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
20 The want_meta argument is advisory for calls that accept it, and it
21 may not be honored. Callers must be able to handle an item.meta value
22 that is either an instance of Metadata or an integer mode, perhaps
23 via item_mode() or augment_item_meta().
25 Setting want_meta=False is rarely desirable since it can limit the VFS
26 to only the metadata that git itself can represent, and so for
27 example, fifos and sockets will appear to be regular files
28 (e.g. S_ISREG(item_mode(item)) will be true). But the option is still
29 provided because it may be more efficient when just the path names or
30 the more limited metadata is sufficient.
32 Any given metadata object's size may be None, in which case the size
33 can be computed via item_size() or augment_item_meta(...,
36 When traversing a directory using functions like contents(), the meta
37 value for any directories other than '.' will be a default directory
38 mode, not a Metadata object. This is because the actual metadata for
39 a directory is stored inside the directory (see
40 fill_in_metadata_if_dir() or ensure_item_has_metadata()).
42 Commit items represent commits (e.g. /.tag/some-commit or
43 /foo/latest), and for most purposes, they appear as the underlying
44 tree. S_ISDIR(item_mode(item)) will return true for both tree Items
45 and Commits and the commit's oid is the tree hash; the commit hash is
50 from __future__ import absolute_import, print_function
51 from collections import namedtuple
52 from errno import ELOOP, ENOENT, ENOTDIR
53 from itertools import chain, dropwhile, groupby, izip, 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
57 import exceptions, re, sys
59 from bup import client, git, metadata
60 from bup.compat import 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.metadata import Metadata
64 from bup.repo import LocalRepo, RemoteRepo
67 class IOError(exceptions.IOError):
68 def __init__(self, errno, message, terminus=None):
69 exceptions.IOError.__init__(self, errno, message)
70 self.terminus = terminus
72 default_file_mode = S_IFREG | 0o644
73 default_dir_mode = S_IFDIR | 0o755
74 default_symlink_mode = S_IFLNK | 0o755
76 def _default_mode_for_gitmode(gitmode):
78 return default_file_mode
80 return default_dir_mode
82 return default_symlink_mode
83 raise Exception('unexpected git mode ' + oct(gitmode))
85 def _normal_or_chunked_file_size(repo, oid):
86 """Return the size of the normal or chunked file indicated by oid."""
87 # FIXME: --batch-format CatPipe?
88 it = repo.cat(oid.encode('hex'))
89 _, obj_t, size = next(it)
91 while obj_t == 'tree':
92 mode, name, last_oid = last(tree_decode(''.join(it)))
94 it = repo.cat(last_oid.encode('hex'))
95 _, obj_t, size = next(it)
96 return ofs + sum(len(b) for b in it)
98 def _tree_chunks(repo, tree, startofs):
99 "Tree should be a sequence of (name, mode, hash) as per tree_decode()."
100 assert(startofs >= 0)
101 # name is the chunk's hex offset in the original file
102 tree = dropwhile(lambda (_1, name, _2): int(name, 16) < startofs, tree)
103 for mode, name, oid in tree:
105 skipmore = startofs - ofs
108 it = repo.cat(oid.encode('hex'))
109 _, obj_t, size = next(it)
112 assert obj_t == 'tree'
113 for b in _tree_chunks(repo, tree_decode(data), skipmore):
116 assert obj_t == 'blob'
117 yield data[skipmore:]
120 def __init__(self, repo, oid, startofs):
121 it = repo.cat(oid.encode('hex'))
122 _, obj_t, size = next(it)
123 isdir = obj_t == 'tree'
126 self.it = _tree_chunks(repo, tree_decode(data), startofs)
130 self.blob = data[startofs:]
133 def next(self, size):
135 while len(out) < size:
136 if self.it and not self.blob:
138 self.blob = self.it.next()
139 except StopIteration:
142 want = size - len(out)
143 out += self.blob[:want]
144 self.blob = self.blob[want:]
147 debug2('next(%d) returned %d\n' % (size, len(out)))
151 class _FileReader(object):
152 def __init__(self, repo, oid, known_size=None):
153 assert len(oid) == 20
158 self._size = known_size
160 def _compute_size(self):
162 self._size = _normal_or_chunked_file_size(self._repo, self.oid)
167 raise IOError(errno.EINVAL, 'Invalid argument')
168 if ofs > self._compute_size():
169 raise IOError(errno.EINVAL, 'Invalid argument')
175 def read(self, count=-1):
177 count = self._compute_size() - self.ofs
178 if not self.reader or self.reader.ofs != self.ofs:
179 self.reader = _ChunkReader(self._repo, self.oid, self.ofs)
181 buf = self.reader.next(count)
184 raise # our offsets will be all screwed up otherwise
193 def __exit__(self, type, value, traceback):
197 _multiple_slashes_rx = re.compile(r'//+')
199 def _decompose_path(path):
200 """Return a boolean indicating whether the path is absolute, and a
201 reversed list of path elements, omitting any occurrences of "."
