X-Git-Url: https://arthur.barton.de/gitweb/?p=bup.git;a=blobdiff_plain;f=lib%2Fbup%2Findex.py;h=ae6201853f45b9f4e7dc7816d03cdc7fda8895ec;hp=a315e8898811548b980a6bfa14a456101ea71499;hb=c7013fde92fc73e0064462954918b09541e9a84f;hpb=9268378e79d20d66647240b926cfcc1d8e95901f diff --git a/lib/bup/index.py b/lib/bup/index.py index a315e88..ae62018 100644 --- a/lib/bup/index.py +++ b/lib/bup/index.py @@ -1,14 +1,17 @@ -import errno, metadata, os, stat, struct, tempfile -from bup import xstat -from bup._helpers import UINT_MAX +from __future__ import absolute_import, print_function +import errno, os, stat, struct, tempfile + +from bup import compat, metadata, xstat +from bup._helpers import UINT_MAX, bytescmp +from bup.compat import range from bup.helpers import (add_error, log, merge_iter, mmap_readwrite, progress, qprogress, resolve_parent, slashappend) -EMPTY_SHA = '\0'*20 -FAKE_SHA = '\x01'*20 +EMPTY_SHA = b'\0' * 20 +FAKE_SHA = b'\x01' * 20 -INDEX_HDR = 'BUPI\0\0\0\6' +INDEX_HDR = b'BUPI\0\0\0\7' # Time values are handled as integer nanoseconds since the epoch in # memory, but are written as xstat/metadata timespecs. This behavior @@ -17,7 +20,21 @@ INDEX_HDR = 'BUPI\0\0\0\6' # Record times (mtime, ctime, atime) as xstat/metadata timespecs, and # store all of the times in the index so they won't interfere with the # forthcoming metadata cache. -INDEX_SIG = '!QQQqQqQqQQII20sHIIQ' +INDEX_SIG = ('!' + 'Q' # dev + 'Q' # ino + 'Q' # nlink + 'qQ' # ctime_s, ctime_ns + 'qQ' # mtime_s, mtime_ns + 'qQ' # atime_s, atime_ns + 'Q' # size + 'I' # mode + 'I' # gitmode + '20s' # sha + 'H' # flags + 'Q' # children_ofs + 'I' # children_n + 'Q') # meta_ofs ENTLEN = struct.calcsize(INDEX_SIG) FOOTER_SIG = '!Q' @@ -148,15 +165,17 @@ def _golevel(level, f, ename, newentry, metastore, tmax): class Entry: def __init__(self, basename, name, meta_ofs, tmax): - self.basename = str(basename) - self.name = str(name) + assert basename is None or type(basename) == bytes + assert name is None or type(name) == bytes + self.basename = basename + self.name = name self.meta_ofs = meta_ofs self.tmax = tmax self.children_ofs = 0 self.children_n = 0 def __repr__(self): - return ("(%s,0x%04x,%d,%d,%d,%d,%d,%d,%s/%s,0x%04x,%d,0x%08x/%d)" + return ("(%r,0x%04x,%d,%d,%d,%d,%d,%d,%s/%s,0x%04x,%d,0x%08x/%d)" % (self.name, self.dev, self.ino, self.nlink, self.ctime, self.mtime, self.atime, self.size, self.mode, self.gitmode, @@ -181,13 +200,34 @@ class Entry: log('pack error: %s (%r)\n' % (e, self)) raise - def from_stat(self, st, meta_ofs, tstart, check_device=True): - old = (self.dev if check_device else 0, - self.ino, self.nlink, self.ctime, self.mtime, - self.size, self.flags & IX_EXISTS) - new = (st.st_dev if check_device else 0, - st.st_ino, st.st_nlink, st.st_ctime, st.st_mtime, - st.st_size, IX_EXISTS) + def stale(self, st, tstart, check_device=True): + if self.size != st.st_size: + return True + if self.mtime != st.st_mtime: + return True + if self.sha == EMPTY_SHA: + return True + if not self.gitmode: + return True + if self.ctime != st.st_ctime: + return True + if self.ino != st.st_ino: + return True + if self.nlink != st.st_nlink: + return True + if not (self.flags & IX_EXISTS): + return True + if check_device and (self.dev != st.st_dev): + return True + # Check that the ctime's "second" is at or after tstart's. + ctime_sec_in_ns = xstat.fstime_floor_secs(st.st_ctime) * 10**9 + if ctime_sec_in_ns >= tstart: + return True + return False + + def update_from_stat(self, st, meta_ofs): + # Should