self.assertIs(type(dir_time), str)
self.assertIs(type(dir_size), str)
self.assertEqual(dir_size, str(actual_dir_size))
def test_path_object_and_str_are_valid_types_get_mtime_and_size(tmp_dir):
tmp_dir.gen(
{"dir": {"dir_file": "dir file content"}, "file": "file_content"}
)
tree = CleanTree(WorkingTree(tmp_dir))
time, size = get_mtime_and_size("dir", tree)
object_time, object_size = get_mtime_and_size(PathInfo("dir"), tree)
assert time == object_time
assert size == object_size
time, size = get_mtime_and_size("file", tree)
object_time, object_size = get_mtime_and_size(PathInfo("file"), tree)
assert time == object_time
assert size == object_size
def test(self):
dvcignore = DvcIgnoreFilter(self.root_dir, WorkingTree(self.root_dir))
file_time, file_size = get_mtime_and_size(self.DATA, dvcignore)
dir_time, dir_size = get_mtime_and_size(self.DATA_DIR, dvcignore)
actual_file_size = os.path.getsize(self.DATA)
actual_dir_size = os.path.getsize(self.DATA) + os.path.getsize(
self.DATA_SUB)
self.assertIs(type(file_time), str)
self.assertIs(type(file_size), str)
self.assertEqual(file_size, str(actual_file_size))
self.assertIs(type(dir_time), str)
self.assertIs(type(dir_size), str)
self.assertEqual(dir_size, str(actual_dir_size))
def test_ignore_collecting_dvcignores(tmp_dir, dvc, dname):
tmp_dir.gen({"dir": {"subdir": {}}})
top_ignore_file = (tmp_dir / dname).with_name(DvcIgnore.DVCIGNORE_FILE)
top_ignore_file.write_text(os.path.basename(dname))
ignore_file = tmp_dir / dname / DvcIgnore.DVCIGNORE_FILE
ignore_file.write_text("foo")
assert len(dvc.tree.dvcignore.ignores) == 3
assert DvcIgnoreDirs([".git", ".hg", ".dvc"]) in dvc.tree.dvcignore.ignores
assert (DvcIgnorePatterns(os.fspath(top_ignore_file),
WorkingTree(dvc.root_dir))
in dvc.tree.dvcignore.ignores)
assert any(i for i in dvc.tree.dvcignore.ignores
if isinstance(i, DvcIgnoreRepo))
def _ls(repo, path_info, recursive=None, dvc=False):
from dvc.ignore import CleanTree
from dvc.repo.tree import DvcTree
from dvc.scm.tree import WorkingTree
if dvc:
tree = DvcTree(repo)
else:
tree = CleanTree(WorkingTree(repo.root_dir))
ret = {}
try:
for root, dirs, files in tree.walk(path_info.fspath):
for fname in files:
info = PathInfo(root) / fname
path = str(info.relative_to(path_info))
ret[path] = {
"isout": dvc,
"isdir": False,
"isexec": False if dvc else tree.isexec(info.fspath),
}
if not recursive:
for dname in dirs:
info = PathInfo(root) / dname
path = str(info.relative_to(path_info))
ret[path] = {
"isout": tree.isdvc(info.fspath) if dvc else False,
"isdir": True,
"isexec": False if dvc else tree.isexec(info.fspath),
}
break
except NotADirectoryError:
return {
path_info.name: {
"isout": dvc,
"isdir": False,
"isexec": False if dvc else tree.isexec(path_info.fspath),
}
}
except FileNotFoundError:
return {}
return ret
def open(self, path, mode="r", encoding="utf-8"):
try:
outs = self._find_outs(path, strict=False)
except OutputNotFoundError as exc:
raise FileNotFoundError from exc
if len(outs) != 1 or outs[0].is_dir_checksum:
raise IOError(errno.EISDIR)
out = outs[0]
# temporary hack to make cache use WorkingTree and not GitTree, because
# cache dir doesn't exist in the latter.
saved_tree = self.repo.tree
self.repo.tree = WorkingTree(self.repo.root_dir)
try:
if out.changed_cache():
raise FileNotFoundError
finally:
self.repo.tree = saved_tree
return open(out.cache_path, mode=mode, encoding=encoding)
def _ls_files_repo(path_info, recursive=None):
from dvc.compat import fspath
from dvc.ignore import CleanTree
from dvc.path_info import PathInfo
from dvc.scm.tree import WorkingTree
if not os.path.exists(fspath(path_info)):
return []
files = []
tree = CleanTree(WorkingTree(path_info))
try:
for dirpath, dirnames, filenames in tree.walk(path_info):
files.extend(PathInfo(dirpath, f) for f in filenames)
if not recursive:
files.extend(PathInfo(dirpath, d) for d in dirnames)
break
except NotADirectoryError:
if os.path.isfile(fspath(path_info)):
files = [path_info]
return [_get_fs_node(f) for f in files]
def _ls_files_repo(target_path_info, recursive=None):
from dvc.compat import fspath
from dvc.ignore import CleanTree
from dvc.path_info import PathInfo
from dvc.scm.tree import WorkingTree
if not os.path.exists(fspath(target_path_info)):
return []
files = []
tree = CleanTree(WorkingTree(target_path_info))
try:
for dirpath, dirnames, filenames in tree.walk(target_path_info):
files.extend(map(lambda f: PathInfo(dirpath, f), filenames))
if not recursive:
files.extend(map(lambda d: PathInfo(dirpath, d), dirnames))
break
except NotADirectoryError:
if os.path.isfile(fspath(target_path_info)):
return [target_path_info]
return files
def open(self, path, mode="r", encoding="utf-8", remote=None):
try:
outs = self._find_outs(path, strict=False)
except OutputNotFoundError as exc:
raise FileNotFoundError from exc
if len(outs) != 1 or outs[0].is_dir_checksum:
raise IsADirectoryError
out = outs[0]