202 and ignoring any leading or trailing slash. If the path is
203 effectively '/' or '.', return an empty list.
206 path = re.sub(_multiple_slashes_rx, '/', path)
208 return True, True, []
209 is_absolute = must_be_dir = False
210 if path.startswith('/'):
213 for suffix in ('/', '/.'):
214 if path.endswith(suffix):
216 path = path[:-len(suffix)]
217 parts = [x for x in path.split('/') if x != '.']
220 must_be_dir = True # e.g. path was effectively '.' or '/', etc.
221 return is_absolute, must_be_dir, parts
224 Item = namedtuple('Item', ('meta', 'oid'))
225 Chunky = namedtuple('Chunky', ('meta', 'oid'))
226 Root = namedtuple('Root', ('meta'))
227 Tags = namedtuple('Tags', ('meta'))
228 RevList = namedtuple('RevList', ('meta', 'oid'))
229 Commit = namedtuple('Commit', ('meta', 'oid', 'coid'))
231 item_types = frozenset((Item, Chunky, Root, Tags, RevList, Commit))
232 real_tree_types = frozenset((Item, Commit))
234 _root = Root(meta=default_dir_mode)
235 _tags = Tags(meta=default_dir_mode)
240 ### A general purpose shared cache with (currently) cheap random
241 ### eviction. There is currently no weighting so a single commit item
242 ### is just as likely to be evicted as an entire "rev-list". See
243 ### is_valid_cache_key for a description of the expected content.
247 _cache_max_items = 30000
250 global _cache, _cache_keys
254 def is_valid_cache_key(x):
255 """Return logically true if x looks like it could be a valid cache key
256 (with respect to structure). Current valid cache entries:
258 commit_oid + ':r' -> rev-list
259 i.e. rev-list -> {'.', commit, '2012...', next_commit, ...}
261 # Suspect we may eventually add "(container_oid, name) -> ...", and others.
266 if len(x) == 22 and x.endswith(b':r'):
271 assert is_valid_cache_key(key)
272 return _cache.get(key)
274 def cache_notice(key, value):
275 global _cache, _cache_keys, _cache_max_items
276 assert is_valid_cache_key(key)
279 if len(_cache) < _cache_max_items:
280 _cache_keys.append(key)
283 victim_i = randrange(0, len(_cache_keys))
284 victim = _cache_keys[victim_i]
286 _cache_keys[victim_i] = key
290 def cache_get_commit_item(oid, need_meta=True):
291 """Return the requested tree item if it can be found in the cache.
292 When need_meta is true don't return a cached item that only has a
294 # tree might be stored independently, or as '.' with its entries.
295 item = cache_get(oid)
299 if isinstance(item.meta, Metadata):
301 entries = cache_get(oid + b':r')
305 def cache_get_revlist_item(oid, need_meta=True):
306 commit = cache_get_commit_item(oid, need_meta=need_meta)
308 return RevList(oid=oid, meta=commit.meta)
312 """Return a completely independent copy of item, such that
313 modifications will not affect the original.
316 meta = getattr(item, 'meta', None)
317 if isinstance(meta, Metadata):
318 return(item._replace(meta=meta.copy()))
322 """Return the integer mode (stat st_mode) for item."""
