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.
55
The iter method here will generally read through the whole index file
56
in one go. With readahead in the kernel and python/libc (typically
57
128kB) this means that there should be no seeks and often only one
58
read() call to get everything into memory.
62
# TODO: Something like pread() would make this slightly simpler and
63
# perhaps more efficient.
65
# TODO: Could also try to mmap things... Might be faster for the
66
# index in particular?
68
# TODO: Some kind of faster lookup of SHAs? The bad thing is that probably means
69
# rewriting existing records, which is not so nice.
71
# TODO: Something to check that regions identified in the index file
72
# completely butt up and do not overlap. Strictly it's not a problem
73
# if there are gaps and that can happen if we're interrupted while
74
# writing to the datafile. Overlapping would be very bad though.
78
import sys, zlib, struct, mdiff, stat, os, sha
79
from binascii import hexlify, unhexlify
85
_HEADER = "bzr revfile v1\n"
86
_HEADER = _HEADER + ('\xff' * (_RECORDSIZE - len(_HEADER)))
87
_NO_RECORD = 0xFFFFFFFFL
89
# fields in the index record
98
# maximum number of patches in a row before recording a whole text.
102
class RevfileError(Exception):
105
class LimitHitException(Exception):
108
class Revfile(object):
109
def __init__(self, basename, mode):
110
# TODO: Lock file while open
112
# TODO: advise of random access
114
self.basename = basename
116
if mode not in ['r', 'w']:
117
raise RevfileError("invalid open mode %r" % mode)
120
idxname = basename + '.irev'
121
dataname = basename + '.drev'
123
idx_exists = os.path.exists(idxname)
124
data_exists = os.path.exists(dataname)
126
if idx_exists != data_exists:
127
raise RevfileError("half-assed revfile")
131
raise RevfileError("Revfile %r does not exist" % basename)
133
self.idxfile = open(idxname, 'w+b')
134
self.datafile = open(dataname, 'w+b')
136
print 'init empty file'
137
self.idxfile.write(_HEADER)
145
self.idxfile = open(idxname, diskmode)
146
self.datafile = open(dataname, diskmode)
148
h = self.idxfile.read(_RECORDSIZE)
150
raise RevfileError("bad header %r in index of %r"
151
% (h, self.basename))
154
def _check_index(self, idx):
155
if idx < 0 or idx > len(self):
156
raise RevfileError("invalid index %r" % idx)
158
def _check_write(self):
160
raise RevfileError("%r is open readonly" % self.basename)
163
def find_sha(self, s):
164
assert isinstance(s, str)
167
for idx, idxrec in enumerate(self):
168
if idxrec[I_SHA] == s:
175
def _add_compressed(self, text_sha, data, base, compress):
176
# well, maybe compress
181
# don't do compression if it's too small; it's unlikely to win
182
# enough to be worthwhile
183
compr_data = zlib.compress(data)
184
compr_len = len(compr_data)
185
if compr_len < data_len:
188
##print '- compressed %d -> %d, %.1f%%' \
189
## % (data_len, compr_len, float(compr_len)/float(data_len) * 100.0)
190
return self._add_raw(text_sha, data, base, flags)
194
def _add_raw(self, text_sha, data, base, flags):
195
"""Add pre-processed data, can be either full text or delta.
197
This does the compression if that makes sense."""
199
self.datafile.seek(0, 2) # to end
200
self.idxfile.seek(0, 2)
201
assert self.idxfile.tell() == _RECORDSIZE * (idx + 1)
202
data_offset = self.datafile.tell()
204
assert isinstance(data, str) # not unicode or anything weird
206
self.datafile.write(data)
207
self.datafile.flush()
209
assert isinstance(text_sha, str)
211
entry += struct.pack(">IIII12x", base, flags, data_offset, len(data))
212
assert len(entry) == _RECORDSIZE
214
self.idxfile.write(entry)
221
def _add_full_text(self, text, text_sha, compress):
222
"""Add a full text to the file.
224
This is not compressed against any reference version.
226
Returns the index for that text."""
227
return self._add_compressed(text_sha, text, _NO_RECORD, compress)
230
def _add_delta(self, text, text_sha, base, compress):
231
"""Add a text stored relative to a previous text."""
232
self._check_index(base)
235
base_text = self.get(base, recursion_limit=CHAIN_LIMIT)
236
except LimitHitException:
237
return self._add_full_text(text, text_sha, compress)
239
data = mdiff.bdiff(base_text, text)
241
# If the delta is larger than the text, we might as well just
242
# store the text. (OK, the delta might be more compressible,
243
# but the overhead of applying it probably still makes it
244
# bad, and I don't want to compress both of them to find out.)
245
if len(data) >= len(text):
246
return self._add_full_text(text, text_sha, compress)
248
return self._add_compressed(text_sha, data, base, compress)
251
def add(self, text, base=_NO_RECORD, compress=True):
252
"""Add a new text to the revfile.
254
If the text is already present them its existing id is
255
returned and the file is not changed.
257
If compress is true then gzip compression will be used if it
260
If a base index is specified, that text *may* be used for
261
delta compression of the new text. Delta compression will
262
only be used if it would be a size win and if the existing
263
base is not at too long of a delta chain already.
