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
print 'init empty file'
153
self.idxfile.write(_HEADER)
161
self.idxfile = open(idxname, diskmode)
162
self.datafile = open(dataname, diskmode)
164
h = self.idxfile.read(_RECORDSIZE)
166
raise RevfileError("bad header %r in index of %r"
167
% (h, self.basename))
170
def _check_index(self, idx):
171
if idx < 0 or idx > len(self):
172
raise RevfileError("invalid index %r" % idx)
174
def _check_write(self):
176
raise RevfileError("%r is open readonly" % self.basename)
179
def find_sha(self, s):
180
assert isinstance(s, str)
183
for idx, idxrec in enumerate(self):
184
if idxrec[I_SHA] == s:
191
def _add_compressed(self, text_sha, data, base, compress):
192
# well, maybe compress
197
# don't do compression if it's too small; it's unlikely to win
198
# enough to be worthwhile
199
compr_data = zlib.compress(data)
200
compr_len = len(compr_data)
201
if compr_len < data_len:
204
##print '- compressed %d -> %d, %.1f%%' \
205
## % (data_len, compr_len, float(compr_len)/float(data_len) * 100.0)
206
return self._add_raw(text_sha, data, base, flags)
210
def _add_raw(self, text_sha, data, base, flags):
211
"""Add pre-processed data, can be either full text or delta.
213
This does the compression if that makes sense."""
215
self.datafile.seek(0, 2) # to end
216
self.idxfile.seek(0, 2)
217
assert self.idxfile.tell() == _RECORDSIZE * (idx + 1)
218
data_offset = self.datafile.tell()
220
assert isinstance(data, str) # not unicode or anything weird
222
self.datafile.write(data)
223
self.datafile.flush()
225
assert isinstance(text_sha, str)
227
entry += struct.pack(">IIII12x", base, flags, data_offset, len(data))
228
assert len(entry) == _RECORDSIZE
230
self.idxfile.write(entry)
237
def _add_full_text(self, text, text_sha, compress):
238
"""Add a full text to the file.
240
This is not compressed against any reference version.
242
Returns the index for that text."""
243
return self._add_compressed(text_sha, text, _NO_RECORD, compress)
247
def _choose_base(self, seed, base):
249
if base == _NO_RECORD:
252
if idxrec[I_BASE] == _NO_RECORD:
255
base = idxrec[I_BASE]
258
return base # relative to this full text
262
def _add_delta(self, text, text_sha, base, compress):
263
"""Add a text stored relative to a previous text."""
264
self._check_index(base)
267
base_text = self.get(base, CHAIN_LIMIT)
268
except LimitHitException:
269
return self._add_full_text(text, text_sha, compress)
271
data = mdiff.bdiff(base_text, text)
273
# If the delta is larger than the text, we might as well just
274
# store the text. (OK, the delta might be more compressible,
275
# but the overhead of applying it probably still makes it
276
# bad, and I don't want to compress both of them to find out.)
277
if len(data) >= len(text):
278
return self._add_full_text(text, text_sha, compress)
280
return self._add_compressed(text_sha, data, base, compress)
283
def add(self, text, base=_NO_RECORD, compress=True):
284
"""Add a new text to the revfile.
286
If the text is already present them its existing id is
287
returned and the file is not changed.
289
If compress is true then gzip compression will be used if it
292
If a base index is specified, that text *may* be used for
293
delta compression of the new text. Delta compression will
294
only be used if it would be a size win and if the existing
295
base is not at too long of a delta chain already.
299
text_sha = sha.new(text).digest()
301
idx = self.find_sha(text_sha)
302
if idx != _NO_RECORD:
303
# TODO: Optional paranoid mode where we read out that record and make sure
304
# it's the same, in case someone ever breaks SHA-1.
305
return idx # already present
307
# base = self._choose_base(ord(text_sha[0]), base)
309
if base == _NO_RECORD:
310
return self._add_full_text(text, text_sha, compress)
312
return self._add_delta(text, text_sha, base, compress)
316
def get(self, idx, recursion_limit=None):
317
"""Retrieve text of a previous revision.