324 if isinstance(m, Metadata):
328 def _read_dir_meta(bupm):
329 # This is because save writes unmodified Metadata() entries for
330 # fake parents -- test-save-strip-graft.sh demonstrates.
331 m = Metadata.read(bupm)
333 return default_dir_mode
334 assert m.mode is not None
339 def tree_data_and_bupm(repo, oid):
340 """Return (tree_bytes, bupm_oid) where bupm_oid will be None if the
341 tree has no metadata (i.e. older bup save, or non-bup tree).
344 assert len(oid) == 20
345 it = repo.cat(oid.encode('hex'))
346 _, item_t, size = next(it)
348 if item_t == 'commit':
349 commit = parse_commit(data)
350 it = repo.cat(commit.tree)
351 _, item_t, size = next(it)
353 assert item_t == 'tree'
354 elif item_t != 'tree':
355 raise Exception('%r is not a tree or commit' % oid.encode('hex'))
356 for _, mangled_name, sub_oid in tree_decode(data):
357 if mangled_name == '.bupm':
359 if mangled_name > '.bupm':
363 def _find_treeish_oid_metadata(repo, oid):
364 """Return the metadata for the tree or commit oid, or None if the tree
365 has no metadata (i.e. older bup save, or non-bup tree).
368 tree_data, bupm_oid = tree_data_and_bupm(repo, oid)
370 with _FileReader(repo, bupm_oid) as meta_stream:
371 return _read_dir_meta(meta_stream)
374 def _readlink(repo, oid):
375 return ''.join(repo.join(oid.encode('hex')))
377 def readlink(repo, item):
378 """Return the link target of item, which must be a symlink. Reads the
379 target from the repository if necessary."""
381 assert S_ISLNK(item_mode(item))
382 if isinstance(item.meta, Metadata):
383 target = item.meta.symlink_target
386 return _readlink(repo, item.oid)
388 def _compute_item_size(repo, item):
389 mode = item_mode(item)
391 size = _normal_or_chunked_file_size(repo, item.oid)
394 return len(_readlink(repo, item.oid))
397 def item_size(repo, item):
398 """Return the size of item, computing it if necessary."""
400 if isinstance(m, Metadata) and m.size is not None:
402 return _compute_item_size(repo, item)
404 def tree_data_reader(repo, oid):
405 """Return an open reader for all of the data contained within oid. If
406 oid refers to a tree, recursively concatenate all of its contents."""
407 return _FileReader(repo, oid)
409 def fopen(repo, item):
410 """Return an open reader for the given file item."""
411 assert S_ISREG(item_mode(item))
412 return tree_data_reader(repo, item.oid)
414 def _commit_item_from_data(oid, data):
415 info = parse_commit(data)
416 return Commit(meta=default_dir_mode,
417 oid=info.tree.decode('hex'),
420 def _commit_item_from_oid(repo, oid, require_meta):
421 commit = cache_get_commit_item(oid, need_meta=require_meta)
422 if commit and ((not require_meta) or isinstance(commit.meta, Metadata)):
424 it = repo.cat(oid.encode('hex'))
425 _, typ, size = next(it)
426 assert typ == 'commit'
427 commit = _commit_item_from_data(oid, ''.join(it))
429 meta = _find_treeish_oid_metadata(repo, commit.oid)
431 commit = commit._replace(meta=meta)
432 cache_notice(oid, commit)
435 def _revlist_item_from_oid(repo, oid, require_meta):
436 commit = _commit_item_from_oid(repo, oid, require_meta)
437 return RevList(oid=oid, meta=commit.meta)
439 def root_items(repo, names=None, want_meta=True):
440 """Yield (name, item) for the items in '/' in the VFS. Return
441 everything if names is logically false, otherwise return only
442 items with a name in the collection.
445 # FIXME: what about non-leaf refs like 'refs/heads/foo/bar/baz?
451 # FIXME: maybe eventually support repo.clone() or something
452 # and pass in two repos, so we can drop the tuple() and stream
453 # in parallel (i.e. meta vs refs).
