diff --git a/dulwich/index.py b/dulwich/index.py index c5fc3da1..20c0ed1b 100644 --- a/dulwich/index.py +++ b/dulwich/index.py @@ -1,846 +1,843 @@ # index.py -- File parser/writer for the git index file # Copyright (C) 2008-2013 Jelmer Vernooij # # Dulwich is dual-licensed under the Apache License, Version 2.0 and the GNU # General Public License as public by the Free Software Foundation; version 2.0 # or (at your option) any later version. You can redistribute it and/or # modify it under the terms of either of these two licenses. # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # # You should have received a copy of the licenses; if not, see # for a copy of the GNU General Public License # and for a copy of the Apache # License, Version 2.0. # """Parser for the git index file format.""" import collections import errno import os import stat import struct import sys from dulwich.file import GitFile from dulwich.objects import ( Blob, S_IFGITLINK, S_ISGITLINK, Tree, hex_to_sha, sha_to_hex, ) from dulwich.pack import ( SHA1Reader, SHA1Writer, ) IndexEntry = collections.namedtuple( 'IndexEntry', [ 'ctime', 'mtime', 'dev', 'ino', 'mode', 'uid', 'gid', 'size', 'sha', 'flags']) FLAG_STAGEMASK = 0x3000 FLAG_VALID = 0x8000 FLAG_EXTENDED = 0x4000 def pathsplit(path): """Split a /-delimited path into a directory part and a basename. Args: path: The path to split. Returns: Tuple with directory name and basename """ try: (dirname, basename) = path.rsplit(b"/", 1) except ValueError: return (b"", path) else: return (dirname, basename) def pathjoin(*args): """Join a /-delimited path. """ return b"/".join([p for p in args if p]) def read_cache_time(f): """Read a cache time. Args: f: File-like object to read from Returns: Tuple with seconds and nanoseconds """ return struct.unpack(">LL", f.read(8)) def write_cache_time(f, t): """Write a cache time. Args: f: File-like object to write to t: Time to write (as int, float or tuple with secs and nsecs) """ if isinstance(t, int): t = (t, 0) elif isinstance(t, float): (secs, nsecs) = divmod(t, 1.0) t = (int(secs), int(nsecs * 1000000000)) elif not isinstance(t, tuple): raise TypeError(t) f.write(struct.pack(">LL", *t)) def read_cache_entry(f): """Read an entry from a cache file. Args: f: File-like object to read from Returns: tuple with: device, inode, mode, uid, gid, size, sha, flags """ beginoffset = f.tell() ctime = read_cache_time(f) mtime = read_cache_time(f) (dev, ino, mode, uid, gid, size, sha, flags, ) = \ struct.unpack(">LLLLLL20sH", f.read(20 + 4 * 6 + 2)) name = f.read((flags & 0x0fff)) # Padding: real_size = ((f.tell() - beginoffset + 8) & ~7) f.read((beginoffset + real_size) - f.tell()) return (name, ctime, mtime, dev, ino, mode, uid, gid, size, sha_to_hex(sha), flags & ~0x0fff) def write_cache_entry(f, entry): """Write an index entry to a file. Args: f: File object entry: Entry to write, tuple with: (name, ctime, mtime, dev, ino, mode, uid, gid, size, sha, flags) """ beginoffset = f.tell() (name, ctime, mtime, dev, ino, mode, uid, gid, size, sha, flags) = entry write_cache_time(f, ctime) write_cache_time(f, mtime) flags = len(name) | (flags & ~0x0fff) f.write(struct.pack( b'>LLLLLL20sH', dev & 0xFFFFFFFF, ino & 0xFFFFFFFF, mode, uid, gid, size, hex_to_sha(sha), flags)) f.write(name) real_size = ((f.tell() - beginoffset + 8) & ~7) f.write(b'\0' * ((beginoffset + real_size) - f.tell())) def read_index(f): """Read an index file, yielding the individual entries.""" header = f.read(4) if header != b'DIRC': raise AssertionError("Invalid index file header: %r" % header) (version, num_entries) = struct.unpack(b'>LL', f.read(4 * 2)) assert version in (1, 2) for i in range(num_entries): yield read_cache_entry(f) def read_index_dict(f): """Read an index file and return it as a dictionary. Args: f: File object to read from """ ret = {} for x in read_index(f): ret[x[0]] = IndexEntry(*x[1:]) return ret def write_index(f, entries): """Write an index file. Args: f: File-like object to write to entries: Iterable over the entries to write """ f.write(b'DIRC') f.write(struct.pack(b'>LL', 2, len(entries))) for x in entries: write_cache_entry(f, x) def write_index_dict(f, entries): """Write an index file based on the contents of a dictionary. """ entries_list = [] for name in sorted(entries): entries_list.append((name,) + tuple(entries[name])) write_index(f, entries_list) def cleanup_mode(mode): """Cleanup a mode value. This will return a mode that can be stored in a tree object. Args: mode: Mode to clean up. """ if stat.S_ISLNK(mode): return stat.S_IFLNK elif stat.S_ISDIR(mode): return stat.S_IFDIR elif S_ISGITLINK(mode): return S_IFGITLINK ret = stat.S_IFREG | 0o644 if mode & 0o100: ret |= 0o111 return ret class Index(object): """A Git Index file.""" def __init__(self, filename): """Open an index file. Args: filename: Path to the index file """ self._filename = filename self.clear() self.read() @property def path(self): return self._filename def __repr__(self): return "%s(%r)" % (self.__class__.