3 from bup import _helpers, helpers
4 from bup.helpers import sc_page_size
6 _fmincore = getattr(helpers, 'fmincore', None)
8 BLOB_MAX = 8192*4 # 8192 is the "typical" blob size for bupsplit
9 BLOB_READ_SIZE = 1024*1024
11 progress_callback = None
14 GIT_MODE_FILE = 0100644
15 GIT_MODE_TREE = 040000
16 GIT_MODE_SYMLINK = 0120000
17 assert(GIT_MODE_TREE != 40000) # 0xxx should be treated as octal
19 # The purpose of this type of buffer is to avoid copying on peek(), get(),
20 # and eat(). We do copy the buffer contents on put(), but that should
21 # be ok if we always only put() large amounts of data at a time.
29 self.data = buffer(self.data, self.start) + s
32 def peek(self, count):
33 return buffer(self.data, self.start, count)
39 v = buffer(self.data, self.start, count)
44 return len(self.data) - self.start
47 def _fadvise_pages_done(fd, first_page, count):
48 assert(first_page >= 0)
51 _helpers.fadvise_done(fd,
52 first_page * sc_page_size,
56 def _nonresident_page_regions(status_bytes, max_region_len=None):
57 """Return (start_page, count) pairs in ascending start_page order for
58 each contiguous region of nonresident pages indicated by the
59 mincore() status_bytes. Limit the number of pages in each region
61 assert(max_region_len is None or max_region_len > 0)
63 for i, x in enumerate(status_bytes):
64 in_core = x & helpers.MINCORE_INCORE
73 elif max_region_len and count >= max_region_len:
77 yield (start, len(status_bytes) - start)
80 def _uncache_ours_upto(fd, offset, first_region, remaining_regions):
81 """Uncache the pages of fd indicated by first_region and
82 remaining_regions that are before offset, where each region is a
83 (start_page, count) pair. The final region must have a start_page
85 rstart, rlen = first_region
86 while rstart is not None and (rstart + rlen) * sc_page_size <= offset:
87 _fadvise_pages_done(fd, rstart, rlen)
88 rstart, rlen = next(remaining_regions, (None, None))
92 def readfile_iter(files, progress=None):
93 for filenum,f in enumerate(files):
96 fd = rpr = rstart = rlen = None
97 if _fmincore and hasattr(f, 'fileno'):
99 max_chunk = max(1, (8 * 1024 * 1024) / sc_page_size)
100 rpr = _nonresident_page_regions(_fmincore(fd), max_chunk)
101 rstart, rlen = next(rpr, (None, None))
104 progress(filenum, len(b))
105 b = f.read(BLOB_READ_SIZE)
108 rstart, rlen = _uncache_ours_upto(fd, ofs, (rstart, rlen), rpr)
113 rstart, rlen = _uncache_ours_upto(fd, ofs, (rstart, rlen), rpr)
116 def _splitbuf(buf, basebits, fanbits):
118 b = buf.peek(buf.used())
119 (ofs, bits) = _helpers.splitbuf(b)
125 level = (bits-basebits)//fanbits # integer division
127 yield buffer(b, 0, ofs), level
130 while buf.used() >= BLOB_MAX:
131 # limit max blob size
132 yield buf.get(BLOB_MAX), 0
135 def _hashsplit_iter(files, progress):
136 assert(BLOB_READ_SIZE > BLOB_MAX)
137 basebits = _helpers.blobbits()
138 fanbits = int(math.log(fanout or 128, 2))
140 for inblock in readfile_iter(files, progress):
142 for buf_and_level in _splitbuf(buf, basebits, fanbits):
145 yield buf.get(buf.used()), 0
148 def _hashsplit_iter_keep_boundaries(files, progress):
149 for real_filenum,f in enumerate(files):
151 def prog(filenum, nbytes):
152 # the inner _hashsplit_iter doesn't know the real file count,
153 # so we'll replace it here.
154 return progress(real_filenum, nbytes)
157 for buf_and_level in _hashsplit_iter([f], progress=prog):
161 def hashsplit_iter(files, keep_boundaries, progress):
163 return _hashsplit_iter_keep_boundaries(files, progress)
165 return _hashsplit_iter(files, progress)
169 def split_to_blobs(makeblob, files, keep_boundaries, progress):
171 for (blob, level) in hashsplit_iter(files, keep_boundaries, progress):
173 total_split += len(blob)
174 if progress_callback:
175 progress_callback(len(blob))
176 yield (sha, len(blob), level)
179 def _make_shalist(l):
182 total = sum(size for mode,sha,size, in l)
183 vlen = len('%x' % total)
185 for (mode, sha, size) in l:
186 shalist.append((mode, '%0*x' % (vlen,ofs), sha))
189 return (shalist, total)
192 def _squish(maketree, stacks, n):
194 while i < n or len(stacks[i]) >= MAX_PER_TREE:
195 while len(stacks) <= i+1:
197 if len(stacks[i]) == 1:
198 stacks[i+1] += stacks[i]
200 (shalist, size) = _make_shalist(stacks[i])
201 tree = maketree(shalist)
202 stacks[i+1].append((GIT_MODE_TREE, tree, size))
207 def split_to_shalist(makeblob, maketree, files,
208 keep_boundaries, progress=None):
209 sl = split_to_blobs(makeblob, files, keep_boundaries, progress)
213 for (sha,size,level) in sl:
214 shal.append((GIT_MODE_FILE, sha, size))
215 return _make_shalist(shal)[0]
218 for (sha,size,level) in sl:
219 stacks[0].append((GIT_MODE_FILE, sha, size))
220 _squish(maketree, stacks, level)
221 #log('stacks: %r\n' % [len(i) for i in stacks])
222 _squish(maketree, stacks, len(stacks)-1)
223 #log('stacks: %r\n' % [len(i) for i in stacks])
224 return _make_shalist(stacks[-1])[0]
227 def split_to_blob_or_tree(makeblob, maketree, files,
228 keep_boundaries, progress=None):
229 shalist = list(split_to_shalist(makeblob, maketree,
230 files, keep_boundaries, progress))
231 if len(shalist) == 1:
232 return (shalist[0][0], shalist[0][2])
233 elif len(shalist) == 0:
234 return (GIT_MODE_FILE, makeblob(''))
236 return (GIT_MODE_TREE, maketree(shalist))
239 def open_noatime(name):
240 fd = _helpers.open_noatime(name)
242 return os.fdopen(fd, 'rb', 1024*1024)