3
# (C) 2005 Canonical Ltd
5
# based on an idea by Matt Mackall
6
# modified to squish into bzr by Martin Pool
8
# This program is free software; you can redistribute it and/or modify
9
# it under the terms of the GNU General Public License as published by
10
# the Free Software Foundation; either version 2 of the License, or
11
# (at your option) any later version.
13
# This program is distributed in the hope that it will be useful,
14
# but WITHOUT ANY WARRANTY; without even the implied warranty of
15
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16
# GNU General Public License for more details.
18
# You should have received a copy of the GNU General Public License
19
# along with this program; if not, write to the Free Software
20
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23
"""Packed file revision storage.
25
A Revfile holds the text history of a particular source file, such
26
as Makefile. It can represent a tree of text versions for that
27
file, allowing for microbranches within a single repository.
29
This is stored on disk as two files: an index file, and a data file.
30
The index file is short and always read completely into memory; the
31
data file is much longer and only the relevant bits of it,
32
identified by the index file, need to be read.
34
Each text version is identified by the SHA-1 of the full text of
35
that version. It also has a sequence number within the file.
37
The index file has a short header and then a sequence of fixed-length
40
* byte[20] SHA-1 of text (as binary, not hex)
41
* uint32 sequence number this is based on, or -1 for full text
42
* uint32 flags: 1=zlib compressed
43
* uint32 offset in text file of start
44
* uint32 length of compressed delta in text file
49
The header is also 48 bytes for tidyness and easy calculation.
51
Both the index and the text are only ever appended to; a consequence
52
is that sequence numbers are stable references. But not every
53
repository in the world will assign the same sequence numbers,
54
therefore the SHA-1 is the only universally unique reference.
56
This is meant to scale to hold 100,000 revisions of a single file, by
57
which time the index file will be ~4.8MB and a bit big to read
60
Some of the reserved fields could be used to implement a (semi?)
61
balanced tree indexed by SHA1 so we can much more efficiently find the
62
index associated with a particular hash. For 100,000 revs we would be
63
able to find it in about 17 random reads, which is not too bad.
65
This performs pretty well except when trying to calculate deltas of
66
really large files. For that the main thing would be to plug in
67
something faster than difflib, which is after all pure Python.
68
Another approach is to just store the gzipped full text of big files,
69
though perhaps that's too perverse?
71
The iter method here will generally read through the whole index file
72
in one go. With readahead in the kernel and python/libc (typically
73
128kB) this means that there should be no seeks and often only one
74
read() call to get everything into memory.
78
# TODO: Something like pread() would make this slightly simpler and
79
# perhaps more efficient.
81
# TODO: Could also try to mmap things... Might be faster for the
82
# index in particular?
84
# TODO: Some kind of faster lookup of SHAs? The bad thing is that probably means
85
# rewriting existing records, which is not so nice.
87
# TODO: Something to check that regions identified in the index file
88
# completely butt up and do not overlap. Strictly it's not a problem
89
# if there are gaps and that can happen if we're interrupted while
90
# writing to the datafile. Overlapping would be very bad though.
92
# TODO: Shouldn't need to lock if we always write in append mode and
93
# then ftell after writing to see where it went. In any case we
94
# assume the whole branch is protected by a lock.
96
import sys, zlib, struct, mdiff, stat, os, sha
97
from binascii import hexlify, unhexlify
101
_HEADER = "bzr revfile v1\n"
102
_HEADER = _HEADER + ('\xff' * (_RECORDSIZE - len(_HEADER)))
103
_NO_RECORD = 0xFFFFFFFFL
105
# fields in the index record
114
# maximum number of patches in a row before recording a whole text.
118
class RevfileError(Exception):
121
class LimitHitException(Exception):
124
class Revfile(object):
125
def __init__(self, basename, mode):
126
# TODO: Lock file while open
128
# TODO: advise of random access
130
self.basename = basename
132
if mode not in ['r', 'w']:
133
raise RevfileError("invalid open mode %r" % mode)
136
idxname = basename + '.irev'
137
dataname = basename + '.drev'
139
idx_exists = os.path.exists(idxname)
140
data_exists = os.path.exists(dataname)
142
if idx_exists != data_exists:
143
raise RevfileError("half-assed revfile")
147
raise RevfileError("Revfile %r does not exist" % basename)
149
self.idxfile = open(idxname, 'w+b')
150
self.datafile = open(dataname, 'w+b')
152
self.idxfile.write(_HEADER)
160
self.idxfile = open(idxname, diskmode)
161
self.datafile = open(dataname, diskmode)
163
h = self.idxfile.read(_RECORDSIZE)
165
raise RevfileError("bad header %r in index of %r"
166
% (h, self.basename))
169
def _check_index(self, idx):
170
if idx < 0 or idx > len(self):
171
raise RevfileError("invalid index %r" % idx)
173
def _check_write(self):
175
raise RevfileError("%r is open readonly" % self.basename)
178
def find_sha(self, s):
179
assert isinstance(s, str)
182
for idx, idxrec in enumerate(self):
183
if idxrec[I_SHA] == s:
190
def _add_compressed(self, text_sha, data, base, compress):
191
# well, maybe compress
196
# don't do compression if it's too small; it's unlikely to win
197
# enough to be worthwhile
198
compr_data = zlib.compress(data)
199
compr_len = len(compr_data)
200
if compr_len < data_len:
203
##print '- compressed %d -> %d, %.1f%%' \
204
## % (data_len, compr_len, float(compr_len)/float(data_len) * 100.0)
205
return self._add_raw(text_sha, data, base, flags)
209
def _add_raw(self, text_sha, data, base, flags):
210
"""Add pre-processed data, can be either full text or delta.