# temporary hack to make cache use WorkingTree and not GitTree, because
# cache dir doesn't exist in the latter.
saved_tree = self.repo.tree
self.repo.tree = WorkingTree(self.repo.root_dir)
try:
if out.changed_cache():
if not self.fetch and not self.stream:
raise FileNotFoundError
remote_obj = self.repo.cloud.get_remote(remote)
if self.stream:
try:
remote_info = remote_obj.checksum_to_path_info(
out.checksum)
return remote_obj.open(remote_info,
mode=mode,
encoding=encoding)
except RemoteActionNotImplemented:
pass
with self.repo.state:
cache_info = out.get_used_cache(remote=remote)
self.repo.cloud.pull(cache_info, remote=remote)
finally:
self.repo.tree = saved_tree
return open(out.cache_path, mode=mode, encoding=encoding)
def __init__(
self,
dvc_dir=None,
validate=True,
tree=None,
): # pylint: disable=super-init-not-called
from dvc.scm.tree import WorkingTree
self.dvc_dir = dvc_dir
if not dvc_dir:
try:
from dvc.repo import Repo
self.dvc_dir = os.path.join(Repo.find_dvc_dir())
except NotDvcRepoError:
self.dvc_dir = None
else:
self.dvc_dir = os.path.abspath(os.path.realpath(dvc_dir))
self.wtree = WorkingTree(self.dvc_dir)
self.tree = tree.tree if tree else self.wtree
self.load(validate=validate)
class Repo(object):
DVC_DIR = ".dvc"
from dvc.repo.destroy import destroy
from dvc.repo.install import install
from dvc.repo.add import add
from dvc.repo.remove import remove
from dvc.repo.lock import lock as lock_stage
from dvc.repo.move import move
from dvc.repo.run import run
from dvc.repo.imp import imp
from dvc.repo.imp_url import imp_url
from dvc.repo.reproduce import reproduce
from dvc.repo.checkout import _checkout
from dvc.repo.push import push
from dvc.repo.fetch import _fetch
from dvc.repo.pull import pull
from dvc.repo.status import status
from dvc.repo.gc import gc
from dvc.repo.commit import commit
from dvc.repo.diff import diff
from dvc.repo.brancher import brancher
from dvc.repo.get import get
from dvc.repo.get_url import get_url
from dvc.repo.update import update
def __init__(self, root_dir=None):
from dvc.state import State
from dvc.lock import Lock
from dvc.scm import SCM
from dvc.cache import Cache
from dvc.data_cloud import DataCloud
from dvc.repo.metrics import Metrics
from dvc.scm.tree import WorkingTree
from dvc.repo.tag import Tag
root_dir = self.find_root(root_dir)
self.root_dir = os.path.abspath(os.path.realpath(root_dir))
self.dvc_dir = os.path.join(self.root_dir, self.DVC_DIR)
self.config = Config(self.dvc_dir)
self.scm = SCM(self.root_dir)
self.tree = WorkingTree(self.root_dir)
self.lock = Lock(
os.path.join(self.dvc_dir, "lock"),
tmp_dir=os.path.join(self.dvc_dir, "tmp"),
)
# NOTE: storing state and link_state in the repository itself to avoid
# any possible state corruption in 'shared cache dir' scenario.
self.state = State(self, self.config.config)
core = self.config.config[Config.SECTION_CORE]
level = core.get(Config.SECTION_CORE_LOGLEVEL)
if level:
logger.setLevel(level.upper())
self.cache = Cache(self)
self.cloud = DataCloud(self)
self.metrics = Metrics(self)
self.tag = Tag(self)
self._ignore()
@property
def tree(self):
return self._tree
@tree.setter
def tree(self, tree):
self._tree = tree
# Our graph cache is no longer valid, as it was based on the previous
# tree.
self._reset()
def __repr__(self):
return "Repo: '{root_dir}'".format(root_dir=self.root_dir)
@classmethod
def find_root(cls, root=None):
if root is None:
root = os.getcwd()
else:
root = os.path.abspath(os.path.realpath(root))
while True:
dvc_dir = os.path.join(root, cls.DVC_DIR)
if os.path.isdir(dvc_dir):
return root
if os.path.ismount(root):
break
root = os.path.dirname(root)
raise NotDvcRepoError(root)
@classmethod
def find_dvc_dir(cls, root=None):
root_dir = cls.find_root(root)
return os.path.join(root_dir, cls.DVC_DIR)
@staticmethod
def init(root_dir=os.curdir, no_scm=False, force=False):
from dvc.repo.init import init
init(root_dir=root_dir, no_scm=no_scm, force=force)
return Repo(root_dir)
def unprotect(self, target):
return self.cache.local.unprotect(PathInfo(target))
def _ignore(self):
from dvc.updater import Updater
updater = Updater(self.dvc_dir)
flist = ([self.config.config_local_file, updater.updater_file] +
self.state.files + self.lock.files + updater.lock.files)
if self.cache.local.cache_dir.startswith(self.root_dir + os.sep):
flist += [self.cache.local.cache_dir]
self.scm.ignore_list(flist)
def check_modified_graph(self, new_stages):
"""Generate graph including the new stage to check for errors"""
self._collect_graph(self.stages + new_stages)
def collect(self, target, with_deps=False, recursive=False, graph=None):
import networkx as nx
from dvc.stage import Stage
G = graph or self.graph
if not target:
return get_stages(G)
target = os.path.abspath(target)
if recursive and os.path.isdir(target):
attrs = nx.get_node_attributes(G, "stage")
nodes = [node for node in nx.dfs_postorder_nodes(G)]
ret = []
for node in nodes:
stage = attrs[node]
if stage.path.startswith(target + os.sep):
ret.append(stage)
return ret
stage = Stage.load(self, target)
if not with_deps:
return [stage]
node = relpath(stage.path, self.root_dir)
pipeline = get_pipeline(get_pipelines(G), node)
return [
pipeline.node[n]["stage"]
for n in nx.dfs_postorder_nodes(pipeline, node)
]
def used_cache(
self,
targets=None,
all_branches=False,
with_deps=False,
all_tags=False,
all_commits=False,
remote=None,
force=False,
jobs=None,
recursive=False,
):
"""Get the stages related to the given target and collect
the `info` of its outputs.