454 for name, oid in tuple(repo.refs([], limit_to_heads=True)):
455 assert(name.startswith('refs/heads/'))
456 yield name[11:], _revlist_item_from_oid(repo, oid, want_meta)
464 if ref in ('.', '.tag'):
466 it = repo.cat('refs/heads/' + ref)
467 oidx, typ, size = next(it)
471 assert typ == 'commit'
472 commit = parse_commit(''.join(it))
473 yield ref, _revlist_item_from_oid(repo, oidx.decode('hex'), want_meta)
475 def ordered_tree_entries(tree_data, bupm=None):
476 """Yields (name, mangled_name, kind, gitmode, oid) for each item in
477 tree, sorted by name.
480 # Sadly, the .bupm entries currently aren't in git tree order,
481 # i.e. they don't account for the fact that git sorts trees
482 # (including our chunked trees) as if their names ended with "/",
483 # so "fo" sorts after "fo." iff fo is a directory. This makes
484 # streaming impossible when we need the metadata.
485 def result_from_tree_entry(tree_entry):
486 gitmode, mangled_name, oid = tree_entry
487 name, kind = git.demangle_name(mangled_name, gitmode)
488 return name, mangled_name, kind, gitmode, oid
490 tree_ents = (result_from_tree_entry(x) for x in tree_decode(tree_data))
492 tree_ents = sorted(tree_ents, key=lambda x: x[0])
493 for ent in tree_ents:
496 def tree_items(oid, tree_data, names=frozenset(), bupm=None):
498 def tree_item(ent_oid, kind, gitmode):
499 if kind == BUP_CHUNKED:
500 meta = Metadata.read(bupm) if bupm else default_file_mode
501 return Chunky(oid=ent_oid, meta=meta)
504 # No metadata here (accessable via '.' inside ent_oid).
505 return Item(meta=default_dir_mode, oid=ent_oid)
507 return Item(oid=ent_oid,
508 meta=(Metadata.read(bupm) if bupm \
509 else _default_mode_for_gitmode(gitmode)))
511 assert len(oid) == 20
513 dot_meta = _read_dir_meta(bupm) if bupm else default_dir_mode
514 yield '.', Item(oid=oid, meta=dot_meta)
515 tree_entries = ordered_tree_entries(tree_data, bupm)
516 for name, mangled_name, kind, gitmode, ent_oid in tree_entries:
517 if mangled_name == '.bupm':
520 yield name, tree_item(ent_oid, kind, gitmode)
523 # Assumes the tree is properly formed, i.e. there are no
524 # duplicates, and entries will be in git tree order.
525 if type(names) not in (frozenset, set):
526 names = frozenset(names)
527 remaining = len(names)
529 # Account for the bupm sort order issue (cf. ordered_tree_entries above)
530 last_name = max(names) if bupm else max(names) + '/'
533 dot_meta = _read_dir_meta(bupm) if bupm else default_dir_mode
534 yield '.', Item(oid=oid, meta=dot_meta)
539 tree_entries = ordered_tree_entries(tree_data, bupm)
540 for name, mangled_name, kind, gitmode, ent_oid in tree_entries:
541 if mangled_name == '.bupm':
544 if name not in names:
546 break # given bupm sort order, we're finished
547 if (kind == BUP_CHUNKED or not S_ISDIR(gitmode)) and bupm:
550 yield name, tree_item(ent_oid, kind, gitmode)
555 def tree_items_with_meta(repo, oid, tree_data, names):
556 # For now, the .bupm order doesn't quite match git's, and we don't
557 # load the tree data incrementally anyway, so we just work in RAM
559 assert len(oid) == 20
561 for _, mangled_name, sub_oid in tree_decode(tree_data):
562 if mangled_name == '.bupm':
563 bupm = _FileReader(repo, sub_oid)
565 if mangled_name > '.bupm':
567 for item in tree_items(oid, tree_data, names, bupm):
570 _save_name_rx = re.compile(r'^\d\d\d\d-\d\d-\d\d-\d{6}(-\d+)?$')
572 def _reverse_suffix_duplicates(strs):
573 """Yields the elements of strs, with any runs of duplicate values
574 suffixed with -N suffixes, where the zero padded integer N
575 decreases to 0 by 1 (e.g. 10, 09, ..., 00).
