1 """Virtual File System interface to bup repository content.
3 This module provides a path-based interface to the content of a bup
6 The VFS is structured like this:
9 /SAVE-NAME/SAVE-DATE/...
12 Each path is represented by an item that has least an item.meta which
13 may be either a Metadata object, or an integer mode. Functions like
14 item_mode() and item_size() will return the mode and size in either
15 case. Any item.meta Metadata instances must not be modified directly.
16 Make a copy to modify via item.meta.copy() if needed, or call
19 The want_meta argument is advisory for calls that accept it, and it
20 may not be honored. Callers must be able to handle an item.meta value
21 that is either an instance of Metadata or an integer mode, perhaps
22 via item_mode() or augment_item_meta().
24 Setting want_meta=False is rarely desirable since it can limit the VFS
25 to only the metadata that git itself can represent, and so for
26 example, fifos and sockets will appear to be regular files
27 (e.g. S_ISREG(item_mode(item)) will be true). But the option is still
28 provided because it may be more efficient when just the path names or
29 the more limited metadata is sufficient.
31 Any given metadata object's size may be None, in which case the size
32 can be computed via item_size() or augment_item_meta(...,
35 When traversing a directory using functions like contents(), the meta
36 value for any directories other than '.' will be a default directory
37 mode, not a Metadata object. This is because the actual metadata for
38 a directory is stored inside the directory (see
39 fill_in_metadata_if_dir() or ensure_item_has_metadata()).
41 Commit items represent commits (e.g. /.tag/some-commit or
42 /foo/latest), and for most purposes, they appear as the underlying
43 tree. S_ISDIR(item_mode(item)) will return true for both tree Items
44 and Commits and the commit's oid is the tree hash; the commit hash is
49 from __future__ import absolute_import, print_function
50 from binascii import hexlify, unhexlify
51 from collections import namedtuple
52 from errno import EINVAL, ELOOP, ENOENT, ENOTDIR
53 from itertools import chain, dropwhile, groupby, tee
54 from random import randrange
55 from stat import S_IFDIR, S_IFLNK, S_IFREG, S_ISDIR, S_ISLNK, S_ISREG
56 from time import localtime, strftime
59 from bup import git, metadata, vint
60 from bup.compat import hexstr, range
61 from bup.git import BUP_CHUNKED, cp, get_commit_items, parse_commit, tree_decode
62 from bup.helpers import debug2, last
63 from bup.io import path_msg
64 from bup.metadata import Metadata
65 from bup.vint import read_bvec, write_bvec
66 from bup.vint import read_vint, write_vint
67 from bup.vint import read_vuint, write_vuint
69 if sys.version_info[0] < 3:
70 from exceptions import IOError as py_IOError
74 # We currently assume that it's always appropriate to just forward IOErrors
77 class IOError(py_IOError):
78 def __init__(self, errno, message, terminus=None):
79 py_IOError.__init__(self, errno, message)
80 self.terminus = terminus
82 def write_ioerror(port, ex):
83 assert isinstance(ex, IOError)
85 (1 if ex.errno is not None else 0)
86 | (2 if ex.strerror is not None else 0)
87 | (4 if ex.terminus is not None else 0))
88 if ex.errno is not None:
89 write_vint(port, ex.errno)
90 if ex.strerror is not None:
91 write_bvec(port, ex.strerror.encode('utf-8'))
92 if ex.terminus is not None:
93 write_resolution(port, ex.terminus)
95 def read_ioerror(port):
96 mask = read_vuint(port)
97 no = read_vint(port) if 1 & mask else None
98 msg = read_bvec(port).decode('utf-8') if 2 & mask else None
99 term = read_resolution(port) if 4 & mask else None
100 return IOError(errno=no, message=msg, terminus=term)
103 default_file_mode = S_IFREG | 0o644
104 default_dir_mode = S_IFDIR | 0o755
105 default_symlink_mode = S_IFLNK | 0o755
107 def _default_mode_for_gitmode(gitmode):
109 return default_file_mode
111 return default_dir_mode
113 return default_symlink_mode
114 raise Exception('unexpected git mode ' + oct(gitmode))
116 def _normal_or_chunked_file_size(repo, oid):
117 """Return the size of the normal or chunked file indicated by oid."""
118 # FIXME: --batch-format CatPipe?