This is useful to know what files from the cache are _in use_
(namely, a file described as an output on a stage).
The scope is, by default, the working directory, but you can use
`all_branches` or `all_tags` to expand scope.
Returns:
A dictionary with Schemes (representing output's location) as keys,
and a list with the outputs' `dumpd` as values.
"""
from dvc.cache import NamedCache
cache = NamedCache()
for branch in self.brancher(
all_branches=all_branches,
all_tags=all_tags,
all_commits=all_commits,
):
if targets:
stages = []
for target in targets:
collected = self.collect(target,
recursive=recursive,
with_deps=with_deps)
stages.extend(collected)
else:
stages = self.stages
for stage in stages:
if stage.is_repo_import:
dep, = stage.deps
cache.external[dep.repo_pair].add(dep.def_path)
continue
for out in stage.outs:
used_cache = out.get_used_cache(remote=remote,
force=force,
jobs=jobs)
suffix = "({})".format(branch) if branch else ""
cache.update(used_cache, suffix=suffix)
return cache
def _collect_graph(self, stages=None):
"""Generate a graph by using the given stages on the given directory
The nodes of the graph are the stage's path relative to the root.
Edges are created when the output of one stage is used as a
dependency in other stage.
The direction of the edges goes from the stage to its dependency:
For example, running the following:
$ dvc run -o A "echo A > A"
$ dvc run -d A -o B "echo B > B"
$ dvc run -d B -o C "echo C > C"
Will create the following graph:
ancestors <--
|
C.dvc -> B.dvc -> A.dvc
| |
| --> descendants
|
------- pipeline ------>
|
v
(weakly connected components)
Args:
stages (list): used to build a graph, if None given, collect stages
in the repository.
Raises:
OutputDuplicationError: two outputs with the same path
StagePathAsOutputError: stage inside an output directory
OverlappingOutputPathsError: output inside output directory
CyclicGraphError: resulting graph has cycles
"""
import networkx as nx
from dvc.exceptions import (
OutputDuplicationError,
StagePathAsOutputError,
OverlappingOutputPathsError,
)
G = nx.DiGraph()
stages = stages or self.collect_stages()
stages = [stage for stage in stages if stage]
outs = {}
for stage in stages:
for out in stage.outs:
if out.path_info in outs:
stages = [stage.relpath, outs[out.path_info].stage.relpath]
raise OutputDuplicationError(str(out), stages)
outs[out.path_info] = out
for stage in stages:
for out in stage.outs:
for p in out.path_info.parents:
if p in outs:
raise OverlappingOutputPathsError(outs[p], out)
for stage in stages:
stage_path_info = PathInfo(stage.path)
for p in chain([stage_path_info], stage_path_info.parents):
if p in outs:
raise StagePathAsOutputError(stage.wdir, stage.relpath)
for stage in stages:
node = relpath(stage.path, self.root_dir)
G.add_node(node, stage=stage)
for dep in stage.deps:
if dep.path_info is None:
continue
for out in outs:
if (out == dep.path_info or dep.path_info.isin(out)
or out.isin(dep.path_info)):
dep_stage = outs[out].stage
dep_node = relpath(dep_stage.path, self.root_dir)
G.add_node(dep_node, stage=dep_stage)
G.add_edge(node, dep_node)
check_acyclic(G)
return G
@cached_property
def graph(self):
return self._collect_graph()
@cached_property
def pipelines(self):
return get_pipelines(self.graph)
@staticmethod
def _filter_out_dirs(dirs, outs, root_dir):
def filter_dirs(dname):
path = os.path.join(root_dir, dname)
for out in outs:
if path == os.path.normpath(out):
return False
return True
return list(filter(filter_dirs, dirs))
def collect_stages(self):
"""
Walks down the root directory looking for Dvcfiles,
skipping the directories that are related with
any SCM (e.g. `.git`), DVC itself (`.dvc`), or directories
tracked by DVC (e.g. `dvc add data` would skip `data/`)
NOTE: For large repos, this could be an expensive
operation. Consider using some memoization.