212
This does the compression if that makes sense."""
214
self.datafile.seek(0, 2) # to end
215
self.idxfile.seek(0, 2)
216
assert self.idxfile.tell() == _RECORDSIZE * (idx + 1)
217
data_offset = self.datafile.tell()
219
assert isinstance(data, str) # not unicode or anything weird
221
self.datafile.write(data)
222
self.datafile.flush()
224
assert isinstance(text_sha, str)
226
entry += struct.pack(">IIII12x", base, flags, data_offset, len(data))
227
assert len(entry) == _RECORDSIZE
229
self.idxfile.write(entry)
236
def _add_full_text(self, text, text_sha, compress):
237
"""Add a full text to the file.
239
This is not compressed against any reference version.
241
Returns the index for that text."""
242
return self._add_compressed(text_sha, text, _NO_RECORD, compress)
246
def _choose_base(self, seed, base):
248
if base == _NO_RECORD:
251
if idxrec[I_BASE] == _NO_RECORD:
254
base = idxrec[I_BASE]
257
return base # relative to this full text
261
def _add_delta(self, text, text_sha, base, compress):
262
"""Add a text stored relative to a previous text."""
263
self._check_index(base)
266
base_text = self.get(base, CHAIN_LIMIT)
267
except LimitHitException:
268
return self._add_full_text(text, text_sha, compress)
270
data = mdiff.bdiff(base_text, text)
273
if True: # paranoid early check for bad diff
274
result = mdiff.bpatch(base_text, data)
275
assert result == text
278
# If the delta is larger than the text, we might as well just
279
# store the text. (OK, the delta might be more compressible,
280
# but the overhead of applying it probably still makes it
281
# bad, and I don't want to compress both of them to find out.)
282
if len(data) >= len(text):
283
return self._add_full_text(text, text_sha, compress)
285
return self._add_compressed(text_sha, data, base, compress)
288
def add(self, text, base=None, compress=True):
289
"""Add a new text to the revfile.
291
If the text is already present them its existing id is
292
returned and the file is not changed.
294
If compress is true then gzip compression will be used if it
297
If a base index is specified, that text *may* be used for
298
delta compression of the new text. Delta compression will
299
only be used if it would be a size win and if the existing
300
base is not at too long of a delta chain already.
307
text_sha = sha.new(text).digest()
309
idx = self.find_sha(text_sha)
310
if idx != _NO_RECORD:
311
# TODO: Optional paranoid mode where we read out that record and make sure
312
# it's the same, in case someone ever breaks SHA-1.
313
return idx # already present
315
# base = self._choose_base(ord(text_sha[0]), base)
317
if base == _NO_RECORD:
318
return self._add_full_text(text, text_sha, compress)
320
return self._add_delta(text, text_sha, base, compress)
324
def get(self, idx, recursion_limit=None):
325
"""Retrieve text of a previous revision.
327
If recursion_limit is an integer then walk back at most that
328
many revisions and then raise LimitHitException, indicating
329
that we ought to record a new file text instead of another
330
delta. Don't use this when trying to get out an existing
334
base = idxrec[I_BASE]
335
if base == _NO_RECORD:
336
text = self._get_full_text(idx, idxrec)
338
text = self._get_patched(idx, idxrec, recursion_limit)
340
if sha.new(text).digest() != idxrec[I_SHA]:
341
raise RevfileError("corrupt SHA-1 digest on record %d in %s"
342
% (idx, self.basename))
348
def _get_raw(self, idx, idxrec):
349
flags = idxrec[I_FLAGS]
351
raise RevfileError("unsupported index flags %#x on index %d"
358
self.datafile.seek(idxrec[I_OFFSET])
360
data = self.datafile.read(l)
362
raise RevfileError("short read %d of %d "
363
"getting text for record %d in %r"
364
% (len(data), l, idx, self.basename))
367
data = zlib.decompress(data)
372
def _get_full_text(self, idx, idxrec):
373
assert idxrec[I_BASE] == _NO_RECORD
375
text = self._get_raw(idx, idxrec)
380
def _get_patched(self, idx, idxrec, recursion_limit):
381
base = idxrec[I_BASE]
383
assert base < idx # no loops!