__name__, self._filename) def write(self): """Write current contents of index to disk.""" f = GitFile(self._filename, 'wb') try: f = SHA1Writer(f) write_index_dict(f, self._byname) finally: f.close() def read(self): """Read current contents of index from disk.""" if not os.path.exists(self._filename): return f = GitFile(self._filename, 'rb') try: f = SHA1Reader(f) for x in read_index(f): self[x[0]] = IndexEntry(*x[1:]) # FIXME: Additional data? f.read(os.path.getsize(self._filename)-f.tell()-20) f.check_sha() finally: f.close() def __len__(self): """Number of entries in this index file.""" return len(self._byname) def __getitem__(self, name): """Retrieve entry by relative path. Returns: tuple with (ctime, mtime, dev, ino, mode, uid, gid, size, sha, flags) """ return self._byname[name] def __iter__(self): """Iterate over the paths in this index.""" return iter(self._byname) def get_sha1(self, path): """Return the (git object) SHA1 for the object at a path.""" return self[path].sha def get_mode(self, path): """Return the POSIX file mode for the object at a path.""" return self[path].mode def iterobjects(self): """Iterate over path, sha, mode tuples for use with commit_tree.""" for path in self: entry = self[path] yield path, entry.sha, cleanup_mode(entry.mode) def iterblobs(self): import warnings warnings.warn('Use iterobjects() instead.', PendingDeprecationWarning) return self.iterobjects() def clear(self): """Remove all contents from this index.""" self._byname = {} def __setitem__(self, name, x): assert isinstance(name, bytes) assert len(x) == 10 # Remove the old entry if any self._byname[name] = IndexEntry(*x) def __delitem__(self, name): assert isinstance(name, bytes) del self._byname[name] def iteritems(self): return self._byname.items() def items(self): return self._byname.items() def update(self, entries): for name, value in entries.items(): self[name] = value def changes_from_tree(self, object_store, tree, want_unchanged=False): """Find the differences between the contents of this index and a tree. Args: object_store: Object store to use for retrieving tree contents tree: SHA1 of the root tree want_unchanged: Whether unchanged files should be reported Returns: Iterator over tuples with (oldpath, newpath), (oldmode, newmode), (oldsha, newsha) """ def lookup_entry(path): entry = self[path] return entry.sha, cleanup_mode(entry.mode) for (name, mode, sha) in changes_from_tree( self._byname.keys(), lookup_entry, object_store, tree, want_unchanged=want_unchanged): yield (name, mode, sha) def commit(self, object_store): """Create a new tree from an index. Args: object_store: Object store to save the tree in Returns: Root tree SHA """ return commit_tree(object_store, self.iterobjects()) def commit_tree(object_store, blobs): """Commit a new tree. Args: object_store: Object store to add trees to blobs: Iterable over blob path, sha, mode entries Returns: SHA1 of the created tree. """ trees = {b'': {}} def add_tree(path): if path in trees: return trees[path] dirname, basename = pathsplit(path) t = add_tree(dirname) assert isinstance(basename, bytes) newtree = {} t[basename] = newtree trees[path] = newtree return newtree for path, sha, mode in blobs: tree_path, basename = pathsplit(path) tree = add_tree(tree_path) tree[basename] = (mode, sha) def build_tree(path): tree = Tree() for basename, entry in trees[path].items(): if isinstance(entry, dict): mode = stat.S_IFDIR sha = build_tree(pathjoin(path, basename)) else: (mode, sha) = entry tree.add(basename, mode, sha) object_store.add_object(tree) return tree.id return build_tree(b'') def commit_index(object_store, index): """Create a new tree from an index. Args: object_store: Object store to save the tree in index: Index file Note: This function is deprecated, use index.commit() instead. Returns: Root tree sha. """ return commit_tree(object_store, index.iterobjects()) def changes_from_tree(names, lookup_entry, object_store, tree, want_unchanged=False): """Find the differences between the contents of a tree and a working copy. Args: names: Iterable of names in the working copy lookup_entry: Function to lookup an entry in the working copy object_store: Object store to use for retrieving tree contents tree: SHA1 of the root tree, or None for an empty tree want_unchanged: Whether unchanged files should be reported Returns: Iterator over tuples with (oldpath, newpath), (oldmode, newmode), (oldsha, newsha) """ # TODO(jelmer): Support a include_trees option other_names = set(names) if tree is not None: for (name, mode, sha) in object_store.iter_tree_contents(tree): try: (other_sha, other_mode) = lookup_entry(name) except KeyError: # Was removed yield ((name, None), (mode, None), (sha, None)) else: other_names.remove(name) if (want_unchanged or other_sha != sha or other_mode != mode): yield ((name, name), (mode, other_mode), (sha, other_sha)) # Mention added files for name in other_names: try: (other_sha, other_mode) = lookup_entry(name) except KeyError: pass else: yield ((None, name), (None, other_mode), (None, other_sha)) def index_entry_from_stat(stat_val, hex_sha, flags, mode=None): """Create a new index entry from a stat value. Args: stat_val: POSIX stat_result instance hex_sha: Hex sha of the object flags: Index flags """ if mode is None: mode = cleanup_mode(stat_val.st_mode) return IndexEntry( stat_val.st_ctime, stat_val.st_mtime, stat_val.st_dev, stat_val.st_ino, mode, stat_val.st_uid, stat_val.st_gid, stat_val.st_size, hex_sha, flags) -def build_file_from_blob(blob, mode, target_path, honor_filemode=True): +def build_file_from_blob(blob, mode, target_path, honor_filemode=True, + tree_encoding='utf-8'): """Build a file or symlink on disk based on a Git object. Args: obj: The git object mode: File mode target_path: Path to write to honor_filemode: An optional flag to honor core.filemode setting in config file, default is core.filemode=True, change executable bit Returns: stat object for the file """ try: oldstat = os.lstat(target_path) except OSError as e: if e.errno == errno.ENOENT: oldstat = None else: raise contents = blob.as_raw_string() if stat.S_ISLNK(mode): # FIXME: This will fail on Windows. What should we do instead? if oldstat: os.unlink(target_path) if sys.platform == 'win32': # os.readlink on Python3 on Windows requires a unicode string. - # TODO(jelmer): Don't assume tree_encoding == fs_encoding - tree_encoding = sys.getfilesystemencoding() contents = contents.decode(tree_encoding) target_path = target_path.decode(tree_encoding) os.symlink(contents, target_path) else: if oldstat is not None and oldstat.st_size == len(contents): with open(target_path, 'rb') as f: if f.read() == contents: return oldstat with open(target_path, 'wb') as f: # Write out file f.write(contents) if honor_filemode: os.chmod(target_path, mode) return os.lstat(target_path) INVALID_DOTNAMES = (b".git", b".", b"..", b"") def validate_path_element_default(element): return element.lower() not in INVALID_DOTNAMES def validate_path_element_ntfs(element): stripped = element.rstrip(b". ").lower() if stripped in INVALID_DOTNAMES: return False if stripped == b"git~1": return False return True def validate_path(path, element_validator=validate_path_element_default): """Default path validator that just checks for .git/.""" parts = path.split(b"/") for p in parts: if not element_validator(p): return False else: return True def build_index_from_tree(root_path, index_path, object_store, tree_id, honor_filemode=True, validate_path_element=validate_path_element_default): """Generate and materialize index from a tree Args: tree_id: Tree to materialize root_path: Target dir for materialized index files index_path: Target path for generated index object_store: Non-empty object store holding tree contents honor_filemode: An optional flag to honor core.filemode setting in config file, default is core.filemode=True, change executable bit validate_path_element: Function to validate path elements to check out; default just refuses .git and .. directories. Note: existing index is wiped and contents are not merged in a working dir. Suitable only for fresh clones. """ index = Index(index_path) if not isinstance(root_path, bytes): root_path = os.fsencode(root_path) for entry in object_store.iter_tree_contents(tree_id): if not validate_path(entry.path, validate_path_element): continue full_path = _tree_to_fs_path(root_path, entry.path) if not os.path.exists(os.path.dirname(full_path)): os.makedirs(os.path.dirname(full_path)) # TODO(jelmer): Merge new index into working tree if S_ISGITLINK(entry.mode): if not os.path.isdir(full_path): os.mkdir(full_path) st = os.lstat(full_path) # TODO(jelmer): record and return submodule paths else: obj = object_store[entry.sha] st = build_file_from_blob( obj, entry.mode, full_path, honor_filemode=honor_filemode) # Add file to index if not honor_filemode or S_ISGITLINK(entry.mode): # we can not use tuple slicing to build a new tuple, # because on windows that will convert the times to # longs, which causes errors further along st_tuple = (entry.mode, st.st_ino, st.st_dev, st.st_nlink, st.st_uid, st.st_gid, st.st_size, st.st_atime, st.st_mtime, st.st_ctime) st = st.__class__(st_tuple) index[entry.path] = index_entry_from_stat(st, entry.sha, 0) index.write() -def blob_from_path_and_stat(fs_path, st): +def blob_from_path_and_stat(fs_path, st, tree_encoding='utf-8'): """Create a blob from a path and a stat object. Args: fs_path: Full file system path to file st: A stat object Returns: A `Blob` object """ assert isinstance(fs_path, bytes) blob = Blob() if stat.S_ISLNK(st.st_mode): if sys.platform == 'win32': # os.readlink on Python3 on Windows requires a unicode string. - # TODO(jelmer): Don't assume tree_encoding == fs_encoding - tree_encoding = sys.getfilesystemencoding() - fs_path = fs_path.decode(tree_encoding) + fs_path = os.fsdecode(fs_path) blob.data = os.readlink(fs_path).encode(tree_encoding) else: blob.data = os.readlink(fs_path) else: with open(fs_path, 'rb') as f: blob.data = f.read() return blob def read_submodule_head(path): """Read the head commit of a submodule. Args: path: path to the submodule Returns: HEAD sha, None if not a valid head/repository """ from dulwich.errors import NotGitRepository from dulwich.repo import Repo # Repo currently expects a "str", so decode if necessary. # TODO(jelmer): Perhaps move this into Repo() ? if not isinstance(path, str): path = os.fsdecode(path) try: repo = Repo(path) except NotGitRepository: return None try: return repo.head() except KeyError: return None def _has_directory_changed(tree_path, entry): """Check if a directory has changed after getting an error. When handling an error trying to create a blob from a path, call this function. It will check if the path is a directory. If it's a directory and a submodule, check the submodule head to see if it's has changed. If not, consider the file as changed as Git tracked a file and not a directory. Return true if the given path should be considered as changed and False otherwise or if the path is not a directory. """ # This is actually a directory if os.path.exists(os.path.join(tree_path, b'.git')): # Submodule head = read_submodule_head(tree_path) if entry.sha != head: return True else: # The file was changed to a directory, so consider it removed. return True return False def get_unstaged_changes(index, root_path, filter_blob_callback=None): """Walk through an index and check for differences against working tree. Args: index: index to check root_path: path in which to find files Returns: iterator over paths with unstaged changes """ # For each entry in the index check the sha1 & ensure not staged if not isinstance(root_path, bytes): root_path = os.fsencode(root_path) for tree_path, entry in index.iteritems(): full_path = _tree_to_fs_path(root_path, tree_path) try: st = os.lstat(full_path) if stat.S_ISDIR(st.st_mode): if _has_directory_changed(tree_path, entry): yield tree_path continue if not stat.S_ISREG(st.st_mode) and not stat.S_ISLNK(st.st_mode): continue blob = blob_from_path_and_stat(full_path, st) if filter_blob_callback is not None: blob = filter_blob_callback(blob, tree_path) except EnvironmentError as e: if e.errno == errno.ENOENT: # The file was removed, so we assume that counts as # different from whatever