"""
from dvc.stage import Stage
stages = []
outs = []
for root, dirs, files in self.tree.walk(self.root_dir,
dvcignore=self.dvcignore):
for fname in files:
path = os.path.join(root, fname)
if not Stage.is_valid_filename(path):
continue
stage = Stage.load(self, path)
for out in stage.outs:
if out.scheme == "local":
outs.append(out.fspath + out.sep)
stages.append(stage)
dirs[:] = self._filter_out_dirs(dirs, outs, root)
return stages
@cached_property
def stages(self):
return get_stages(self.graph)
def find_outs_by_path(self, path, outs=None, recursive=False):
if not outs:
outs = [out for stage in self.stages for out in stage.outs]
abs_path = os.path.abspath(path)
is_dir = self.tree.isdir(abs_path)
def func(out):
if out.scheme == "local" and out.fspath == abs_path:
return True
if is_dir and recursive and out.path_info.isin(abs_path):
return True
return False
matched = list(filter(func, outs))
if not matched:
raise OutputNotFoundError(path)
return matched
def find_out_by_relpath(self, relpath):
path = os.path.join(self.root_dir, relpath)
out, = self.find_outs_by_path(path)
return out
def is_dvc_internal(self, path):
path_parts = os.path.normpath(path).split(os.path.sep)
return self.DVC_DIR in path_parts
@contextmanager
def open(self, path, remote=None, mode="r", encoding=None):
"""Opens a specified resource as a file descriptor"""
cause = None
try:
out, = self.find_outs_by_path(path)
except OutputNotFoundError as e:
out = None
cause = e
if out and out.use_cache:
try:
with self._open_cached(out, remote, mode, encoding) as fd:
yield fd
return
except FileNotFoundError as e:
raise FileMissingError(relpath(path, self.root_dir), cause=e)
if self.tree.exists(path):
with self.tree.open(path, mode, encoding) as fd:
yield fd
return
raise FileMissingError(relpath(path, self.root_dir), cause=cause)
def _open_cached(self, out, remote=None, mode="r", encoding=None):
if out.isdir():
raise ValueError("Can't open a dir")
cache_file = self.cache.local.checksum_to_path_info(out.checksum)
cache_file = fspath_py35(cache_file)
if os.path.exists(cache_file):
return _open(cache_file, mode=mode, encoding=encoding)
try:
remote_obj = self.cloud.get_remote(remote)
remote_info = remote_obj.checksum_to_path_info(out.checksum)
return remote_obj.open(remote_info, mode=mode, encoding=encoding)
except RemoteActionNotImplemented:
with self.state:
cache_info = out.get_used_cache(remote=remote)
self.cloud.pull(cache_info, remote=remote)
return _open(cache_file, mode=mode, encoding=encoding)
@cached_property
def dvcignore(self):
return DvcIgnoreFilter(self.root_dir)
def close(self):
self.scm.close()
@locked
def checkout(self, *args, **kwargs):
return self._checkout(*args, **kwargs)
@locked
def fetch(self, *args, **kwargs):
return self._fetch(*args, **kwargs)
def _reset(self):
self.__dict__.pop("graph", None)
self.__dict__.pop("stages", None)
self.__dict__.pop("pipelines", None)
self.__dict__.pop("dvcignore", None)
class Repo(object):
DVC_DIR = ".dvc"
from dvc.repo.destroy import destroy
from dvc.repo.install import install
from dvc.repo.add import add
from dvc.repo.remove import remove
from dvc.repo.lock import lock as lock_stage
from dvc.repo.move import move
from dvc.repo.run import run
from dvc.repo.imp import imp
from dvc.repo.reproduce import reproduce
from dvc.repo.checkout import checkout
from dvc.repo.push import push
from dvc.repo.fetch import fetch
from dvc.repo.pull import pull
from dvc.repo.status import status
from dvc.repo.gc import gc
from dvc.repo.commit import commit
from dvc.repo.diff import diff
from dvc.repo.brancher import brancher
def __init__(self, root_dir=None):
from dvc.state import State
from dvc.lock import Lock
from dvc.scm import SCM
from dvc.cache import Cache
from dvc.data_cloud import DataCloud
from dvc.repo.metrics import Metrics
from dvc.scm.tree import WorkingTree
from dvc.repo.tag import Tag
from dvc.repo.pkg import Pkg
root_dir = self.find_root(root_dir)
self.root_dir = os.path.abspath(os.path.realpath(root_dir))
self.dvc_dir = os.path.join(self.root_dir, self.DVC_DIR)
self.config = Config(self.dvc_dir)
self.tree = WorkingTree(self.root_dir)
self.scm = SCM(self.root_dir, repo=self)
self.lock = Lock(self.dvc_dir)
# NOTE: storing state and link_state in the repository itself to avoid
# any possible state corruption in 'shared cache dir' scenario.
self.state = State(self, self.config.config)
core = self.config.config[Config.SECTION_CORE]
level = core.get(Config.SECTION_CORE_LOGLEVEL)
if level:
logger.setLevel(level.upper())
self.cache = Cache(self)
self.cloud = DataCloud(self)
self.metrics = Metrics(self)
self.tag = Tag(self)
self.pkg = Pkg(self)
self._ignore()
def __repr__(self):
return "Repo: '{root_dir}'".format(root_dir=self.root_dir)
@classmethod
def find_root(cls, root=None):
if root is None:
root = os.getcwd()
else:
root = os.path.abspath(os.path.realpath(root))
while True:
dvc_dir = os.path.join(root, cls.DVC_DIR)
if os.path.isdir(dvc_dir):
return root
if os.path.ismount(root):
break
root = os.path.dirname(root)
raise NotDvcRepoError(root)
@classmethod
def find_dvc_dir(cls, root=None):
root_dir = cls.find_root(root)
return os.path.join(root_dir, cls.DVC_DIR)
@staticmethod
def init(root_dir=os.curdir, no_scm=False, force=False):
from dvc.repo.init import init
init(root_dir=root_dir, no_scm=no_scm, force=force)
return Repo(root_dir)
def unprotect(self, target):
return self.cache.local.unprotect(PathInfo(target))
def _ignore(self):
from dvc.updater import Updater
updater = Updater(self.dvc_dir)
flist = [
self.state.state_file,
self.lock.lock_file,
self.config.config_local_file,
updater.updater_file,
updater.lock.lock_file,
] + self.state.temp_files
if self.cache.local.cache_dir.startswith(self.root_dir):
flist += [self.cache.local.cache_dir]
self.scm.ignore_list(flist)
def check_dag(self, stages):
"""Generate graph including the new stage to check for errors"""
self.graph(stages=stages)
@staticmethod
def _check_cyclic_graph(graph):
import networkx as nx
from dvc.exceptions import CyclicGraphError
cycles = list(nx.simple_cycles(graph))
if cycles:
raise CyclicGraphError(cycles[0])
def _get_pipeline(self, node):
pipelines = [i for i in self.pipelines() if i.has_node(node)]
assert len(pipelines) == 1
return pipelines[0]
def collect(self, target, with_deps=False, recursive=False):
import networkx as nx
from dvc.stage import Stage
if not target or (recursive and os.path.isdir(target)):
return self.active_stages(target)
stage = Stage.load(self, target)
if not with_deps:
return [stage]
node = os.path.relpath(stage.path, self.root_dir)
G = self._get_pipeline(node)
ret = []
for n in nx.dfs_postorder_nodes(G, node):
ret.append(G.node[n]["stage"])
return ret
def _collect_dir_cache(self,
out,
branch=None,
remote=None,
force=False,
jobs=None):
"""Get a list of `info`s retaled to the given directory.