385
if recursion_limit == None:
388
sub_limit = recursion_limit - 1
390
raise LimitHitException()
392
base_text = self.get(base, sub_limit)
393
patch = self._get_raw(idx, idxrec)
395
text = mdiff.bpatch(base_text, patch)
402
"""Return number of revisions."""
403
l = os.fstat(self.idxfile.fileno())[stat.ST_SIZE]
405
raise RevfileError("bad length %d on index of %r" % (l, self.basename))
407
raise RevfileError("no header present in index of %r" % (self.basename))
408
return int(l / _RECORDSIZE) - 1
411
def __getitem__(self, idx):
412
"""Index by sequence id returns the index field"""
413
## TODO: Can avoid seek if we just moved there...
414
self._seek_index(idx)
415
idxrec = self._read_next_index()
417
raise IndexError("no index %d" % idx)
422
def _seek_index(self, idx):
424
raise RevfileError("invalid index %r" % idx)
425
self.idxfile.seek((idx + 1) * _RECORDSIZE)
430
"""Read back all index records.
432
Do not seek the index file while this is underway!"""
433
## sys.stderr.write(" ** iter called ** \n")
436
idxrec = self._read_next_index()
442
def _read_next_index(self):
443
rec = self.idxfile.read(_RECORDSIZE)
446
elif len(rec) != _RECORDSIZE:
447
raise RevfileError("short read of %d bytes getting index %d from %r"
448
% (len(rec), idx, self.basename))
450
return struct.unpack(">20sIIII12x", rec)
453
def dump(self, f=sys.stdout):
454
f.write('%-8s %-40s %-8s %-8s %-8s %-8s\n'
455
% tuple('idx sha1 base flags offset len'.split()))
456
f.write('-------- ---------------------------------------- ')
457
f.write('-------- -------- -------- --------\n')
459
for i, rec in enumerate(self):
460
f.write("#%-7d %40s " % (i, hexlify(rec[0])))
461
if rec[1] == _NO_RECORD:
464
f.write("#%-7d " % rec[1])
466
f.write("%8x %8d %8d\n" % (rec[2], rec[3], rec[4]))
469
def total_text_size(self):
470
"""Return the sum of sizes of all file texts.
472
This is how much space they would occupy if they were stored without
473
delta and gzip compression.
475
As a side effect this completely validates the Revfile, checking that all
476
texts can be reproduced with the correct SHA-1."""
478
for idx in range(len(self)):
479
t += len(self.get(idx))
483
def check(self, pb=None):
484
"""Extract every version and check its hash."""
486
for i in range(total):
488
pb.update("check revision", i, total)
489
# the get method implicitly checks the SHA-1
501
sys.stderr.write("usage: revfile dump REVFILE\n"
502
" revfile add REVFILE < INPUT\n"
503
" revfile add-delta REVFILE BASE < INPUT\n"
504
" revfile add-series REVFILE BASE FILE...\n"
505
" revfile get REVFILE IDX\n"
506
" revfile find-sha REVFILE HEX\n"
507
" revfile total-text-size REVFILE\n"
508
" revfile last REVFILE\n")
512
return Revfile(filename, 'w')
515
return Revfile(filename, 'r')
518
print rw().add(sys.stdin.read())
519
elif cmd == 'add-delta':
520
print rw().add(sys.stdin.read(), int(argv[3]))
521
elif cmd == 'add-series':
526
rev = r.add(file(fn).read(), rev)
533
sys.stderr.write("usage: revfile get FILE IDX\n")
538
if idx < 0 or idx >= len(r):
539
sys.stderr.write("invalid index %r\n" % idx)
542
sys.stdout.write(r.get(idx))
543
elif cmd == 'find-sha':
545
s = unhexlify(argv[3])
547
sys.stderr.write("usage: revfile find-sha FILE HEX\n")
550
idx = ro().find_sha(s)
551
if idx == _NO_RECORD:
552
sys.stderr.write("no such record\n")
556
elif cmd == 'total-text-size':
557
print ro().total_text_size()
561
import bzrlib.progress
562
pb = bzrlib.progress.ProgressBar()
565
sys.stderr.write("unknown command %r\n" % cmd)
569
if __name__ == '__main__':
571
sys.exit(main(sys.argv) or 0)
added
removed
bzrlib/revfile.py