119 it = repo.cat(hexlify(oid))
120 _, obj_t, size = next(it)
122 while obj_t == b'tree':
123 mode, name, last_oid = last(tree_decode(b''.join(it)))
125 it = repo.cat(hexlify(last_oid))
126 _, obj_t, size = next(it)
127 return ofs + sum(len(b) for b in it)
129 def _skip_chunks_before_offset(tree, offset):
130 prev_ent = next(tree, None)
135 ent_ofs = int(ent[1], 16)
137 return chain([prev_ent, ent], tree)
138 if ent_ofs == offset:
139 return chain([ent], tree)
143 def _tree_chunks(repo, tree, startofs):
144 "Tree should be a sequence of (name, mode, hash) as per tree_decode()."
145 assert(startofs >= 0)
146 # name is the chunk's hex offset in the original file
147 for mode, name, oid in _skip_chunks_before_offset(tree, startofs):
149 skipmore = startofs - ofs
152 it = repo.cat(hexlify(oid))
153 _, obj_t, size = next(it)
156 assert obj_t == b'tree'
157 for b in _tree_chunks(repo, tree_decode(data), skipmore):
160 assert obj_t == b'blob'
161 yield data[skipmore:]
164 def __init__(self, repo, oid, startofs):
165 it = repo.cat(hexlify(oid))
166 _, obj_t, size = next(it)
167 isdir = obj_t == b'tree'
170 self.it = _tree_chunks(repo, tree_decode(data), startofs)
174 self.blob = data[startofs:]
177 def next(self, size):
179 while len(out) < size:
180 if self.it and not self.blob:
182 self.blob = next(self.it)
183 except StopIteration:
186 want = size - len(out)
187 out += self.blob[:want]
188 self.blob = self.blob[want:]
191 debug2('next(%d) returned %d\n' % (size, len(out)))
195 class _FileReader(object):
196 def __init__(self, repo, oid, known_size=None):
197 assert len(oid) == 20
202 self._size = known_size
204 def _compute_size(self):
206 self._size = _normal_or_chunked_file_size(self._repo, self.oid)
210 if ofs < 0 or ofs > self._compute_size():
211 raise IOError(EINVAL, 'Invalid seek offset: %d' % ofs)
217 def read(self, count=-1):
218 size = self._compute_size()
222 count = size - self.ofs
223 if not self.reader or self.reader.ofs != self.ofs:
224 self.reader = _ChunkReader(self._repo, self.oid, self.ofs)
226 buf = self.reader.next(count)
229 raise # our offsets will be all screwed up otherwise
238 def __exit__(self, type, value, traceback):
242 _multiple_slashes_rx = re.compile(br'//+')
244 def _decompose_path(path):
245 """Return a boolean indicating whether the path is absolute, and a
246 reversed list of path elements, omitting any occurrences of "."
247 and ignoring any leading or trailing slash. If the path is
248 effectively '/' or '.', return an empty list.
251 path = re.sub(_multiple_slashes_rx, b'/', path)
253 return True, True, []
254 is_absolute = must_be_dir = False
255 if path.startswith(b'/'):
258 for suffix in (b'/', b'/.'):
259 if path.endswith(suffix):
261 path = path[:-len(suffix)]
262 parts = [x for x in path.split(b'/') if x != b'.']
265 must_be_dir = True # e.g. path was effectively '.' or '/', etc.