- Pull the directory entry from the remote cache if it was changed.
Example:
Given the following commands:
$ echo "foo" > directory/foo
$ echo "bar" > directory/bar
$ dvc add directory
It will return something similar to the following list:
[
{ 'path': 'directory', 'md5': '168fd6761b9c.dir', ... },
{ 'path': 'directory/foo', 'md5': 'c157a79031e1', ... },
{ 'path': 'directory/bar', 'md5': 'd3b07384d113', ... },
]
"""
info = out.dumpd()
ret = [info]
r = out.remote
md5 = info[r.PARAM_CHECKSUM]
if self.cache.local.changed_cache_file(md5):
try:
self.cloud.pull(ret,
jobs=jobs,
remote=remote,
show_checksums=False)
except DvcException as exc:
msg = "Failed to pull cache for '{}': {}"
logger.debug(msg.format(out, exc))
if self.cache.local.changed_cache_file(md5):
msg = ("Missing cache for directory '{}'. "
"Cache for files inside will be lost. "
"Would you like to continue? Use '-f' to force. ")
if not force and not prompt.confirm(msg):
raise DvcException(
"unable to fully collect used cache"
" without cache for directory '{}'".format(out))
else:
return ret
for i in out.dir_cache:
i["branch"] = branch
i[r.PARAM_PATH] = os.path.join(info[r.PARAM_PATH],
i[r.PARAM_RELPATH])
ret.append(i)
return ret
def _collect_used_cache(self,
out,
branch=None,
remote=None,
force=False,
jobs=None):
"""Get a dumpd of the given `out`, with an entry including the branch.
The `used_cache` of an output is no more than its `info`.
In case that the given output is a directory, it will also
include the `info` of its files.
"""
if not out.use_cache or not out.info:
if not out.info:
logger.warning("Output '{}'({}) is missing version "
"info. Cache for it will not be collected. "
"Use dvc repro to get your pipeline up to "
"date.".format(out, out.stage))
return []
info = out.dumpd()
info["branch"] = branch
ret = [info]
if out.scheme != "local":
return ret
if not out.is_dir_checksum:
return ret
return self._collect_dir_cache(out,
branch=branch,
remote=remote,
force=force,
jobs=jobs)
def used_cache(
self,
target=None,
all_branches=False,
active=True,
with_deps=False,
all_tags=False,
remote=None,
force=False,
jobs=None,
recursive=False,
):
"""Get the stages related to the given target and collect
the `info` of its outputs.
This is useful to know what files from the cache are _in use_
(namely, a file described as an output on a stage).
The scope is, by default, the working directory, but you can use
`all_branches` or `all_tags` to expand scope.
Returns:
A dictionary with Schemes (representing output's location) as keys,
and a list with the outputs' `dumpd` as values.
"""
cache = {}
cache["local"] = []
cache["s3"] = []
cache["gs"] = []
cache["hdfs"] = []
cache["ssh"] = []
cache["azure"] = []
for branch in self.brancher(all_branches=all_branches,
all_tags=all_tags):
if target:
if recursive and os.path.isdir(target):
stages = self.stages(target)
else:
stages = self.collect(target, with_deps=with_deps)
elif active:
stages = self.active_stages()
else:
stages = self.stages()
for stage in stages:
if active and not target and stage.locked:
logger.warning(
"DVC-file '{path}' is locked. Its dependencies are"
" not going to be pushed/pulled/fetched.".format(
path=stage.relpath))
for out in stage.outs:
scheme = out.path_info.scheme
cache[scheme].extend(
self._collect_used_cache(
out,
branch=branch,
remote=remote,
force=force,
jobs=jobs,
))
return cache
def graph(self, stages=None, from_directory=None):
"""Generate a graph by using the given stages on the given directory
The nodes of the graph are the stage's path relative to the root.
Edges are created when the output of one stage is used as a
dependency in other stage.
The direction of the edges goes from the stage to its dependency:
For example, running the following:
$ dvc run -o A "echo A > A"
$ dvc run -d A -o B "echo B > B"
$ dvc run -d B -o C "echo C > C"
Will create the following graph:
ancestors <--
|
C.dvc -> B.dvc -> A.dvc
| |
| --> descendants
|
------- pipeline ------>
|
v
(weakly connected components)
Args:
stages (list): used to build a graph, if None given, use the ones
on the `from_directory`.