578 for name, duplicates in groupby(strs):
579 ndup = len(tuple(duplicates))
583 ndig = len(str(ndup - 1))
584 fmt = '%s-' + '%0' + str(ndig) + 'd'
585 for i in range(ndup - 1, -1, -1):
586 yield fmt % (name, i)
589 items = f.readline().split(None)
590 assert len(items) == 2
591 tree, auth_sec = items
592 return tree.decode('hex'), int(auth_sec)
594 def _name_for_rev(rev):
595 commit_oidx, (tree_oid, utc) = rev
596 return strftime('%Y-%m-%d-%H%M%S', localtime(utc))
598 def _item_for_rev(rev):
599 commit_oidx, (tree_oid, utc) = rev
600 coid = commit_oidx.decode('hex')
601 item = cache_get_commit_item(coid, need_meta=False)
604 item = Commit(meta=default_dir_mode, oid=tree_oid, coid=coid)
605 cache_notice(item.coid, item)
608 def cache_commit(repo, oid):
609 """Build, cache, and return a "name -> commit_item" dict of the entire
613 # For now, always cache with full metadata
615 entries['.'] = _revlist_item_from_oid(repo, oid, True)
616 revs = repo.rev_list((oid.encode('hex'),), format='%T %at',
618 rev_items, rev_names = tee(revs)
619 revs = None # Don't disturb the tees
620 rev_names = _reverse_suffix_duplicates(_name_for_rev(x) for x in rev_names)
621 rev_items = (_item_for_rev(x) for x in rev_items)
623 for item in rev_items:
624 latest = latest or item
625 name = next(rev_names)
627 entries['latest'] = latest
628 cache_notice(latest.coid + b':r', entries)
631 def revlist_items(repo, oid, names):
632 assert len(oid) == 20
634 # Special case '.' instead of caching the whole history since it's
635 # the only way to get the metadata for the commit.
636 if names and all(x == '.' for x in names):
637 yield '.', _revlist_item_from_oid(repo, oid, True)
640 # For now, don't worry about the possibility of the contents being
641 # "too big" for the cache.
642 entries = cache_get(oid + b':r')
644 entries = cache_commit(repo, oid)
647 for name in sorted(entries.keys()):
648 yield name, entries[name]
651 names = frozenset(name for name in names
652 if _save_name_rx.match(name) or name in ('.', 'latest'))
655 yield '.', entries['.']
656 for name in (n for n in names if n != '.'):
657 commit = entries.get(name)
661 def tags_items(repo, names):
665 assert len(oid) == 20
666 oidx = oid.encode('hex')
668 _, typ, size = next(it)
670 return cache_get_commit_item(oid, need_meta=False) \
671 or _commit_item_from_data(oid, ''.join(it))
674 return Item(meta=default_file_mode, oid=oid)
676 return Item(meta=default_dir_mode, oid=oid)
677 raise Exception('unexpected tag type ' + typ + ' for tag ' + name)
681 # We have to pull these all into ram because tag_item calls cat()
682 for name, oid in tuple(repo.refs(names, limit_to_tags=True)):
683 assert(name.startswith('refs/tags/'))
685 yield name, tag_item(oid)
688 # Assumes no duplicate refs
689 if type(names) not in (frozenset, set):
690 names = frozenset(names)
691 remaining = len(names)
692 last_name = max(names)
699 for name, oid in repo.refs(names, limit_to_tags=True):
700 assert(name.startswith('refs/tags/'))
704 if name not in names:
706 yield name, tag_item(oid)
711 def contents(repo, item, names=None, want_meta=True):
712 """Yields information about the items contained in item. Yields
713 (name, item) for each name in names, if the name exists, in an
714 unspecified order. If there are no names, then yields (name,
715 item) for all items, including, a first item named '.'
716 representing the container itself.
718 The meta value for any directories other than '.' will be a
719 default directory mode, not a Metadata object. This is because
720 the actual metadata for a directory is stored inside the directory
721 (see fill_in_metadata_if_dir() or ensure_item_has_metadata()).
723 Note that want_meta is advisory. For any given item, item.meta
724 might be a Metadata instance or a mode, and if the former,
725 meta.size might be None. Missing sizes can be computed via via
726 item_size() or augment_item_meta(..., include_size=True).