266 return is_absolute, must_be_dir, parts
269 Item = namedtuple('Item', ('meta', 'oid'))
270 Chunky = namedtuple('Chunky', ('meta', 'oid'))
271 FakeLink = namedtuple('FakeLink', ('meta', 'target'))
272 Root = namedtuple('Root', ('meta'))
273 Tags = namedtuple('Tags', ('meta'))
274 RevList = namedtuple('RevList', ('meta', 'oid'))
275 Commit = namedtuple('Commit', ('meta', 'oid', 'coid'))
277 item_types = frozenset((Item, Chunky, Root, Tags, RevList, Commit))
278 real_tree_types = frozenset((Item, Commit))
280 def write_item(port, item):
282 name = bytes(kind.__name__.encode('ascii'))
284 has_meta = 1 if isinstance(meta, Metadata) else 0
285 if kind in (Item, Chunky, RevList):
286 assert len(item.oid) == 20
288 vint.send(port, 'sVs', name, has_meta, item.oid)
289 Metadata.write(meta, port, include_path=False)
291 vint.send(port, 'sVsV', name, has_meta, item.oid, item.meta)
292 elif kind in (Root, Tags):
294 vint.send(port, 'sV', name, has_meta)
295 Metadata.write(meta, port, include_path=False)
297 vint.send(port, 'sVV', name, has_meta, item.meta)
299 assert len(item.oid) == 20
300 assert len(item.coid) == 20
302 vint.send(port, 'sVss', name, has_meta, item.oid, item.coid)
303 Metadata.write(meta, port, include_path=False)
305 vint.send(port, 'sVssV', name, has_meta, item.oid, item.coid,
307 elif kind == FakeLink:
309 vint.send(port, 'sVs', name, has_meta, item.target)
310 Metadata.write(meta, port, include_path=False)
312 vint.send(port, 'sVsV', name, has_meta, item.target, item.meta)
317 def read_m(port, has_meta):
319 m = Metadata.read(port)
321 return read_vuint(port)
322 kind, has_meta = vint.recv(port, 'sV')
324 oid, meta = read_bvec(port), read_m(port, has_meta)
325 return Item(oid=oid, meta=meta)
326 if kind == b'Chunky':
327 oid, meta = read_bvec(port), read_m(port, has_meta)
328 return Chunky(oid=oid, meta=meta)
329 if kind == b'RevList':
330 oid, meta = read_bvec(port), read_m(port, has_meta)
331 return RevList(oid=oid, meta=meta)
333 return Root(meta=read_m(port, has_meta))
335 return Tags(meta=read_m(port, has_meta))
336 if kind == b'Commit':
337 oid, coid = vint.recv(port, 'ss')
338 meta = read_m(port, has_meta)
339 return Commit(oid=oid, coid=coid, meta=meta)
340 if kind == b'FakeLink':
341 target, meta = read_bvec(port), read_m(port, has_meta)
342 return FakeLink(target=target, meta=meta)
345 def write_resolution(port, resolution):
346 write_vuint(port, len(resolution))
347 for name, item in resolution:
348 write_bvec(port, name)
351 write_item(port, item)
355 def read_resolution(port):
359 name = read_bvec(port)
360 have_item = ord(port.read(1))
361 assert have_item in (0, 1)
362 item = read_item(port) if have_item else None
363 result.append((name, item))
367 _root = Root(meta=default_dir_mode)
368 _tags = Tags(meta=default_dir_mode)
373 ### A general purpose shared cache with (currently) cheap random
374 ### eviction. At the moment there is no weighting so a single commit
375 ### item is just as likely to be evicted as an entire "rev-list". See
376 ### is_valid_cache_key for a description of the expected content.
380 _cache_max_items = 30000
383 global _cache, _cache_keys
387 def is_valid_cache_key(x):
388 """Return logically true if x looks like it could be a valid cache key
389 (with respect to structure). Current valid cache entries:
390 res:... -> resolution
392 rvl:OID -> {'.', commit, '2012...', next_commit, ...}
394 # Suspect we may eventually add "(container_oid, name) -> ...", and others.
398 if tag in (b'itm:', b'rvl:') and len(x) == 24:
405 if not is_valid_cache_key(key):
406 raise Exception('invalid cache key: ' + repr(key))
407 return _cache.get(key)
409 def cache_notice(key, value):
410 global _cache, _cache_keys, _cache_max_items
411 if not is_valid_cache_key(key):
412 raise Exception('invalid cache key: ' + repr(key))
415 if len(_cache) < _cache_max_items:
416 _cache_keys.append(key)
419 victim_i = randrange(0, len(_cache_keys))
420 victim = _cache_keys[victim_i]
422 _cache_keys[victim_i] = key
425 def cache_get_commit_item(oid, need_meta=True):
426 """Return the requested tree item if it can be found in the cache.
427 When need_meta is true don't return a cached item that only has a
429 # tree might be stored independently, or as '.' with its entries.
430 commit_key = b'itm:' + oid
431 item = cache_get(commit_key)
435 if isinstance(item.meta, Metadata):
437 entries = cache_get(b'rvl:' + oid)
441 def cache_get_revlist_item(oid, need_meta=True):
442 commit = cache_get_commit_item(oid, need_meta=need_meta)
444 return RevList(oid=oid, meta=commit.meta)
447 """Return a completely independent copy of item, such that
448 modifications will not affect the original.
451 meta = getattr(item, 'meta', None)
452 if isinstance(meta, Metadata):
453 return(item._replace(meta=meta.copy()))
457 """Return the integer mode (stat st_mode) for item."""