from_directory (str): directory where to look at for stages, if
None is given, use the current working directory
Raises:
OutputDuplicationError: two outputs with the same path
StagePathAsOutputError: stage inside an output directory
OverlappingOutputPathsError: output inside output directory
CyclicGraphError: resulting graph has cycles
"""
import networkx as nx
from dvc.exceptions import (
OutputDuplicationError,
StagePathAsOutputError,
OverlappingOutputPathsError,
)
G = nx.DiGraph()
G_active = nx.DiGraph()
stages = stages or self.stages(from_directory, check_dag=False)
stages = [stage for stage in stages if stage]
outs = []
for stage in stages:
for out in stage.outs:
existing = []
for o in outs:
if o.path_info == out.path_info:
existing.append(o.stage)
in_o_dir = out.path_info.isin(o.path_info)
in_out_dir = o.path_info.isin(out.path_info)
if in_o_dir or in_out_dir:
raise OverlappingOutputPathsError(o, out)
if existing:
stages = [stage.relpath, existing[0].relpath]
raise OutputDuplicationError(str(out), stages)
outs.append(out)
for stage in stages:
stage_path_info = PathInfo(stage.path)
for out in outs:
if stage_path_info.isin(out.path_info):
raise StagePathAsOutputError(stage.wdir, stage.relpath)
for stage in stages:
node = os.path.relpath(stage.path, self.root_dir)
G.add_node(node, stage=stage)
G_active.add_node(node, stage=stage)
for dep in stage.deps:
for out in outs:
if (out.path_info != dep.path_info
and not dep.path_info.isin(out.path_info)
and not out.path_info.isin(dep.path_info)):
continue
dep_stage = out.stage
dep_node = os.path.relpath(dep_stage.path, self.root_dir)
G.add_node(dep_node, stage=dep_stage)
G.add_edge(node, dep_node)
if not stage.locked:
G_active.add_node(dep_node, stage=dep_stage)
G_active.add_edge(node, dep_node)
self._check_cyclic_graph(G)
return G, G_active
def pipelines(self, from_directory=None):
import networkx as nx
G, G_active = self.graph(from_directory=from_directory)
return [
G.subgraph(c).copy() for c in nx.weakly_connected_components(G)
]
def stages(self, from_directory=None, check_dag=True):
"""
Walks down the root directory looking for Dvcfiles,
skipping the directories that are related with
any SCM (e.g. `.git`), DVC itself (`.dvc`), or directories
tracked by DVC (e.g. `dvc add data` would skip `data/`)
NOTE: For large repos, this could be an expensive
operation. Consider using some memoization.
"""
from dvc.stage import Stage
if not from_directory:
from_directory = self.root_dir
elif not os.path.isdir(from_directory):
raise TargetNotDirectoryError(from_directory)
stages = []
outs = []
ignore_file_handler = DvcIgnoreFileHandler(self.tree)
for root, dirs, files in self.tree.walk(
from_directory, ignore_file_handler=ignore_file_handler):
for fname in files:
path = os.path.join(root, fname)
if not Stage.is_valid_filename(path):
continue
stage = Stage.load(self, path)
for out in stage.outs:
if out.scheme == "local":
outs.append(out.fspath + out.sep)
stages.append(stage)
def filter_dirs(dname, root=root):
path = os.path.join(root, dname)
if path in (self.dvc_dir, self.scm.dir):
return False
for out in outs:
if path == os.path.normpath(out) or path.startswith(out):
return False
return True
dirs[:] = list(filter(filter_dirs, dirs))
if check_dag:
self.check_dag(stages)
return stages
def active_stages(self, from_directory=None):
import networkx as nx
stages = []
for G in self.pipelines(from_directory):
stages.extend(list(nx.get_node_attributes(G, "stage").values()))
return stages
def find_outs_by_path(self, path, outs=None, recursive=False):
if not outs:
# there is no `from_directory=path` argument because some data
# files might be generated to an upper level, and so it is
# needed to look at all the files (project root_dir)
stages = self.stages()
outs = [out for stage in stages for out in stage.outs]
abs_path = os.path.abspath(path)
is_dir = self.tree.isdir(abs_path)
def func(out):
if out.scheme == "local" and out.fspath == abs_path:
return True
if is_dir and recursive and out.path_info.isin(abs_path):
return True
return False
matched = list(filter(func, outs))
if not matched:
raise OutputNotFoundError(path)
return matched
def is_dvc_internal(self, path):
path_parts = os.path.normpath(path).split(os.path.sep)
return self.DVC_DIR in path_parts
def open(self, path, remote=None, mode="r", encoding=None):
"""Opens a specified resource as a file descriptor"""
out, = self.find_outs_by_path(path)
if out.isdir():
raise ValueError("Can't open a dir")
with self.state:
cache_info = self._collect_used_cache(out, remote=remote)
self.cloud.pull(cache_info, remote=remote)
cache_file = self.cache.local.checksum_to_path_info(out.checksum)
return _open(cache_file.fspath, mode=mode, encoding=encoding)
class Repo(object):
DVC_DIR = ".dvc"
from dvc.repo.destroy import destroy
from dvc.repo.install import install
from dvc.repo.add import add
from dvc.repo.remove import remove
from dvc.repo.ls import ls
from dvc.repo.lock import lock as lock_stage
from dvc.repo.move import move
from dvc.repo.run import run
from dvc.repo.imp import imp
from dvc.repo.imp_url import imp_url
from dvc.repo.reproduce import reproduce
from dvc.repo.checkout import _checkout
from dvc.repo.push import push
from dvc.repo.fetch import _fetch
from dvc.repo.pull import pull
from dvc.repo.status import status
from dvc.repo.gc import gc
from dvc.repo.commit import commit
from dvc.repo.diff import diff
from dvc.repo.brancher import brancher
from dvc.repo.get import get
from dvc.repo.get_url import get_url
from dvc.repo.update import update
def __init__(self, root_dir=None):
from dvc.state import State
from dvc.lock import make_lock
from dvc.scm import SCM
from dvc.cache import Cache
from dvc.data_cloud import DataCloud
from dvc.repo.metrics import Metrics
from dvc.scm.tree import WorkingTree
from dvc.repo.tag import Tag
from dvc.utils.fs import makedirs
root_dir = self.find_root(root_dir)
self.root_dir = os.path.abspath(os.path.realpath(root_dir))
self.dvc_dir = os.path.join(self.root_dir, self.DVC_DIR)
self.config = Config(self.dvc_dir)
no_scm = self.config["core"].get("no_scm", False)
self.scm = SCM(self.root_dir, no_scm=no_scm)
self.tree = WorkingTree(self.root_dir)
self.tmp_dir = os.path.join(self.dvc_dir, "tmp")
makedirs(self.tmp_dir, exist_ok=True)
hardlink_lock = self.config["core"].get("hardlink_lock", False)
self.lock = make_lock(
os.path.join(self.dvc_dir, "lock"),
tmp_dir=os.path.join(self.dvc_dir, "tmp"),
hardlink_lock=hardlink_lock,
friendly=True,
)