file used to exist. yield tree_path else: raise else: if blob.id != entry.sha: yield tree_path os_sep_bytes = os.sep.encode('ascii') def _tree_to_fs_path(root_path, tree_path): """Convert a git tree path to a file system path. Args: root_path: Root filesystem path tree_path: Git tree path as bytes Returns: File system path. """ assert isinstance(tree_path, bytes) if os_sep_bytes != b'/': sep_corrected_path = tree_path.replace(b'/', os_sep_bytes) else: sep_corrected_path = tree_path return os.path.join(root_path, sep_corrected_path) def _fs_to_tree_path(fs_path, fs_encoding=None): """Convert a file system path to a git tree path. Args: fs_path: File system path. fs_encoding: File system encoding Returns: Git tree path as bytes """ if not isinstance(fs_path, bytes): fs_path_bytes = fs_path.encode(fs_encoding) else: fs_path_bytes = fs_path if os_sep_bytes != b'/': tree_path = fs_path_bytes.replace(os_sep_bytes, b'/') else: tree_path = fs_path_bytes return tree_path def index_entry_from_path(path, object_store=None): """Create an index from a filesystem path. This returns an index value for files, symlinks and tree references. for directories and non-existant files it returns None Args: path: Path to create an index entry for object_store: Optional object store to save new blobs in Returns: An index entry; None for directories """ assert isinstance(path, bytes) st = os.lstat(path) if stat.S_ISDIR(st.st_mode): if os.path.exists(os.path.join(path, b'.git')): head = read_submodule_head(path) if head is None: return None return index_entry_from_stat( st, head, 0, mode=S_IFGITLINK) return None if stat.S_ISREG(st.st_mode) or stat.S_ISLNK(st.st_mode): blob = blob_from_path_and_stat(path, st) if object_store is not None: object_store.add_object(blob) return index_entry_from_stat(st, blob.id, 0) return None def iter_fresh_entries(paths, root_path, object_store=None): """Iterate over current versions of index entries on disk. Args: paths: Paths to iterate over root_path: Root path to access from store: Optional store to save new blobs in Returns: Iterator over path, index_entry """ for path in paths: p = _tree_to_fs_path(root_path, path) try: entry = index_entry_from_path(p, object_store=object_store) except EnvironmentError as e: if e.errno in (errno.ENOENT, errno.EISDIR): entry = None else: raise yield path, entry def iter_fresh_blobs(index, root_path): """Iterate over versions of blobs on disk referenced by index. Don't use this function; it removes missing entries from index. Args: index: Index file root_path: Root path to access from include_deleted: Include deleted entries with sha and mode set to None Returns: Iterator over path, sha, mode """ import warnings warnings.warn(PendingDeprecationWarning, "Use iter_fresh_objects instead.") for entry in iter_fresh_objects( index, root_path, include_deleted=True): if entry[1] is None: del index[entry[0]] else: yield entry def iter_fresh_objects(paths, root_path, include_deleted=False, object_store=None): """Iterate over versions of objecs on disk referenced by index. Args: index: Index file root_path: Root path to access from include_deleted: Include deleted entries with sha and mode set to None object_store: Optional object store to report new items to Returns: Iterator over path, sha, mode """ for path, entry in iter_fresh_entries(paths, root_path, object_store=object_store): if entry is None: if include_deleted: yield path, None, None else: entry = IndexEntry(*entry) yield path, entry.sha, cleanup_mode(entry.mode) def refresh_index(index, root_path): """Refresh the contents of an index. This is the equivalent to running 'git commit -a'. Args: index: Index to update root_path: Root filesystem path """ for path, entry in iter_fresh_entries(index, root_path): index[path] = path diff --git a/dulwich/refs.py b/dulwich/refs.py index 1e27feac..0b5a3c68 100644 --- a/dulwich/refs.py +++ b/dulwich/refs.py @@ -1,972 +1,968 @@ # refs.py -- For dealing with git refs # Copyright (C) 2008-2013 Jelmer Vernooij # # Dulwich is dual-licensed under the Apache License, Version 2.0 and the GNU # General Public License as public by the Free Software Foundation; version 2.0 # or (at your option) any later version. You can redistribute it and/or # modify it under the terms of either of these two licenses. # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # # You should have received a copy of the licenses; if not, see # for a copy of the GNU General Public License # and for a copy of the Apache # License, Version 2.0. # """Ref handling. """ import errno import os import sys from dulwich.errors import ( PackedRefsException, RefFormatError, ) from dulwich.objects import ( git_line, valid_hexsha, ZERO_SHA, ) from dulwich.file import ( GitFile, ensure_dir_exists, ) SYMREF = b'ref: ' LOCAL_BRANCH_PREFIX = b'refs/heads/' LOCAL_TAG_PREFIX = b'refs/tags/' BAD_REF_CHARS = set(b'\177 ~^:?*[') ANNOTATED_TAG_SUFFIX = b'^{}' def parse_symref_value(contents): """Parse a symref value. Args: contents: Contents to parse Returns: Destination """ if contents.startswith(SYMREF): return contents[len(SYMREF):].rstrip(b'\r\n') raise ValueError(contents) def check_ref_format(refname): """Check if a refname is correctly formatted. Implements all the same rules as git-check-ref-format[1]. [1] http://www.kernel.org/pub/software/scm/git/docs/git-check-ref-format.html Args: refname: The refname to check Returns: True if refname is valid, False otherwise """ # These could be combined into one big expression, but are listed # separately to parallel [1]. if b'/.' in refname or refname.startswith(b'.'): return False if b'/' not in refname: return False if b'..' in refname: return False for i, c in enumerate(refname): if ord(refname[i:i+1]) < 0o40 or c in BAD_REF_CHARS: return False if refname[-1] in b'/.': return False if refname.endswith(b'.lock'): return False if b'@{' in refname: return False if b'\\' in refname: return False return True class RefsContainer(object): """A container for refs.""" def __init__(self, logger=None): self._logger = logger def _log(self, ref, old_sha, new_sha, committer=None, timestamp=None, timezone=None, message=None): if self._logger is None: return if message is None: return self._logger(ref, old_sha, new_sha, committer, timestamp, timezone, message) def set_symbolic_ref(self, name, other, committer=None, timestamp=None, timezone=None, message=None): """Make a ref point at another ref. Args: name: Name of the ref to set other: Name of the ref to point at message: Optional message """ raise NotImplementedError(self.set_symbolic_ref) def get_packed_refs(self): """Get contents of the packed-refs file. Returns: Dictionary mapping ref names to SHA1s Note: Will return an empty dictionary when no packed-refs file is present. """ raise NotImplementedError(self.get_packed_refs) def get_peeled(self, name): """Return the cached peeled value of a ref, if available. Args: name: Name of the ref to peel Returns: The peeled value of the ref. If the ref is known not point to a tag, this will be the SHA the ref refers to. If the ref may point to a tag, but no cached information is available, None is returned. """ return None def import_refs(self, base, other, committer=None, timestamp=None, timezone=None, message=None, prune=False): if prune: to_delete = set(self.subkeys(base)) else: to_delete = set() for name, value in other.items(): self.set_if_equals(b'/'.join((base, name)), None, value, message=message) if to_delete: try: to_delete.remove(name) except KeyError: pass for ref in to_delete: self.remove_if_equals(b'/'.join((base, ref)), None) def allkeys(self): """All refs present in this container.""" raise NotImplementedError(self.allkeys) def __iter__(self): return iter(self.allkeys()) def keys(self, base=None): """Refs present in this container. Args: base: An optional base to return refs under. Returns: An unsorted set of valid refs in this container, including packed refs. """ if base is not None: return self.subkeys(base) else: return self.allkeys() def subkeys(self, base): """Refs present in this container under a base. Args: base: The base to return refs under. Returns: A set of valid refs in this container under the base; the base prefix is stripped from the ref names returned. """ keys = set() base_len = len(base) + 1 for refname in self.allkeys(): if refname.startswith(base): keys.add(refname[base_len:]) return keys def as_dict(self, base=None): """Return the contents of this container as a dictionary. """ ret = {} keys = self.keys(base) if base is None: base = b'' else: base = base.rstrip(b'/') for key in keys: try: ret[key] = self[(base + b'/' + key).strip(b'/')] except KeyError: continue # Unable to resolve return ret def _check_refname(self, name): """Ensure a refname is valid and lives in refs or is HEAD. HEAD is not a valid refname according to git-check-ref-format, but this class needs to be able to touch HEAD. Also, check_ref_format expects refnames without the leading 'refs/', but this class requires that so it cannot touch anything outside the refs dir (or HEAD). Args: name: The name of the reference. Raises: KeyError: if a refname is not HEAD or is otherwise not valid. """ if name in (b'HEAD', b'refs/stash'): return if not name.startswith(b'refs/') or not check_ref_format(name[5:]): raise RefFormatError(name) def read_ref(self, refname): """Read a reference without following any references. Args: refname: The name of the reference Returns: The contents of the ref file, or None if it does not exist. """ contents = self.read_loose_ref(refname) if not contents: contents = self.get_packed_refs().get(refname, None) return contents def read_loose_ref(self, name): """Read a loose reference and return its contents. Args: name: the refname to read Returns: The contents of the ref file, or None if it does not exist. """ raise NotImplementedError(self.read_loose_ref) def follow(self, name): """Follow a reference name. Returns: a tuple of (refnames, sha), wheres refnames are the names of references in the chain """ contents = SYMREF + name depth = 0 refnames = [] while contents.startswith(SYMREF): refname = contents[len(SYMREF):] refnames.append(refname) contents = self.read_ref(refname) if not contents: break depth += 1 if depth > 5: raise KeyError(name) return refnames, contents def _follow(self, name): import warnings warnings.warn( "RefsContainer._follow is deprecated. Use RefsContainer.follow " "instead.", DeprecationWarning) refnames, contents = self.follow(name) if not refnames: return (None, contents) return (refnames[-1], contents) def __contains__(self, refname): if self.read_ref(refname): return True return False def __getitem__(self, name): """Get the SHA1 for a reference name. This method follows all symbolic references. """ _, sha = self.follow(name) if sha is None: raise KeyError(name) return sha def set_if_equals(self, name, old_ref, new_ref, committer=None, timestamp=None, timezone=None, message=None): """Set a refname to new_ref only if it currently equals old_ref. This method follows all symbolic references if applicable for the subclass, and can be used to perform an atomic compare-and-swap operation. Args: name: The refname to set. old_ref: The old sha the refname must refer to, or None to set unconditionally. new_ref: The new sha the refname will refer to. message: Message for reflog Returns: True if the set was successful, False otherwise. """ raise NotImplementedError(self.set_if_equals) def add_if_new(self, name, ref): """Add a new reference only if it does not already