459 if isinstance(m, Metadata):
463 def _read_dir_meta(bupm):
464 # This is because save writes unmodified Metadata() entries for
465 # fake parents -- test-save-strip-graft.sh demonstrates.
466 m = Metadata.read(bupm)
468 return default_dir_mode
469 assert m.mode is not None
474 def tree_data_and_bupm(repo, oid):
475 """Return (tree_bytes, bupm_oid) where bupm_oid will be None if the
476 tree has no metadata (i.e. older bup save, or non-bup tree).
479 assert len(oid) == 20
480 it = repo.cat(hexlify(oid))
481 _, item_t, size = next(it)
483 if item_t == b'commit':
484 commit = parse_commit(data)
485 it = repo.cat(commit.tree)
486 _, item_t, size = next(it)
488 assert item_t == b'tree'
489 elif item_t != b'tree':
490 raise Exception('%s is not a tree or commit' % hexstr(oid))
491 for _, mangled_name, sub_oid in tree_decode(data):
492 if mangled_name == b'.bupm':
494 if mangled_name > b'.bupm':
498 def _find_treeish_oid_metadata(repo, oid):
499 """Return the metadata for the tree or commit oid, or None if the tree
500 has no metadata (i.e. older bup save, or non-bup tree).
503 tree_data, bupm_oid = tree_data_and_bupm(repo, oid)
505 with _FileReader(repo, bupm_oid) as meta_stream:
506 return _read_dir_meta(meta_stream)
509 def _readlink(repo, oid):
510 return b''.join(repo.join(hexlify(oid)))
512 def readlink(repo, item):
513 """Return the link target of item, which must be a symlink. Reads the
514 target from the repository if necessary."""
516 assert S_ISLNK(item_mode(item))
517 if isinstance(item, FakeLink):
519 if isinstance(item.meta, Metadata):
520 target = item.meta.symlink_target
523 return _readlink(repo, item.oid)
525 def _compute_item_size(repo, item):
526 mode = item_mode(item)
528 size = _normal_or_chunked_file_size(repo, item.oid)
531 if isinstance(item, FakeLink):
532 return len(item.target)
533 return len(_readlink(repo, item.oid))
536 def item_size(repo, item):
537 """Return the size of item, computing it if necessary."""
539 if isinstance(m, Metadata) and m.size is not None:
541 return _compute_item_size(repo, item)
543 def tree_data_reader(repo, oid):
544 """Return an open reader for all of the data contained within oid. If
545 oid refers to a tree, recursively concatenate all of its contents."""
546 return _FileReader(repo, oid)
548 def fopen(repo, item):
549 """Return an open reader for the given file item."""
550 assert S_ISREG(item_mode(item))
551 return tree_data_reader(repo, item.oid)
553 def _commit_item_from_data(oid, data):
554 info = parse_commit(data)
555 return Commit(meta=default_dir_mode,
556 oid=unhexlify(info.tree),
559 def _commit_item_from_oid(repo, oid, require_meta):
560 commit = cache_get_commit_item(oid, need_meta=require_meta)
561 if commit and ((not require_meta) or isinstance(commit.meta, Metadata)):
563 it = repo.cat(hexlify(oid))
564 _, typ, size = next(it)
565 assert typ == b'commit'
566 commit = _commit_item_from_data(oid, b''.join(it))
568 meta = _find_treeish_oid_metadata(repo, commit.oid)
570 commit = commit._replace(meta=meta)
571 commit_key = b'itm:' + oid
572 cache_notice(commit_key, commit)
575 def _revlist_item_from_oid(repo, oid, require_meta):
576 commit = _commit_item_from_oid(repo, oid, require_meta)
577 return RevList(oid=oid, meta=commit.meta)
579 def root_items(repo, names=None, want_meta=True):
580 """Yield (name, item) for the items in '/' in the VFS. Return
581 everything if names is logically false, otherwise return only
582 items with a name in the collection.
585 # FIXME: what about non-leaf refs like 'refs/heads/foo/bar/baz?
591 # FIXME: maybe eventually support repo.clone() or something
592 # and pass in two repos, so we can drop the tuple() and stream
593 # in parallel (i.e. meta vs refs).
594 for name, oid in tuple(repo.refs([], limit_to_heads=True)):
595 assert(name.startswith(b'refs/heads/'))
596 yield name[11:], _revlist_item_from_oid(repo, oid, want_meta)
604 if ref in (b'.', b'.tag'):
606 it = repo.cat(b'refs/heads/' + ref)
607 oidx, typ, size = next(it)
611 assert typ == b'commit'
612 commit = parse_commit(b''.join(it))
613 yield ref, _revlist_item_from_oid(repo, unhexlify(oidx), want_meta)
615 def ordered_tree_entries(tree_data, bupm=None):
616 """Yields (name, mangled_name, kind, gitmode, oid) for each item in
617 tree, sorted by name.