# NOTE: storing state and link_state in the repository itself to avoid
# any possible state corruption in 'shared cache dir' scenario.
self.state = State(self)
self.cache = Cache(self)
self.cloud = DataCloud(self)
self.metrics = Metrics(self)
self.tag = Tag(self)
self._ignore()
@property
def tree(self):
return self._tree
@tree.setter
def tree(self, tree):
self._tree = tree if isinstance(tree, CleanTree) else CleanTree(tree)
# Our graph cache is no longer valid, as it was based on the previous
# tree.
self._reset()
def __repr__(self):
return "{}: '{}'".format(self.__class__.__name__, self.root_dir)
@classmethod
def find_root(cls, root=None):
root_dir = os.path.realpath(root or os.curdir)
if not os.path.isdir(root_dir):
raise NotDvcRepoError("directory '{}' does not exist".format(root))
while True:
dvc_dir = os.path.join(root_dir, cls.DVC_DIR)
if os.path.isdir(dvc_dir):
return root_dir
if os.path.ismount(root_dir):
break
root_dir = os.path.dirname(root_dir)
message = ("you are not inside of a DVC repository "
"(checked up to mount point '{}')").format(root_dir)
raise NotDvcRepoError(message)
@classmethod
def find_dvc_dir(cls, root=None):
root_dir = cls.find_root(root)
return os.path.join(root_dir, cls.DVC_DIR)
@staticmethod
def init(root_dir=os.curdir, no_scm=False, force=False, subdir=False):
from dvc.repo.init import init
init(root_dir=root_dir, no_scm=no_scm, force=force, subdir=subdir)
return Repo(root_dir)
def unprotect(self, target):
return self.cache.local.unprotect(PathInfo(target))
def _ignore(self):
from dvc.updater import Updater
updater = Updater(self.dvc_dir)
flist = ([self.config.files["local"], updater.updater_file] +
[self.lock.lockfile, updater.lock.lockfile, self.tmp_dir] +
self.state.files)
if path_isin(self.cache.local.cache_dir, self.root_dir):
flist += [self.cache.local.cache_dir]
self.scm.ignore_list(flist)
def check_modified_graph(self, new_stages):
"""Generate graph including the new stage to check for errors"""
self._collect_graph(self.stages + new_stages)
def collect(self, target, with_deps=False, recursive=False, graph=None):
import networkx as nx
from dvc.stage import Stage
G = graph or self.graph
if not target:
return list(G)
target = os.path.abspath(target)
if recursive and os.path.isdir(target):
stages = nx.dfs_postorder_nodes(G)
return [stage for stage in stages if path_isin(stage.path, target)]
stage = Stage.load(self, target)
if not with_deps:
return [stage]
pipeline = get_pipeline(get_pipelines(G), stage)
return list(nx.dfs_postorder_nodes(pipeline, stage))
def collect_granular(self, target, *args, **kwargs):
if not target:
return [(stage, None) for stage in self.stages]
try:
(out, ) = self.find_outs_by_path(target, strict=False)
filter_info = PathInfo(os.path.abspath(target))
return [(out.stage, filter_info)]
except OutputNotFoundError:
stages = self.collect(target, *args, **kwargs)
return [(stage, None) for stage in stages]
def used_cache(
self,
targets=None,
all_branches=False,
with_deps=False,
all_tags=False,
all_commits=False,
remote=None,
force=False,
jobs=None,
recursive=False,
):
"""Get the stages related to the given target and collect
the `info` of its outputs.
This is useful to know what files from the cache are _in use_
(namely, a file described as an output on a stage).
The scope is, by default, the working directory, but you can use
`all_branches` or `all_tags` to expand scope.
Returns:
A dictionary with Schemes (representing output's location) as keys,
and a list with the outputs' `dumpd` as values.
"""
from dvc.cache import NamedCache
cache = NamedCache()
for branch in self.brancher(
all_branches=all_branches,
all_tags=all_tags,
all_commits=all_commits,
):
targets = targets or [None]
pairs = cat(
self.collect_granular(
target, recursive=recursive, with_deps=with_deps)
for target in targets)
suffix = "({})".format(branch) if branch else ""
for stage, filter_info in pairs:
used_cache = stage.get_used_cache(
remote=remote,
force=force,
jobs=jobs,
filter_info=filter_info,
)
cache.update(used_cache, suffix=suffix)
return cache
def _collect_graph(self, stages=None):
"""Generate a graph by using the given stages on the given directory
The nodes of the graph are the stage's path relative to the root.
Edges are created when the output of one stage is used as a
dependency in other stage.
The direction of the edges goes from the stage to its dependency:
For example, running the following:
$ dvc run -o A "echo A > A"
$ dvc run -d A -o B "echo B > B"
$ dvc run -d B -o C "echo C > C"
Will create the following graph:
ancestors <--
|
C.dvc -> B.dvc -> A.dvc
| |
| --> descendants
|
------- pipeline ------>
|
v
(weakly connected components)
Args:
stages (list): used to build a graph, if None given, collect stages
in the repository.