exist. Args: name: Ref name ref: Ref value message: Message for reflog """ raise NotImplementedError(self.add_if_new) def __setitem__(self, name, ref): """Set a reference name to point to the given SHA1. This method follows all symbolic references if applicable for the subclass. Note: This method unconditionally overwrites the contents of a reference. To update atomically only if the reference has not changed, use set_if_equals(). Args: name: The refname to set. ref: The new sha the refname will refer to. """ self.set_if_equals(name, None, ref) def remove_if_equals(self, name, old_ref, committer=None, timestamp=None, timezone=None, message=None): """Remove a refname only if it currently equals old_ref. This method does not follow symbolic references, even if applicable for the subclass. It can be used to perform an atomic compare-and-delete operation. Args: name: The refname to delete. old_ref: The old sha the refname must refer to, or None to delete unconditionally. message: Message for reflog Returns: True if the delete was successful, False otherwise. """ raise NotImplementedError(self.remove_if_equals) def __delitem__(self, name): """Remove a refname. This method does not follow symbolic references, even if applicable for the subclass. Note: This method unconditionally deletes the contents of a reference. To delete atomically only if the reference has not changed, use remove_if_equals(). Args: name: The refname to delete. """ self.remove_if_equals(name, None) def get_symrefs(self): """Get a dict with all symrefs in this container. Returns: Dictionary mapping source ref to target ref """ ret = {} for src in self.allkeys(): try: dst = parse_symref_value(self.read_ref(src)) except ValueError: pass else: ret[src] = dst return ret class DictRefsContainer(RefsContainer): """RefsContainer backed by a simple dict. This container does not support symbolic or packed references and is not threadsafe. """ def __init__(self, refs, logger=None): super(DictRefsContainer, self).__init__(logger=logger) self._refs = refs self._peeled = {} def allkeys(self): return self._refs.keys() def read_loose_ref(self, name): return self._refs.get(name, None) def get_packed_refs(self): return {} def set_symbolic_ref(self, name, other, committer=None, timestamp=None, timezone=None, message=None): old = self.follow(name)[-1] self._refs[name] = SYMREF + other self._log(name, old, old, committer=committer, timestamp=timestamp, timezone=timezone, message=message) def set_if_equals(self, name, old_ref, new_ref, committer=None, timestamp=None, timezone=None, message=None): if old_ref is not None and self._refs.get(name, ZERO_SHA) != old_ref: return False realnames, _ = self.follow(name) for realname in realnames: self._check_refname(realname) old = self._refs.get(realname) self._refs[realname] = new_ref self._log(realname, old, new_ref, committer=committer, timestamp=timestamp, timezone=timezone, message=message) return True def add_if_new(self, name, ref, committer=None, timestamp=None, timezone=None, message=None): if name in self._refs: return False self._refs[name] = ref self._log(name, None, ref, committer=committer, timestamp=timestamp, timezone=timezone, message=message) return True def remove_if_equals(self, name, old_ref, committer=None, timestamp=None, timezone=None, message=None): if old_ref is not None and self._refs.get(name, ZERO_SHA) != old_ref: return False try: old = self._refs.pop(name) except KeyError: pass else: self._log(name, old, None, committer=committer, timestamp=timestamp, timezone=timezone, message=message) return True def get_peeled(self, name): return self._peeled.get(name) def _update(self, refs): """Update multiple refs; intended only for testing.""" # TODO(dborowitz): replace this with a public function that uses # set_if_equal. self._refs.update(refs) def _update_peeled(self, peeled): """Update cached peeled refs; intended only for testing.""" self._peeled.update(peeled) class InfoRefsContainer(RefsContainer): """Refs container that reads refs from a info/refs file.""" def __init__(self, f): self._refs = {} self._peeled = {} for l in f.readlines(): sha, name = l.rstrip(b'\n').split(b'\t') if name.endswith(ANNOTATED_TAG_SUFFIX): name = name[:-3] if not check_ref_format(name): raise ValueError("invalid ref name %r" % name) self._peeled[name] = sha else: if not check_ref_format(name): raise ValueError("invalid ref name %r" % name) self._refs[name] = sha def allkeys(self): return self._refs.keys() def read_loose_ref(self, name): return self._refs.get(name, None) def get_packed_refs(self): return {} def get_peeled(self, name): try: return self._peeled[name] except KeyError: return self._refs[name] class DiskRefsContainer(RefsContainer): """Refs container that reads refs from disk.""" def __init__(self, path, worktree_path=None, logger=None): super(DiskRefsContainer, self).__init__(logger=logger) if getattr(path, 'encode', None) is not None: path = os.fsencode(path) self.path = path if worktree_path is None: worktree_path = path if getattr(worktree_path, 'encode', None) is not None: worktree_path = os.fsencode(worktree_path) self.worktree_path = worktree_path self._packed_refs = None self._peeled_refs = None def __repr__(self): return "%s(%r)" % (self.__class__.