620 # Sadly, the .bupm entries currently aren't in git tree order,
621 # i.e. they don't account for the fact that git sorts trees
622 # (including our chunked trees) as if their names ended with "/",
623 # so "fo" sorts after "fo." iff fo is a directory. This makes
624 # streaming impossible when we need the metadata.
625 def result_from_tree_entry(tree_entry):
626 gitmode, mangled_name, oid = tree_entry
627 name, kind = git.demangle_name(mangled_name, gitmode)
628 return name, mangled_name, kind, gitmode, oid
630 tree_ents = (result_from_tree_entry(x) for x in tree_decode(tree_data))
632 tree_ents = sorted(tree_ents, key=lambda x: x[0])
633 for ent in tree_ents:
636 def tree_items(oid, tree_data, names=frozenset(), bupm=None):
638 def tree_item(ent_oid, kind, gitmode):
639 if kind == BUP_CHUNKED:
640 meta = Metadata.read(bupm) if bupm else default_file_mode
641 return Chunky(oid=ent_oid, meta=meta)
644 # No metadata here (accessable via '.' inside ent_oid).
645 return Item(meta=default_dir_mode, oid=ent_oid)
647 meta = Metadata.read(bupm) if bupm else None
648 # handle the case of metadata being empty/missing in bupm
649 # (or there not being bupm at all)
651 meta = _default_mode_for_gitmode(gitmode)
652 return Item(oid=ent_oid, meta=meta)
654 assert len(oid) == 20
656 dot_meta = _read_dir_meta(bupm) if bupm else default_dir_mode
657 yield b'.', Item(oid=oid, meta=dot_meta)
658 tree_entries = ordered_tree_entries(tree_data, bupm)
659 for name, mangled_name, kind, gitmode, ent_oid in tree_entries:
660 if mangled_name == b'.bupm':
663 yield name, tree_item(ent_oid, kind, gitmode)
666 # Assumes the tree is properly formed, i.e. there are no
667 # duplicates, and entries will be in git tree order.
668 if type(names) not in (frozenset, set):
669 names = frozenset(names)
670 remaining = len(names)
672 # Account for the bupm sort order issue (cf. ordered_tree_entries above)
673 last_name = max(names) if bupm else max(names) + b'/'
676 dot_meta = _read_dir_meta(bupm) if bupm else default_dir_mode
677 yield b'.', Item(oid=oid, meta=dot_meta)
682 tree_entries = ordered_tree_entries(tree_data, bupm)
683 for name, mangled_name, kind, gitmode, ent_oid in tree_entries:
684 if mangled_name == b'.bupm':
687 if name not in names:
689 break # given bupm sort order, we're finished
690 if (kind == BUP_CHUNKED or not S_ISDIR(gitmode)) and bupm:
693 yield name, tree_item(ent_oid, kind, gitmode)
698 def tree_items_with_meta(repo, oid, tree_data, names):
699 # For now, the .bupm order doesn't quite match git's, and we don't
700 # load the tree data incrementally anyway, so we just work in RAM
702 assert len(oid) == 20
704 for _, mangled_name, sub_oid in tree_decode(tree_data):
705 if mangled_name == b'.bupm':
706 bupm = _FileReader(repo, sub_oid)
708 if mangled_name > b'.bupm':
710 for item in tree_items(oid, tree_data, names, bupm):
713 _save_name_rx = re.compile(br'^\d\d\d\d-\d\d-\d\d-\d{6}(-\d+)?$')
715 def _reverse_suffix_duplicates(strs):
716 """Yields the elements of strs, with any runs of duplicate values
717 suffixed with -N suffixes, where the zero padded integer N
718 decreases to 0 by 1 (e.g. 10, 09, ..., 00).