267
text_sha = sha.new(text).digest()
269
idx = self.find_sha(text_sha)
270
if idx != _NO_RECORD:
271
# TODO: Optional paranoid mode where we read out that record and make sure
272
# it's the same, in case someone ever breaks SHA-1.
273
return idx # already present
275
if base == _NO_RECORD:
276
return self._add_full_text(text, text_sha, compress)
278
return self._add_delta(text, text_sha, base, compress)
282
def get(self, idx, recursion_limit=None):
283
"""Retrieve text of a previous revision.
285
If recursion_limit is an integer then walk back at most that
286
many revisions and then raise LimitHitException, indicating
287
that we ought to record a new file text instead of another
288
delta. Don't use this when trying to get out an existing
292
base = idxrec[I_BASE]
293
if base == _NO_RECORD:
294
text = self._get_full_text(idx, idxrec)
296
text = self._get_patched(idx, idxrec, recursion_limit)
298
if sha.new(text).digest() != idxrec[I_SHA]:
299
raise RevfileError("corrupt SHA-1 digest on record %d"
306
def _get_raw(self, idx, idxrec):
307
flags = idxrec[I_FLAGS]
309
raise RevfileError("unsupported index flags %#x on index %d"
316
self.datafile.seek(idxrec[I_OFFSET])
318
data = self.datafile.read(l)
320
raise RevfileError("short read %d of %d "
321
"getting text for record %d in %r"
322
% (len(data), l, idx, self.basename))
325
data = zlib.decompress(data)
330
def _get_full_text(self, idx, idxrec):
331
assert idxrec[I_BASE] == _NO_RECORD
333
text = self._get_raw(idx, idxrec)
338
def _get_patched(self, idx, idxrec, recursion_limit):
339
base = idxrec[I_BASE]
341
assert base < idx # no loops!
343
if recursion_limit == None:
346
sub_limit = recursion_limit - 1
348
raise LimitHitException()
350
base_text = self.get(base, sub_limit)
351
patch = self._get_raw(idx, idxrec)
353
text = mdiff.bpatch(base_text, patch)
360
"""Return number of revisions."""
361
l = os.fstat(self.idxfile.fileno())[stat.ST_SIZE]
363
raise RevfileError("bad length %d on index of %r" % (l, self.basename))
365
raise RevfileError("no header present in index of %r" % (self.basename))
366
return int(l / _RECORDSIZE) - 1
369
def __getitem__(self, idx):
370
"""Index by sequence id returns the index field"""
371
## TODO: Can avoid seek if we just moved there...
372
self._seek_index(idx)
373
idxrec = self._read_next_index()
380
def _seek_index(self, idx):
382
raise RevfileError("invalid index %r" % idx)
383
self.idxfile.seek((idx + 1) * _RECORDSIZE)
388
"""Read back all index records.
390
Do not seek the index file while this is underway!"""
391
sys.stderr.write(" ** iter called ** \n")
394
idxrec = self._read_next_index()
400
def _read_next_index(self):
401
rec = self.idxfile.read(_RECORDSIZE)
404
elif len(rec) != _RECORDSIZE:
405
raise RevfileError("short read of %d bytes getting index %d from %r"
406
% (len(rec), idx, self.basename))
408
return struct.unpack(">20sIIII12x", rec)
411
def dump(self, f=sys.stdout):
412
f.write('%-8s %-40s %-8s %-8s %-8s %-8s\n'
413
% tuple('idx sha1 base flags offset len'.split()))
414
f.write('-------- ---------------------------------------- ')
415
f.write('-------- -------- -------- --------\n')
417
for i, rec in enumerate(self):
418
f.write("#%-7d %40s " % (i, hexlify(rec[0])))
419
if rec[1] == _NO_RECORD:
422
f.write("#%-7d " % rec[1])
424
f.write("%8x %8d %8d\n" % (rec[2], rec[3], rec[4]))
427
def total_text_size(self):
428
"""Return the sum of sizes of all file texts.
430
This is how much space they would occupy if they were stored without
431
delta and gzip compression.
433
As a side effect this completely validates the Revfile, checking that all
434
texts can be reproduced with the correct SHA-1."""
436
for idx in range(len(self)):
437
t += len(self.get(idx))
446
sys.stderr.write("usage: revfile dump\n"
448
" revfile add-delta BASE\n"
450
" revfile find-sha HEX\n"
451
" revfile total-text-size\n"
456
return Revfile('testrev', 'w')
459
return Revfile('testrev', 'r')
462
print rw().add(sys.stdin.read())
463
elif cmd == 'add-delta':
464
print rw().add(sys.stdin.read(), int(argv[2]))
471
sys.stderr.write("usage: revfile get IDX\n")
474
if idx < 0 or idx >= len(r):
475
sys.stderr.write("invalid index %r\n" % idx)
478
sys.stdout.write(ro().get(idx))
479
elif cmd == 'find-sha':
481
s = unhexlify(argv[2])
483
sys.stderr.write("usage: revfile find-sha HEX\n")
486
idx = ro().find_sha(s)
487
if idx == _NO_RECORD:
488
sys.stderr.write("no such record\n")
492
elif cmd == 'total-text-size':
493
print ro().total_text_size()
497
sys.stderr.write("unknown command %r\n" % cmd)
501
if __name__ == '__main__':
503
sys.exit(main(sys.argv) or 0)
added
removed
bzrlib/revfile.py