__name__, self.path) def subkeys(self, base): subkeys = set() path = self.refpath(base) for root, unused_dirs, files in os.walk(path): dir = root[len(path):] if os.path.sep != '/': - dir = dir.replace(os.path.sep.encode( - sys.getfilesystemencoding()), b"/") + dir = dir.replace(os.fsencode(os.path.sep), b"/") dir = dir.strip(b'/') for filename in files: refname = b"/".join(([dir] if dir else []) + [filename]) # check_ref_format requires at least one /, so we prepend the # base before calling it. if check_ref_format(base + b'/' + refname): subkeys.add(refname) for key in self.get_packed_refs(): if key.startswith(base): subkeys.add(key[len(base):].strip(b'/')) return subkeys def allkeys(self): allkeys = set() if os.path.exists(self.refpath(b'HEAD')): allkeys.add(b'HEAD') path = self.refpath(b'') refspath = self.refpath(b'refs') for root, unused_dirs, files in os.walk(refspath): dir = root[len(path):] if os.path.sep != '/': - dir = dir.replace( - os.path.sep.encode(sys.getfilesystemencoding()), b"/") + dir = dir.replace(os.fsencode(os.path.sep), b"/") for filename in files: refname = b"/".join([dir, filename]) if check_ref_format(refname): allkeys.add(refname) allkeys.update(self.get_packed_refs()) return allkeys def refpath(self, name): """Return the disk path of a ref. """ if os.path.sep != "/": - name = name.replace( - b"/", - os.path.sep.encode(sys.getfilesystemencoding())) + name = name.replace(b"/", os.fsencode(os.path.sep)) # TODO: as the 'HEAD' reference is working tree specific, it # should actually not be a part of RefsContainer if name == b'HEAD': return os.path.join(self.worktree_path, name) else: return os.path.join(self.path, name) def get_packed_refs(self): """Get contents of the packed-refs file. Returns: Dictionary mapping ref names to SHA1s Note: Will return an empty dictionary when no packed-refs file is present. """ # TODO: invalidate the cache on repacking if self._packed_refs is None: # set both to empty because we want _peeled_refs to be # None if and only if _packed_refs is also None. self._packed_refs = {} self._peeled_refs = {} path = os.path.join(self.path, b'packed-refs') try: f = GitFile(path, 'rb') except IOError as e: if e.errno == errno.ENOENT: return {} raise with f: first_line = next(iter(f)).rstrip() if (first_line.startswith(b'# pack-refs') and b' peeled' in first_line): for sha, name, peeled in read_packed_refs_with_peeled(f): self._packed_refs[name] = sha if peeled: self._peeled_refs[name] = peeled else: f.seek(0) for sha, name in read_packed_refs(f): self._packed_refs[name] = sha return self._packed_refs def get_peeled(self, name): """Return the cached peeled value of a ref, if available. Args: name: Name of the ref to peel Returns: The peeled value of the ref. If the ref is known not point to a tag, this will be the SHA the ref refers to. If the ref may point to a tag, but no cached information is available, None is returned. """ self.get_packed_refs() if self._peeled_refs is None or name not in self._packed_refs: # No cache: no peeled refs were read, or this ref is loose return None if name in self._peeled_refs: return self._peeled_refs[name] else: # Known not peelable return self[name] def read_loose_ref(self, name): """Read a reference file and return its contents. If the reference file a symbolic reference, only read the first line of the file. Otherwise, only read the first 40 bytes. Args: name: the refname to read, relative to refpath Returns: The contents of the ref file, or None if the file does not exist. Raises: IOError: if any other error occurs """ filename = self.refpath(name) try: with GitFile(filename, 'rb') as f: header = f.read(len(SYMREF)) if header == SYMREF: # Read only the first line return header + next(iter(f)).rstrip(b'\r\n') else: # Read only the first 40 bytes return header + f.read(40 - len(SYMREF)) except IOError as e: if e.errno in (errno.ENOENT, errno.EISDIR, errno.ENOTDIR): return None raise def _remove_packed_ref(self, name): if self._packed_refs is None: return filename = os.path.join(self.path, b'packed-refs') # reread cached refs from disk, while holding the lock f = GitFile(filename, 'wb') try: self._packed_refs = None self.get_packed_refs() if name not in self._packed_refs: return del self._packed_refs[name] if name in self._peeled_refs: del self._peeled_refs[name] write_packed_refs(f, self._packed_refs, self._peeled_refs) f.close() finally: f.abort() def set_symbolic_ref(self, name, other, committer=None, timestamp=None, timezone=None, message=None): """Make a ref point at another ref. Args: name: Name of the ref to set other: Name of the ref to point at message: Optional message to describe the change """ self._check_refname(name) self._check_refname(other) filename = self.refpath(name) f = GitFile(filename, 'wb') try: f.write(SYMREF + other + b'\n') sha = self.follow(name)[-1] self._log(name, sha, sha, committer=committer, timestamp=timestamp, timezone=timezone, message=message) except BaseException: f.abort() raise else: f.close() def set_if_equals(self, name, old_ref, new_ref, committer=None, timestamp=None, timezone=None, message=None): """Set a refname to new_ref only if it currently equals old_ref. This method follows all symbolic references, and can be used to perform an atomic compare-and-swap operation. Args: name: The refname to set. old_ref: The old sha the refname must refer to, or None to set unconditionally. new_ref: The new sha the refname will refer to. message: Set message for reflog Returns: True if the set was successful, False otherwise. """ self._check_refname(name) try: realnames, _ = self.follow(name) realname = realnames[-1] except (KeyError, IndexError): realname = name filename = self.refpath(realname) # make sure none of the ancestor folders is in packed refs probe_ref = os.path.dirname(realname) packed_refs = self.get_packed_refs() while probe_ref: if packed_refs.get(probe_ref, None) is not None: raise OSError(errno.ENOTDIR, 'Not a directory: {}'.format(filename)) probe_ref = os.path.dirname(probe_ref) ensure_dir_exists(os.path.dirname(filename)) with GitFile(filename, 'wb') as f: if old_ref is not None: try: # read again while holding the lock orig_ref = self.read_loose_ref(realname) if orig_ref is None: orig_ref = self.get_packed_refs().get( realname, ZERO_SHA) if