721 for name, duplicates in groupby(strs):
722 ndup = len(tuple(duplicates))
726 ndig = len(str(ndup - 1))
727 fmt = b'%s-' + b'%0' + (b'%d' % ndig) + b'd'
728 for i in range(ndup - 1, -1, -1):
729 yield fmt % (name, i)
732 items = f.readline().split(None)
733 assert len(items) == 2
734 tree, auth_sec = items
735 return unhexlify(tree), int(auth_sec)
737 def _name_for_rev(rev):
738 commit_oidx, (tree_oid, utc) = rev
739 return strftime('%Y-%m-%d-%H%M%S', localtime(utc)).encode('ascii')
741 def _item_for_rev(rev):
742 commit_oidx, (tree_oid, utc) = rev
743 coid = unhexlify(commit_oidx)
744 item = cache_get_commit_item(coid, need_meta=False)
747 item = Commit(meta=default_dir_mode, oid=tree_oid, coid=coid)
748 commit_key = b'itm:' + coid
749 cache_notice(commit_key, item)
752 def cache_commit(repo, oid):
753 """Build, cache, and return a "name -> commit_item" dict of the entire
757 # For now, always cache with full metadata
759 entries[b'.'] = _revlist_item_from_oid(repo, oid, True)
760 revs = repo.rev_list((hexlify(oid),), format=b'%T %at',
762 rev_items, rev_names = tee(revs)
763 revs = None # Don't disturb the tees
764 rev_names = _reverse_suffix_duplicates(_name_for_rev(x) for x in rev_names)
765 rev_items = (_item_for_rev(x) for x in rev_items)
767 for item in rev_items:
768 name = next(rev_names)
769 tip = tip or (name, item)
771 entries[b'latest'] = FakeLink(meta=default_symlink_mode, target=tip[0])
772 revlist_key = b'rvl:' + tip[1].coid
773 cache_notice(revlist_key, entries)
776 def revlist_items(repo, oid, names):
777 assert len(oid) == 20
779 # Special case '.' instead of caching the whole history since it's
780 # the only way to get the metadata for the commit.
781 if names and all(x == b'.' for x in names):
782 yield b'.', _revlist_item_from_oid(repo, oid, True)
785 # For now, don't worry about the possibility of the contents being
786 # "too big" for the cache.
787 revlist_key = b'rvl:' + oid
788 entries = cache_get(revlist_key)
790 entries = cache_commit(repo, oid)
793 for name in sorted(entries.keys()):
794 yield name, entries[name]
797 names = frozenset(name for name in names
798 if _save_name_rx.match(name) or name in (b'.', b'latest'))
801 yield b'.', entries[b'.']
802 for name in (n for n in names if n != b'.'):
803 commit = entries.get(name)
807 def tags_items(repo, names):
811 assert len(oid) == 20
814 _, typ, size = next(it)
816 return cache_get_commit_item(oid, need_meta=False) \
817 or _commit_item_from_data(oid, b''.join(it))
820 return Item(meta=default_file_mode, oid=oid)
822 return Item(meta=default_dir_mode, oid=oid)
823 raise Exception('unexpected tag type ' + typ.decode('ascii')
824 + ' for tag ' + path_msg(name))
828 # We have to pull these all into ram because tag_item calls cat()
829 for name, oid in tuple(repo.refs(names, limit_to_tags=True)):
830 assert(name.startswith(b'refs/tags/'))
832 yield name, tag_item(oid)
835 # Assumes no duplicate refs
836 if type(names) not in (frozenset, set):
837 names = frozenset(names)
838 remaining = len(names)
839 last_name = max(names)
846 for name, oid in repo.refs(names, limit_to_tags=True):
847 assert(name.startswith(b'refs/tags/'))
851 if name not in names:
853 yield name, tag_item(oid)
858 def contents(repo, item, names=None, want_meta=True):
859 """Yields information about the items contained in item. Yields
860 (name, item) for each name in names, if the name exists, in an
861 unspecified order. If there are no names, then yields (name,
862 item) for all items, including, a first item named '.'
863 representing the container itself.
865 The meta value for any directories other than '.' will be a
866 default directory mode, not a Metadata object. This is because
867 the actual metadata for a directory is stored inside the directory
868 (see fill_in_metadata_if_dir() or ensure_item_has_metadata()).
870 Note that want_meta is advisory. For any given item, item.meta
871 might be a Metadata instance or a mode, and if the former,
872 meta.size might be None. Missing sizes can be computed via via
873 item_size() or augment_item_meta(..., include_size=True).
875 Do not modify any item.meta Metadata instances directly. If
876 needed, make a copy via item.meta.copy() and modify that instead.
879 # Q: are we comfortable promising '.' first when no names?