728 Do not modify any item.meta Metadata instances directly. If
729 needed, make a copy via item.meta.copy() and modify that instead.
732 # Q: are we comfortable promising '.' first when no names?
735 assert S_ISDIR(item_mode(item))
738 if item_t in real_tree_types:
739 it = repo.cat(item.oid.encode('hex'))
740 _, obj_type, size = next(it)
742 if obj_type == 'tree':
744 item_gen = tree_items_with_meta(repo, item.oid, data, names)
746 item_gen = tree_items(item.oid, data, names)
747 elif obj_type == 'commit':
749 item_gen = tree_items_with_meta(repo, item.oid, tree_data, names)
751 item_gen = tree_items(item.oid, tree_data, names)
754 raise Exception('unexpected git ' + obj_type)
755 elif item_t == RevList:
756 item_gen = revlist_items(repo, item.oid, names)
758 item_gen = root_items(repo, names, want_meta)
760 item_gen = tags_items(repo, names)
762 raise Exception('unexpected VFS item ' + str(item))
766 def _resolve_path(repo, path, parent=None, want_meta=True, deref=False):
767 def raise_dir_required_but_not_dir(path, parent, past):
768 raise IOError(ENOTDIR,
769 "path %r%s resolves to non-directory %r"
771 ' (relative to %r)' % parent if parent else '',
780 assert type(x[0]) in (bytes, str)
781 assert type(x[1]) in item_types
782 assert parent[0][1] == _root
783 if not S_ISDIR(item_mode(parent[-1][1])):
784 raise IOError(ENOTDIR,
785 'path resolution parent %r is not a directory'
787 is_absolute, must_be_dir, future = _decompose_path(path)
790 if not future: # path was effectively '.' or '/'
792 return (('', _root),)
799 past = list(parent) if parent else [('', _root)]
803 if must_be_dir and not S_ISDIR(item_mode(past[-1][1])):
804 raise_dir_required_but_not_dir(path, parent, past)
806 segment = future.pop()
809 if len(past) > 1: # .. from / is /
810 assert S_ISDIR(item_mode(past[-1][1]))
813 parent_name, parent_item = past[-1]
814 wanted = (segment,) if not want_meta else ('.', segment)
815 items = tuple(contents(repo, parent_item, names=wanted,
816 want_meta=want_meta))
818 item = items[0][1] if items else None
819 else: # First item will be '.' and have the metadata
820 item = items[1][1] if len(items) == 2 else None
821 dot, dot_item = items[0]
823 past[-1] = parent_name, parent_item
825 past.append((segment, None),)
827 mode = item_mode(item)
828 if not S_ISLNK(mode):
829 if not S_ISDIR(mode):
830 past.append((segment, item),)
832 raise IOError(ENOTDIR,
833 'path %r%s ends internally in non-directory here: %r'
835 ' (relative to %r)' % parent if parent else '',
839 raise_dir_required_but_not_dir(path, parent, past)
842 if want_meta and type(item) in real_tree_types:
843 dir_meta = _find_treeish_oid_metadata(repo, item.oid)
845 item = item._replace(meta=dir_meta)
846 past.append((segment, item))
848 if not future and not deref:
849 past.append((segment, item),)
853 'too many symlinks encountered while resolving %r%s'
854 % (path, ' relative to %r' % parent if parent else ''),
855 terminus=tuple(past + [(segment, item)]))
856 target = readlink(repo, item)
857 is_absolute, _, target_future = _decompose_path(target)
859 if not target_future: # path was effectively '/'
860 return (('', _root),)
862 future = target_future
864 future.extend(target_future)
867 def lresolve(repo, path, parent=None, want_meta=True):
868 """Perform exactly the same function as resolve(), except if the final
869 path element is a symbolic link, don't follow it, just return it
873 return _resolve_path(repo, path, parent=parent, want_meta=want_meta,
876 def resolve(repo, path, parent=None, want_meta=True):
877 """Follow the path in the virtual filesystem and return a tuple
878 representing the location, if any, denoted by the path. Each
879 element in the result tuple will be (name, info), where info will
880 be a VFS item that can be passed to functions like item_mode().