Raises:
OutputDuplicationError: two outputs with the same path
StagePathAsOutputError: stage inside an output directory
OverlappingOutputPathsError: output inside output directory
CyclicGraphError: resulting graph has cycles
"""
import networkx as nx
from pygtrie import Trie
from dvc.exceptions import (
OutputDuplicationError,
StagePathAsOutputError,
OverlappingOutputPathsError,
)
G = nx.DiGraph()
stages = stages or self.stages
stages = [stage for stage in stages if stage]
outs = Trie() # Use trie to efficiently find overlapping outs and deps
for stage in stages:
for out in stage.outs:
out_key = out.path_info.parts
# Check for dup outs
if out_key in outs:
dup_stages = [stage, outs[out_key].stage]
raise OutputDuplicationError(str(out), dup_stages)
# Check for overlapping outs
if outs.has_subtrie(out_key):
parent = out
overlapping = first(outs.values(prefix=out_key))
else:
parent = outs.shortest_prefix(out_key).value
overlapping = out
if parent and overlapping:
msg = ("Paths for outs:\n'{}'('{}')\n'{}'('{}')\n"
"overlap. To avoid unpredictable behaviour, "
"rerun command with non overlapping outs paths."
).format(
str(parent),
parent.stage.relpath,
str(overlapping),
overlapping.stage.relpath,
)
raise OverlappingOutputPathsError(parent, overlapping, msg)
outs[out_key] = out
for stage in stages:
out = outs.shortest_prefix(PathInfo(stage.path).parts).value
if out:
raise StagePathAsOutputError(stage, str(out))
# Building graph
G.add_nodes_from(stages)
for stage in stages:
for dep in stage.deps:
if dep.path_info is None:
continue
dep_key = dep.path_info.parts
overlapping = list(n.value for n in outs.prefixes(dep_key))
if outs.has_subtrie(dep_key):
overlapping.extend(outs.values(prefix=dep_key))
G.add_edges_from((stage, out.stage) for out in overlapping)
check_acyclic(G)
return G
@cached_property
def graph(self):
return self._collect_graph()
@cached_property
def pipelines(self):
return get_pipelines(self.graph)
@cached_property
def stages(self):
"""
Walks down the root directory looking for Dvcfiles,
skipping the directories that are related with
any SCM (e.g. `.git`), DVC itself (`.dvc`), or directories
tracked by DVC (e.g. `dvc add data` would skip `data/`)
NOTE: For large repos, this could be an expensive
operation. Consider using some memoization.
"""
from dvc.stage import Stage
stages = []
outs = set()
for root, dirs, files in self.tree.walk(self.root_dir):
for fname in files:
path = os.path.join(root, fname)
if not Stage.is_valid_filename(path):
continue
stage = Stage.load(self, path)
stages.append(stage)
for out in stage.outs:
if out.scheme == "local":
outs.add(out.fspath)
dirs[:] = [d for d in dirs if os.path.join(root, d) not in outs]
return stages
def find_outs_by_path(self, path, outs=None, recursive=False, strict=True):
if not outs:
outs = [out for stage in self.stages for out in stage.outs]
abs_path = os.path.abspath(path)
path_info = PathInfo(abs_path)
match = path_info.__eq__ if strict else path_info.isin_or_eq
def func(out):
if out.scheme == "local" and match(out.path_info):
return True
if recursive and out.path_info.isin(path_info):
return True
return False
matched = list(filter(func, outs))
if not matched:
raise OutputNotFoundError(path, self)
return matched
def find_out_by_relpath(self, relpath):
path = os.path.join(self.root_dir, relpath)
(out, ) = self.find_outs_by_path(path)
return out
def is_dvc_internal(self, path):
path_parts = os.path.normpath(path).split(os.path.sep)
return self.DVC_DIR in path_parts
@contextmanager
def open_by_relpath(self, path, remote=None, mode="r", encoding=None):
"""Opens a specified resource as a file descriptor"""
cause = None
try:
out = self.find_out_by_relpath(path)
except OutputNotFoundError as exc:
out = None
cause = exc
if out and out.use_cache:
try:
with self._open_cached(out, remote, mode, encoding) as fd:
yield fd
return
except FileNotFoundError as exc:
raise FileMissingError(path) from exc
abs_path = os.path.join(self.root_dir, path)
if os.path.exists(abs_path):
with open(abs_path, mode=mode, encoding=encoding) as fd:
yield fd
return
raise FileMissingError(path) from cause
def _open_cached(self, out, remote=None, mode="r", encoding=None):
if out.isdir():
raise ValueError("Can't open a dir")
cache_file = self.cache.local.checksum_to_path_info(out.checksum)
cache_file = fspath_py35(cache_file)
if os.path.exists(cache_file):
return open(cache_file, mode=mode, encoding=encoding)
try:
remote_obj = self.cloud.get_remote(remote)
remote_info = remote_obj.checksum_to_path_info(out.checksum)
return remote_obj.open(remote_info, mode=mode, encoding=encoding)
except RemoteActionNotImplemented:
with self.state:
cache_info = out.get_used_cache(remote=remote)
self.cloud.pull(cache_info, remote=remote)
return open(cache_file, mode=mode, encoding=encoding)
def close(self):
self.scm.close()
@locked
def checkout(self, *args, **kwargs):
return self._checkout(*args, **kwargs)
@locked
def fetch(self, *args, **kwargs):
return self._fetch(*args, **kwargs)
def _reset(self):
self.__dict__.pop("graph", None)
self.__dict__.pop("stages", None)
self.__dict__.pop("pipelines", None)
self.__dict__.pop("dvcignore", None)