882 assert S_ISDIR(item_mode(item))
884 if item_t in real_tree_types:
885 it = repo.cat(hexlify(item.oid))
886 _, obj_t, size = next(it)
890 # Note: it shouldn't be possible to see an Item with type
891 # 'commit' since a 'commit' should always produce a Commit.
892 raise Exception('unexpected git ' + obj_t.decode('ascii'))
894 item_gen = tree_items_with_meta(repo, item.oid, data, names)
896 item_gen = tree_items(item.oid, data, names)
897 elif item_t == RevList:
898 item_gen = revlist_items(repo, item.oid, names)
900 item_gen = root_items(repo, names, want_meta)
902 item_gen = tags_items(repo, names)
904 raise Exception('unexpected VFS item ' + str(item))
908 def _resolve_path(repo, path, parent=None, want_meta=True, follow=True):
909 cache_key = b'res:%d%d%d:%s\0%s' \
910 % (bool(want_meta), bool(follow), repo.id(),
911 (b'/'.join(x[0] for x in parent) if parent else b''),
913 resolution = cache_get(cache_key)
917 def notice_resolution(r):
918 cache_notice(cache_key, r)
921 def raise_dir_required_but_not_dir(path, parent, past):
922 raise IOError(ENOTDIR,
923 "path %s%s resolves to non-directory %r"
925 ' (relative to %r)' % parent if parent else '',
934 assert type(x[0]) in (bytes, str)
935 assert type(x[1]) in item_types
936 assert parent[0][1] == _root
937 if not S_ISDIR(item_mode(parent[-1][1])):
938 raise IOError(ENOTDIR,
939 'path resolution parent %r is not a directory'
941 is_absolute, must_be_dir, future = _decompose_path(path)
944 if not future: # path was effectively '.' or '/'
946 return notice_resolution(((b'', _root),))
948 return notice_resolution(tuple(parent))
949 return notice_resolution(((b'', _root),))
951 past = [(b'', _root)]
953 past = list(parent) if parent else [(b'', _root)]
957 if must_be_dir and not S_ISDIR(item_mode(past[-1][1])):
958 raise_dir_required_but_not_dir(path, parent, past)
959 return notice_resolution(tuple(past))
960 segment = future.pop()
963 if len(past) > 1: # .. from / is /
964 assert S_ISDIR(item_mode(past[-1][1]))
967 parent_name, parent_item = past[-1]
968 wanted = (segment,) if not want_meta else (b'.', segment)
969 items = tuple(contents(repo, parent_item, names=wanted,
970 want_meta=want_meta))
972 item = items[0][1] if items else None
973 else: # First item will be '.' and have the metadata
974 item = items[1][1] if len(items) == 2 else None
975 dot, dot_item = items[0]
977 past[-1] = parent_name, parent_item
979 past.append((segment, None),)
980 return notice_resolution(tuple(past))
981 mode = item_mode(item)
982 if not S_ISLNK(mode):
983 if not S_ISDIR(mode):
984 past.append((segment, item),)
986 raise IOError(ENOTDIR,
987 'path %r%s ends internally in non-directory here: %r'
989 ' (relative to %r)' % parent if parent else '',
993 raise_dir_required_but_not_dir(path, parent, past)
994 return notice_resolution(tuple(past))
996 if want_meta and type(item) in real_tree_types:
997 dir_meta = _find_treeish_oid_metadata(repo, item.oid)
999 item = item._replace(meta=dir_meta)
1000 past.append((segment, item))
1002 if not future and not follow:
1003 past.append((segment, item),)
1006 raise IOError(ELOOP,
1007 'too many symlinks encountered while resolving %r%s'
1008 % (path, ' relative to %r' % parent if parent else ''),
1009 terminus=tuple(past + [(segment, item)]))
1010 target = readlink(repo, item)
1011 is_absolute, _, target_future = _decompose_path(target)
1013 if not target_future: # path was effectively '/'
1014 return notice_resolution(((b'', _root),))
1015 past = [(b'', _root)]
1016 future = target_future
1018 future.extend(target_future)
1021 def resolve(repo, path, parent=None, want_meta=True, follow=True):
1022 """Follow the path in the virtual filesystem and return a tuple
1023 representing the location, if any, denoted by the path. Each
1024 element in the result tuple will be (name, info), where info will
1025 be a VFS item that can be passed to functions like item_mode().
1027 If follow is false, and if the final path element is a symbolic
1028 link, don't follow it, just return it in the result.
1030 If a path segment that does not exist is encountered during
1031 resolution, the result will represent the location of the missing
1032 item, and that item in the result will be None.