319
If recursion_limit is an integer then walk back at most that
320
many revisions and then raise LimitHitException, indicating
321
that we ought to record a new file text instead of another
322
delta. Don't use this when trying to get out an existing
326
base = idxrec[I_BASE]
327
if base == _NO_RECORD:
328
text = self._get_full_text(idx, idxrec)
330
text = self._get_patched(idx, idxrec, recursion_limit)
332
if sha.new(text).digest() != idxrec[I_SHA]:
333
raise RevfileError("corrupt SHA-1 digest on record %d"
340
def _get_raw(self, idx, idxrec):
341
flags = idxrec[I_FLAGS]
343
raise RevfileError("unsupported index flags %#x on index %d"
350
self.datafile.seek(idxrec[I_OFFSET])
352
data = self.datafile.read(l)
354
raise RevfileError("short read %d of %d "
355
"getting text for record %d in %r"
356
% (len(data), l, idx, self.basename))
359
data = zlib.decompress(data)
364
def _get_full_text(self, idx, idxrec):
365
assert idxrec[I_BASE] == _NO_RECORD
367
text = self._get_raw(idx, idxrec)
372
def _get_patched(self, idx, idxrec, recursion_limit):
373
base = idxrec[I_BASE]
375
assert base < idx # no loops!
377
if recursion_limit == None:
380
sub_limit = recursion_limit - 1
382
raise LimitHitException()
384
base_text = self.get(base, sub_limit)
385
patch = self._get_raw(idx, idxrec)
387
text = mdiff.bpatch(base_text, patch)
394
"""Return number of revisions."""
395
l = os.fstat(self.idxfile.fileno())[stat.ST_SIZE]
397
raise RevfileError("bad length %d on index of %r" % (l, self.basename))
399
raise RevfileError("no header present in index of %r" % (self.basename))
400
return int(l / _RECORDSIZE) - 1
403
def __getitem__(self, idx):
404
"""Index by sequence id returns the index field"""
405
## TODO: Can avoid seek if we just moved there...
406
self._seek_index(idx)
407
idxrec = self._read_next_index()
409
raise IndexError("no index %d" % idx)
414
def _seek_index(self, idx):
416
raise RevfileError("invalid index %r" % idx)
417
self.idxfile.seek((idx + 1) * _RECORDSIZE)
422
"""Read back all index records.
424
Do not seek the index file while this is underway!"""
425
## sys.stderr.write(" ** iter called ** \n")
428
idxrec = self._read_next_index()
434
def _read_next_index(self):
435
rec = self.idxfile.read(_RECORDSIZE)
438
elif len(rec) != _RECORDSIZE:
439
raise RevfileError("short read of %d bytes getting index %d from %r"
440
% (len(rec), idx, self.basename))
442
return struct.unpack(">20sIIII12x", rec)
445
def dump(self, f=sys.stdout):
446
f.write('%-8s %-40s %-8s %-8s %-8s %-8s\n'
447
% tuple('idx sha1 base flags offset len'.split()))
448
f.write('-------- ---------------------------------------- ')
449
f.write('-------- -------- -------- --------\n')
451
for i, rec in enumerate(self):
452
f.write("#%-7d %40s " % (i, hexlify(rec[0])))
453
if rec[1] == _NO_RECORD:
456
f.write("#%-7d " % rec[1])
458
f.write("%8x %8d %8d\n" % (rec[2], rec[3], rec[4]))
461
def total_text_size(self):
462
"""Return the sum of sizes of all file texts.
464
This is how much space they would occupy if they were stored without
465
delta and gzip compression.
467
As a side effect this completely validates the Revfile, checking that all
468
texts can be reproduced with the correct SHA-1."""
470
for idx in range(len(self)):
471
t += len(self.get(idx))
481
sys.stderr.write("usage: revfile dump REVFILE\n"
482
" revfile add REVFILE < INPUT\n"
483
" revfile add-delta REVFILE BASE < INPUT\n"
484
" revfile add-series REVFILE BASE FILE...\n"
485
" revfile get REVFILE IDX\n"
486
" revfile find-sha REVFILE HEX\n"
487
" revfile total-text-size REVFILE\n"
488
" revfile last REVFILE\n")
492
return Revfile(filename, 'w')
495
return Revfile(filename, 'r')
498
print rw().add(sys.stdin.read())
499
elif cmd == 'add-delta':
500
print rw().add(sys.stdin.read(), int(argv[3]))
501
elif cmd == 'add-series':
506
rev = r.add(file(fn).read(), rev)
513
sys.stderr.write("usage: revfile get FILE IDX\n")
518
if idx < 0 or idx >= len(r):
519
sys.stderr.write("invalid index %r\n" % idx)
522
sys.stdout.write(r.get(idx))
523
elif cmd == 'find-sha':
525
s = unhexlify(argv[3])
527
sys.stderr.write("usage: revfile find-sha FILE HEX\n")
530
idx = ro().find_sha(s)
531
if idx == _NO_RECORD:
532
sys.stderr.write("no such record\n")
536
elif cmd == 'total-text-size':
537
print ro().total_text_size()
541
sys.stderr.write("unknown command %r\n" % cmd)
545
if __name__ == '__main__':
547
sys.exit(main(sys.argv) or 0)
added
removed
bzrlib/revfile.py