orig_ref != old_ref: f.abort() return False except (OSError, IOError): f.abort() raise try: f.write(new_ref + b'\n') except (OSError, IOError): f.abort() raise self._log(realname, old_ref, new_ref, committer=committer, timestamp=timestamp, timezone=timezone, message=message) return True def add_if_new(self, name, ref, committer=None, timestamp=None, timezone=None, message=None): """Add a new reference only if it does not already exist. This method follows symrefs, and only ensures that the last ref in the chain does not exist. Args: name: The refname to set. ref: The new sha the refname will refer to. message: Optional message for reflog Returns: True if the add was successful, False otherwise. """ try: realnames, contents = self.follow(name) if contents is not None: return False realname = realnames[-1] except (KeyError, IndexError): realname = name self._check_refname(realname) filename = self.refpath(realname) ensure_dir_exists(os.path.dirname(filename)) with GitFile(filename, 'wb') as f: if os.path.exists(filename) or name in self.get_packed_refs(): f.abort() return False try: f.write(ref + b'\n') except (OSError, IOError): f.abort() raise else: self._log(name, None, ref, committer=committer, timestamp=timestamp, timezone=timezone, message=message) return True def remove_if_equals(self, name, old_ref, committer=None, timestamp=None, timezone=None, message=None): """Remove a refname only if it currently equals old_ref. This method does not follow symbolic references. It can be used to perform an atomic compare-and-delete operation. Args: name: The refname to delete. old_ref: The old sha the refname must refer to, or None to delete unconditionally. message: Optional message Returns: True if the delete was successful, False otherwise. """ self._check_refname(name) filename = self.refpath(name) ensure_dir_exists(os.path.dirname(filename)) f = GitFile(filename, 'wb') try: if old_ref is not None: orig_ref = self.read_loose_ref(name) if orig_ref is None: orig_ref = self.get_packed_refs().get(name, ZERO_SHA) if orig_ref != old_ref: return False # remove the reference file itself try: os.remove(filename) except OSError as e: if e.errno != errno.ENOENT: # may only be packed raise self._remove_packed_ref(name) self._log(name, old_ref, None, committer=committer, timestamp=timestamp, timezone=timezone, message=message) finally: # never write, we just wanted the lock f.abort() # outside of the lock, clean-up any parent directory that might now # be empty. this ensures that re-creating a reference of the same # name of what was previously a directory works as expected parent = name while True: try: parent, _ = parent.rsplit(b'/', 1) except ValueError: break parent_filename = self.refpath(parent) try: os.rmdir(parent_filename) except OSError: # this can be caused by the parent directory being # removed by another process, being not empty, etc. # in any case, this is non fatal because we already # removed the reference, just ignore it break return True def _split_ref_line(line): """Split a single ref line into a tuple of SHA1 and name.""" fields = line.rstrip(b'\n\r').split(b' ') if len(fields) != 2: raise PackedRefsException("invalid ref line %r" % line) sha, name = fields if not valid_hexsha(sha): raise PackedRefsException("Invalid hex sha %r" % sha) if not check_ref_format(name): raise PackedRefsException("invalid ref name %r" % name) return (sha, name) def read_packed_refs(f): """Read a packed refs file. Args: f: file-like object to read from Returns: Iterator over tuples with SHA1s and ref names. """ for l in f: if l.startswith(b'#'): # Comment continue if l.startswith(b'^'): raise PackedRefsException( "found peeled ref in packed-refs without peeled") yield _split_ref_line(l) def read_packed_refs_with_peeled(f): """Read a packed refs file including peeled refs. Assumes the "# pack-refs with: peeled" line was already read. Yields tuples with ref names, SHA1s, and peeled SHA1s (or None). Args: f: file-like object to read from, seek'ed to the second line """ last = None for line in f: if line[0] == b'#': continue line = line.rstrip(b'\r\n') if line.startswith(b'^'): if not last: raise PackedRefsException("unexpected peeled ref line") if not valid_hexsha(line[1:]): raise PackedRefsException("Invalid hex sha %r" % line[1:]) sha, name = _split_ref_line(last) last = None yield (sha, name, line[1:]) else: if last: sha, name = _split_ref_line(last) yield (sha, name, None) last = line if last: sha, name = _split_ref_line(last) yield (sha, name, None) def write_packed_refs(f, packed_refs, peeled_refs=None): """Write a packed refs file. Args: f: empty file-like object to write to packed_refs: dict of refname to sha of packed refs to write peeled_refs: dict of refname to peeled value of sha """ if peeled_refs is None: peeled_refs = {} else: f.write(b'# pack-refs with: peeled\n') for refname in sorted(packed_refs.keys()): f.write(git_line(packed_refs[refname], refname)) if refname in peeled_refs: f.write(b'^' + peeled_refs[refname] + b'\n') def read_info_refs(f): ret = {} for l in f.readlines(): (sha, name) = l.rstrip(b"\r\n").split(b"\t", 1) ret[name] = sha return ret def write_info_refs(refs, store): """Generate info refs.""" for name, sha in sorted(refs.items()): # get_refs() includes HEAD as a special case, but we don't want to # advertise it if name == b'HEAD': continue try: o = store[sha] except KeyError: continue peeled = store.peel_sha(sha) yield o.id + b'\t' + name + b'\n' if o.id != peeled.id: yield peeled.id + b'\t' + name + ANNOTATED_TAG_SUFFIX + b'\n' def is_local_branch(x): return x.startswith(LOCAL_BRANCH_PREFIX) def strip_peeled_refs(refs): """Remove all peeled refs""" return {ref: sha for (ref, sha) in refs.items() if not ref.endswith(ANNOTATED_TAG_SUFFIX)}