only be called when the entry is stale(), and + # invalidate() should almost certainly be called afterward. self.dev = st.st_dev self.ino = st.st_ino self.nlink = st.st_nlink @@ -198,13 +238,8 @@ class Entry: self.mode = st.st_mode self.flags |= IX_EXISTS self.meta_ofs = meta_ofs - # Check that the ctime's "second" is at or after tstart's. - ctime_sec_in_ns = xstat.fstime_floor_secs(st.st_ctime) * 10**9 - if ctime_sec_in_ns >= tstart or old != new \ - or self.sha == EMPTY_SHA or not self.gitmode: - self.invalidate() self._fixup() - + def _fixup(self): self.mtime = self._fixup_time(self.mtime) self.ctime = self._fixup_time(self.ctime) @@ -249,13 +284,39 @@ class Entry: def is_fake(self): return not self.ctime - def __cmp__(a, b): - return (cmp(b.name, a.name) - or cmp(a.is_valid(), b.is_valid()) - or cmp(a.is_fake(), b.is_fake())) + def _cmp(self, other): + # Note reversed name ordering + bc = bytescmp(other.name, self.name) + if bc != 0: + return bc + vc = self.is_valid() - other.is_valid() + if vc != 0: + return vc + fc = self.is_fake() - other.is_fake() + if fc != 0: + return fc + return 0 + + def __eq__(self, other): + return self._cmp(other) == 0 + + def __ne__(self, other): + return self._cmp(other) != 0 + + def __lt__(self, other): + return self._cmp(other) < 0 + + def __gt__(self, other): + return self._cmp(other) > 0 + + def __le__(self, other): + return self._cmp(other) <= 0 + + def __ge__(self, other): + return self._cmp(other) >= 0 def write(self, f): - f.write(self.basename + '\0' + self.packed()) + f.write(self.basename + b'\0' + self.packed()) class NewEntry(Entry): @@ -289,7 +350,7 @@ class ExistingEntry(Entry): self.ctime, ctime_ns, self.mtime, mtime_ns, self.atime, atime_ns, self.size, self.mode, self.gitmode, self.sha, self.flags, self.children_ofs, self.children_n, self.meta_ofs - ) = struct.unpack(INDEX_SIG, str(buffer(m, ofs, ENTLEN))) + ) = struct.unpack(INDEX_SIG, m[ofs : ofs + ENTLEN]) self.atime = xstat.timespec_to_nsecs((self.atime, atime_ns)) self.mtime = xstat.timespec_to_nsecs((self.mtime, mtime_ns)) self.ctime = xstat.timespec_to_nsecs((self.ctime, ctime_ns)) @@ -320,22 +381,22 @@ class ExistingEntry(Entry): def iter(self, name=None, wantrecurse=None): dname = name - if dname and not dname.endswith('/'): - dname += '/' + if dname and not dname.endswith(b'/'): + dname += b'/' ofs = self.children_ofs assert(ofs <= len(self._m)) assert(self.children_n <= UINT_MAX) # i.e. python struct 'I' - for i in xrange(self.children_n): - eon = self._m.find('\0', ofs) + for i in range(self.children_n): + eon = self._m.find(b'\0', ofs) assert(eon >= 0) assert(eon >= ofs) assert(eon > ofs) - basename = str(buffer(self._m, ofs, eon-ofs)) + basename = self._m[ofs : ofs + (eon - ofs)] child = ExistingEntry(self, basename, self.name + basename, self._m, eon+1) if (not dname or child.name.startswith(dname) - or child.name.endswith('/') and dname.startswith(child.name)): + or child.name.endswith(b'/') and dname.startswith(child.name)): if not wantrecurse or wantrecurse(child): for e in child.iter(name=name, wantrecurse=wantrecurse): yield e @@ -350,12 +411,12 @@ class ExistingEntry(Entry): class Reader: def __init__(self, filename): self.filename = filename - self.m = '' + self.m = b'' self.writable = False self.count = 0 f = None try: - f = open(filename, 'r+') + f = open(filename, 'rb+') except IOError as e: if e.errno == errno.ENOENT: pass @@ -372,7 +433,8 @@ class Reader: self.m = mmap_readwrite(f) self.writable = True self.count = struct.unpack(FOOTER_SIG, - str(buffer(self.m, st.st_size-FOOTLEN, FOOTLEN)))[0] + self.m[st.st_size - FOOTLEN + : st.st_size])[0] def __del__(self): self.close() @@ -383,20 +445,20 @@ class Reader: def forward_iter(self): ofs = len(INDEX_HDR) while ofs+ENTLEN <= len(self.m)-FOOTLEN: - eon = self.m.find('\0', ofs) + eon = self.m.find(b'\0', ofs) assert(eon >= 0) assert(eon >= ofs) assert(eon > ofs) - basename = str(buffer(self.m, ofs, eon-ofs)) + basename = self.m[ofs : ofs + (eon - ofs)] yield ExistingEntry(None, basename, basename, self.m, eon+1) ofs = eon + 1 + ENTLEN def iter(self, name=None, wantrecurse=None): if len(self.m) > len(INDEX_HDR)+ENTLEN: dname = name - if dname and not dname.endswith('/'): - dname += '/' - root = ExistingEntry(None, '/', '/', + if dname and not dname.endswith(b'/'): + dname += b'/' + root = ExistingEntry(None, b'/', b'/', self.m, len(self.m)-FOOTLEN-ENTLEN) for sub in root.iter(name=name, wantrecurse=wantrecurse): yield sub @@ -446,9 +508,9 @@ class Reader: # in an odd way and depends on a terminating '/' to indicate directories. def pathsplit(p): """Split a path into a list of elements of the file system hierarchy.""" - l = p.split('/') - l = [i+'/' for i in l[:-1]] + l[-1:] - if l[-1] == '': + l = p.split(b'/') + l = [i + b'/' for i in l[:-1]] + l[-1:] + if l[-1] == b'': l.pop() # extra blank caused by terminating '/' return l @@ -464,7 +526,7 @@ class Writer: self.metastore = metastore self.tmax = tmax (dir,name) = os.path.split(filename) - (ffd,self.tmpname) = tempfile.mkstemp('.tmp', filename, dir) + ffd, self.tmpname = tempfile.mkstemp(b'.tmp', filename, dir) self.f = os.fdopen(ffd, 'wb', 65536) self.f.write(INDEX_HDR) @@ -506,7 +568,7 @@ class Writer: self.metastore, self.tmax) def add(self, name, st, meta_ofs, hashgen = None): - endswith = name.endswith('/') + endswith = name.endswith(b'/') ename = pathsplit(name) basename = ename[-1] #log('add: %r %r\n' % (basename, name)) @@ -543,31 +605,48 @@ class Writer: return Reader(self.tmpname) +def _slashappend_or_add_error(p, caller): + """Return p, after ensuring it has a single trailing slash if it names + a directory, unless there's an OSError, in which case, call + add_error() and return None.""" + try: + st = os.lstat(p) + except OSError as e: + add_error('%s: %s' % (caller, e)) + return None + else: + if stat.S_ISDIR(st.st_mode): + return slashappend(p) + return p + + +def unique_resolved_paths(paths): + "Return a collection of unique resolved paths." + rps = (_slashappend_or_add_error(resolve_parent(p), 'unique_resolved_paths') + for p in paths) + return frozenset((x for x in rps if x is not None)) + + def reduce_paths(paths): xpaths = [] for p in paths: - rp = resolve_parent(p) - try: - st = os.lstat(rp) - if stat.S_ISDIR(st.st_mode): - rp = slashappend(rp) - p = slashappend(p) - xpaths.append((rp, p)) - except OSError as e: - add_error('reduce_paths: %s' % e) + rp = _slashappend_or_add_error(resolve_parent(p), 'reduce_paths') + if rp: + xpaths.append((rp, slashappend(p) if rp.endswith(b'/') else p)) xpaths.sort() paths = [] prev = None for (rp, p) in xpaths: if prev and (prev == rp - or (prev.endswith('/') and rp.startswith(prev))): + or (prev.endswith(b'/') and rp.startswith(prev))): continue # already superceded by previous path paths.append((rp, p)) prev = rp paths.sort(reverse=True) return paths + def merge(*iters): def pfunc(count, total): qprogress('bup: merging indexes (%d/%d)\r' % (count, total))