882 If a path segment that does not exist is encountered during
883 resolution, the result will represent the location of the missing
884 item, and that item in the result will be None.
886 Any attempt to traverse a non-directory will raise a VFS ENOTDIR
889 Any symlinks along the path, including at the end, will be
890 resolved. A VFS IOError with the errno attribute set to ELOOP
891 will be raised if too many symlinks are traversed while following
892 the path. That exception is effectively like a normal
893 ELOOP IOError exception, but will include a terminus element
894 describing the location of the failure, which will be a tuple of
895 (name, info) elements.
897 The parent, if specified, must be a sequence of (name, item)
898 tuples, and will provide the starting point for the resolution of
899 the path. If no parent is specified, resolution will start at
902 The result may include elements of parent directly, so they must
903 not be modified later. If this is a concern, pass in "name,
904 copy_item(item) for name, item in parent" instead.
906 When want_meta is true, detailed metadata will be included in each
907 result item if it's avaiable, otherwise item.meta will be an
908 integer mode. The metadata size may or may not be provided, but
909 can be computed by item_size() or augment_item_meta(...,
910 include_size=True). Setting want_meta=False is rarely desirable
911 since it can limit the VFS to just the metadata git itself can
912 represent, and so, as an example, fifos and sockets will appear to
913 be regular files (e.g. S_ISREG(item_mode(item)) will be true) .
914 But the option is provided because it may be more efficient when
915 only the path names or the more limited metadata is sufficient.
917 Do not modify any item.meta Metadata instances directly. If
918 needed, make a copy via item.meta.copy() and modify that instead.
921 result = _resolve_path(repo, path, parent=parent, want_meta=want_meta,
923 _, leaf_item = result[-1]
925 assert not S_ISLNK(item_mode(leaf_item))
928 def try_resolve(repo, path, parent=None, want_meta=True):
929 """If path does not refer to a symlink, does not exist, or refers to a
930 valid symlink, behave exactly like resolve(). If path refers to
931 an invalid symlink, behave like lresolve.
934 res = lresolve(repo, path, parent=parent, want_meta=want_meta)
935 leaf_name, leaf_item = res[-1]
938 if not S_ISLNK(item_mode(leaf_item)):
940 deref = resolve(repo, leaf_name, parent=res[:-1], want_meta=want_meta)
941 deref_name, deref_item = deref[-1]
946 def augment_item_meta(repo, item, include_size=False):
947 """Ensure item has a Metadata instance for item.meta. If item.meta is
948 currently a mode, replace it with a compatible "fake" Metadata
949 instance. If include_size is true, ensure item.meta.size is
950 correct, computing it if needed. If item.meta is a Metadata
951 instance, this call may modify it in place or replace it.
954 # If we actually had parallelism, we'd need locking...
957 if isinstance(m, Metadata):
958 if include_size and m.size is None:
959 m.size = _compute_item_size(repo, item)
960 return item._replace(meta=m)
965 meta.uid = meta.gid = meta.atime = meta.mtime = meta.ctime = 0
967 target = _readlink(repo, item.oid)
968 meta.symlink_target = target
969 meta.size = len(target)
971 meta.size = _compute_item_size(repo, item)
972 return item._replace(meta=meta)
974 def fill_in_metadata_if_dir(repo, item):
975 """If item is a directory and item.meta is not a Metadata instance,
976 attempt to find the metadata for the directory. If found, return
977 a new item augmented to include that metadata. Otherwise, return
978 item. May be useful for the output of contents().
981 if S_ISDIR(item_mode(item)) and not isinstance(item.meta, Metadata):
982 items = tuple(contents(repo, item, ('.',), want_meta=True))
983 assert len(items) == 1
984 assert items[0][0] == '.'
988 def ensure_item_has_metadata(repo, item, include_size=False):
989 """If item is a directory, attempt to find and add its metadata. If
990 the item still doesn't have a Metadata instance for item.meta,
991 give it one via augment_item_meta(). May be useful for the output
995 return augment_item_meta(repo,
996 fill_in_metadata_if_dir(repo, item),
997 include_size=include_size)