lib/bup/hashsplit.py

   1 import math
   2 from bup import _helpers
   3 from bup.helpers import *
   4
   5 BLOB_MAX = 8192*4   # 8192 is the "typical" blob size for bupsplit
   6 BLOB_READ_SIZE = 1024*1024
   7 MAX_PER_TREE = 256
   8 progress_callback = None
   9 fanout = 16
  10
  11 GIT_MODE_FILE = 0100644
  12 GIT_MODE_TREE = 040000
  13 GIT_MODE_SYMLINK = 0120000
  14 assert(GIT_MODE_TREE != 40000)  # 0xxx should be treated as octal
  15
  16 # The purpose of this type of buffer is to avoid copying on peek(), get(),
  17 # and eat().  We do copy the buffer contents on put(), but that should
  18 # be ok if we always only put() large amounts of data at a time.
  19 class Buf:
  20     def __init__(self):
  21         self.data = ''
  22         self.start = 0
  23
  24     def put(self, s):
  25         if s:
  26             self.data = buffer(self.data, self.start) + s
  27             self.start = 0
  28
  29     def peek(self, count):
  30         return buffer(self.data, self.start, count)
  31
  32     def eat(self, count):
  33         self.start += count
  34
  35     def get(self, count):
  36         v = buffer(self.data, self.start, count)
  37         self.start += count
  38         return v
  39
  40     def used(self):
  41         return len(self.data) - self.start
  42
  43
  44 def readfile_iter(files, progress=None):
  45     for filenum,f in enumerate(files):
  46         ofs = 0
  47         b = ''
  48         while 1:
  49             if progress:
  50                 progress(filenum, len(b))
  51             fadvise_done(f, max(0, ofs - 1024*1024))
  52             b = f.read(BLOB_READ_SIZE)
  53             ofs += len(b)
  54             if not b:
  55                 fadvise_done(f, ofs)
  56                 break
  57             yield b
  58
  59
  60 def _splitbuf(buf, basebits, fanbits):
  61     while 1:
  62         b = buf.peek(buf.used())
  63         (ofs, bits) = _helpers.splitbuf(b)
  64         if ofs:
  65             if ofs > BLOB_MAX:
  66                 ofs = BLOB_MAX
  67                 level = 0
  68             else:
  69                 level = (bits-basebits)//fanbits  # integer division
  70             buf.eat(ofs)
  71             yield buffer(b, 0, ofs), level
  72         else:
  73             break
  74     while buf.used() >= BLOB_MAX:
  75         # limit max blob size
  76         yield buf.get(BLOB_MAX), 0
  77
  78
  79 def _hashsplit_iter(files, progress):
  80     assert(BLOB_READ_SIZE > BLOB_MAX)
  81     basebits = _helpers.blobbits()
  82     fanbits = int(math.log(fanout or 128, 2))
  83     buf = Buf()
  84     for inblock in readfile_iter(files, progress):
  85         buf.put(inblock)
  86         for buf_and_level in _splitbuf(buf, basebits, fanbits):
  87             yield buf_and_level
  88     if buf.used():
  89         yield buf.get(buf.used()), 0
  90
  91
  92 def _hashsplit_iter_keep_boundaries(files, progress):
  93     for real_filenum,f in enumerate(files):
  94         if progress:
  95             def prog(filenum, nbytes):
  96                 # the inner _hashsplit_iter doesn't know the real file count,
  97                 # so we'll replace it here.
  98                 return progress(real_filenum, nbytes)
  99         else:
 100             prog = None
 101         for buf_and_level in _hashsplit_iter([f], progress=prog):
 102             yield buf_and_level
 103
 104
 105 def hashsplit_iter(files, keep_boundaries, progress):
 106     if keep_boundaries:
 107         return _hashsplit_iter_keep_boundaries(files, progress)
 108     else:
 109         return _hashsplit_iter(files, progress)
 110
 111
 112 total_split = 0
 113 def split_to_blobs(makeblob, files, keep_boundaries, progress):
 114     global total_split
 115     for (blob, level) in hashsplit_iter(files, keep_boundaries, progress):
 116         sha = makeblob(blob)
 117         total_split += len(blob)
 118         if progress_callback:
 119             progress_callback(len(blob))
 120         yield (sha, len(blob), level)
 121
 122
 123 def _make_shalist(l):
 124     ofs = 0
 125     l = list(l)
 126     total = sum(size for mode,sha,size, in l)
 127     vlen = len('%x' % total)
 128     shalist = []
 129     for (mode, sha, size) in l:
 130         shalist.append((mode, '%0*x' % (vlen,ofs), sha))
 131         ofs += size
 132     assert(ofs == total)
 133     return (shalist, total)
 134
 135
 136 def _squish(maketree, stacks, n):
 137     i = 0
 138     while i < n or len(stacks[i]) >= MAX_PER_TREE:
 139         while len(stacks) <= i+1:
 140             stacks.append([])
 141         if len(stacks[i]) == 1:
 142             stacks[i+1] += stacks[i]
 143         elif stacks[i]:
 144             (shalist, size) = _make_shalist(stacks[i])
 145             tree = maketree(shalist)
 146             stacks[i+1].append((GIT_MODE_TREE, tree, size))
 147         stacks[i] = []
 148         i += 1
 149
 150
 151 def split_to_shalist(makeblob, maketree, files,
 152                      keep_boundaries, progress=None):
 153     sl = split_to_blobs(makeblob, files, keep_boundaries, progress)
 154     assert(fanout != 0)
 155     if not fanout:
 156         shal = []
 157         for (sha,size,level) in sl:
 158             shal.append((GIT_MODE_FILE, sha, size))
 159         return _make_shalist(shal)[0]
 160     else:
 161         stacks = [[]]
 162         for (sha,size,level) in sl:
 163             stacks[0].append((GIT_MODE_FILE, sha, size))
 164             _squish(maketree, stacks, level)
 165         #log('stacks: %r\n' % [len(i) for i in stacks])
 166         _squish(maketree, stacks, len(stacks)-1)
 167         #log('stacks: %r\n' % [len(i) for i in stacks])
 168         return _make_shalist(stacks[-1])[0]
 169
 170
 171 def split_to_blob_or_tree(makeblob, maketree, files,
 172                           keep_boundaries, progress=None):
 173     shalist = list(split_to_shalist(makeblob, maketree,
 174                                     files, keep_boundaries, progress))
 175     if len(shalist) == 1:
 176         return (shalist[0][0], shalist[0][2])
 177     elif len(shalist) == 0:
 178         return (GIT_MODE_FILE, makeblob(''))
 179     else:
 180         return (GIT_MODE_TREE, maketree(shalist))
 181
 182
 183 def open_noatime(name):
 184     fd = _helpers.open_noatime(name)
 185     try:
 186         return os.fdopen(fd, 'rb', 1024*1024)
 187     except:
 188         try:
 189             os.close(fd)
 190         except:
 191             pass
 192         raise
 193
 194
 195 def fadvise_done(f, ofs):
 196     assert(ofs >= 0)
 197     if ofs > 0 and hasattr(f, 'fileno'):
 198         _helpers.fadvise_done(f.fileno(), ofs)