1034 Any attempt to traverse a non-directory will raise a VFS ENOTDIR
1037 Any symlinks along the path, including at the end, will be
1038 resolved. A VFS IOError with the errno attribute set to ELOOP
1039 will be raised if too many symlinks are traversed while following
1040 the path. That exception is effectively like a normal
1041 ELOOP IOError exception, but will include a terminus element
1042 describing the location of the failure, which will be a tuple of
1043 (name, info) elements.
1045 The parent, if specified, must be a sequence of (name, item)
1046 tuples, and will provide the starting point for the resolution of
1047 the path. If no parent is specified, resolution will start at
1050 The result may include elements of parent directly, so they must
1051 not be modified later. If this is a concern, pass in "name,
1052 copy_item(item) for name, item in parent" instead.
1054 When want_meta is true, detailed metadata will be included in each
1055 result item if it's avaiable, otherwise item.meta will be an
1056 integer mode. The metadata size may or may not be provided, but
1057 can be computed by item_size() or augment_item_meta(...,
1058 include_size=True). Setting want_meta=False is rarely desirable
1059 since it can limit the VFS to just the metadata git itself can
1060 represent, and so, as an example, fifos and sockets will appear to
1061 be regular files (e.g. S_ISREG(item_mode(item)) will be true) .
1062 But the option is provided because it may be more efficient when
1063 only the path names or the more limited metadata is sufficient.
1065 Do not modify any item.meta Metadata instances directly. If
1066 needed, make a copy via item.meta.copy() and modify that instead.
1069 if repo.is_remote():
1070 # Redirect to the more efficient remote version
1071 return repo.resolve(path, parent=parent, want_meta=want_meta,
1073 result = _resolve_path(repo, path, parent=parent, want_meta=want_meta,
1075 _, leaf_item = result[-1]
1076 if leaf_item and follow:
1077 assert not S_ISLNK(item_mode(leaf_item))
1080 def try_resolve(repo, path, parent=None, want_meta=True):
1081 """If path does not refer to a symlink, does not exist, or refers to a
1082 valid symlink, behave exactly like resolve(..., follow=True). If
1083 path refers to an invalid symlink, behave like resolve(...,
1087 res = resolve(repo, path, parent=parent, want_meta=want_meta, follow=False)
1088 leaf_name, leaf_item = res[-1]
1091 if not S_ISLNK(item_mode(leaf_item)):
1093 follow = resolve(repo, leaf_name, parent=res[:-1], want_meta=want_meta)
1094 follow_name, follow_item = follow[-1]
1099 def augment_item_meta(repo, item, include_size=False):
1100 """Ensure item has a Metadata instance for item.meta. If item.meta is
1101 currently a mode, replace it with a compatible "fake" Metadata
1102 instance. If include_size is true, ensure item.meta.size is
1103 correct, computing it if needed. If item.meta is a Metadata
1104 instance, this call may modify it in place or replace it.
1107 # If we actually had parallelism, we'd need locking...
1110 if isinstance(m, Metadata):
1111 if include_size and m.size is None:
1112 m.size = _compute_item_size(repo, item)
1113 return item._replace(meta=m)
1118 meta.uid = meta.gid = meta.atime = meta.mtime = meta.ctime = 0
1120 if isinstance(item, FakeLink):
1121 target = item.target
1123 target = _readlink(repo, item.oid)
1124 meta.symlink_target = target
1125 meta.size = len(target)
1127 meta.size = _compute_item_size(repo, item)
1128 return item._replace(meta=meta)
1130 def fill_in_metadata_if_dir(repo, item):
1131 """If item is a directory and item.meta is not a Metadata instance,
1132 attempt to find the metadata for the directory. If found, return
1133 a new item augmented to include that metadata. Otherwise, return
1134 item. May be useful for the output of contents().
1137 if S_ISDIR(item_mode(item)) and not isinstance(item.meta, Metadata):
1138 items = tuple(contents(repo, item, (b'.',), want_meta=True))
1139 assert len(items) == 1
1140 assert items[0][0] == b'.'
1144 def ensure_item_has_metadata(repo, item, include_size=False):
1145 """If item is a directory, attempt to find and add its metadata. If
1146 the item still doesn't have a Metadata instance for item.meta,
1147 give it one via augment_item_meta(). May be useful for the output
1151 return augment_item_meta(repo,
1152 fill_in_metadata_if_dir(repo, item),
1153 include_size=include_size)