1
# Copyright (C) 2006-2011 Canonical Ltd
1
# Copyright (C) 2005, 2006 Canonical Ltd
2
2
# Written by Robert Collins <robert.collins@canonical.com>
4
4
# This program is free software; you can redistribute it and/or modify
14
14
# You should have received a copy of the GNU General Public License
15
15
# along with this program; if not, write to the Free Software
16
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
16
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18
18
"""Bzrlib specific gzip tunings. We plan to feed these to the upstream gzip."""
22
22
# make GzipFile faster:
24
from gzip import FEXTRA, FCOMMENT, FNAME, FHCRC
24
from gzip import U32, LOWU32, FEXTRA, FCOMMENT, FNAME, FHCRC
29
29
# we want a \n preserved, break on \n only splitlines.
30
from bzrlib import symbol_versioning
32
32
__all__ = ["GzipFile", "bytes_to_gzip"]
36
"""Return i as an unsigned integer, assuming it fits in 32 bits.
38
If it's >= 2GB when viewed as a 32-bit unsigned int, return a long.
46
"""Return the low-order 32 bits of an int, as a non-negative int."""
47
return i & 0xFFFFFFFFL
50
35
def bytes_to_gzip(bytes, factory=zlib.compressobj,
51
36
level=zlib.Z_DEFAULT_COMPRESSION, method=zlib.DEFLATED,
52
37
width=-zlib.MAX_WBITS, mem=zlib.DEF_MEM_LEVEL,
54
39
"""Create a gzip file containing bytes and return its content."""
55
return chunks_to_gzip([bytes])
58
def chunks_to_gzip(chunks, factory=zlib.compressobj,
59
level=zlib.Z_DEFAULT_COMPRESSION, method=zlib.DEFLATED,
60
width=-zlib.MAX_WBITS, mem=zlib.DEF_MEM_LEVEL,
62
"""Create a gzip file containing chunks and return its content.
64
:param chunks: An iterable of strings. Each string can have arbitrary
68
41
'\037\213' # self.fileobj.write('\037\213') # magic header
69
42
'\010' # self.fileobj.write('\010') # compression method
81
54
# using a compressobj avoids a small header and trailer that the compress()
82
55
# utility function adds.
83
56
compress = factory(level, method, width, mem, 0)
87
crc = crc32(chunk, crc)
88
total_len += len(chunk)
89
zbytes = compress.compress(chunk)
57
result.append(compress.compress(bytes))
92
58
result.append(compress.flush())
59
result.append(struct.pack("<L", LOWU32(crc32(bytes))))
93
60
# size may exceed 2GB, or even 4GB
94
result.append(struct.pack("<LL", LOWU32(crc), LOWU32(total_len)))
61
result.append(struct.pack("<L", LOWU32(len(bytes))))
95
62
return ''.join(result)
118
85
Yes, its only 1.6 seconds, but they add up.
121
def __init__(self, *args, **kwargs):
122
symbol_versioning.warn(
123
symbol_versioning.deprecated_in((2, 3, 0))
124
% 'bzrlib.tuned_gzip.GzipFile',
125
DeprecationWarning, stacklevel=2)
126
gzip.GzipFile.__init__(self, *args, **kwargs)
128
88
def _add_read_data(self, data):
129
89
# 4169 calls in 183
130
90
# temp var for len(data) and switch to +='s.
139
99
"""A tuned version of gzip._write_gzip_header
141
101
We have some extra constrains that plain Gzip does not.
142
1) We want to write the whole blob at once. rather than multiple
102
1) We want to write the whole blob at once. rather than multiple
143
103
calls to fileobj.write().
144
104
2) We never have a filename
145
105
3) We don't care about the time
162
122
def _read(self, size=1024):
163
123
# various optimisations:
164
# reduces lsprof count from 2500 to
124
# reduces lsprof count from 2500 to
165
125
# 8337 calls in 1272, 365 internal
166
126
if self.fileobj is None:
167
127
raise EOFError, "Reached EOF"
191
151
self._add_read_data(self.decompress.flush())
192
if len(self.decompress.unused_data) < 8:
193
raise AssertionError("what does flush do?")
152
assert len(self.decompress.unused_data) >= 8, "what does flush do?"
194
153
self._gzip_tail = self.decompress.unused_data[0:8]
196
155
# tell the driving read() call we have stuffed all the data
216
175
self._gzip_tail = self.decompress.unused_data[0:8]
217
176
elif seek_length < 0:
218
177
# we haven't read enough to check the checksum.
219
if not (-8 < seek_length):
220
raise AssertionError("too great a seek")
178
assert -8 < seek_length, "too great a seek."
221
179
buf = self.fileobj.read(-seek_length)
222
180
self._gzip_tail = self.decompress.unused_data + buf
232
190
"""tuned to reduce function calls and eliminate file seeking:
234
192
reduces lsprof count from 800 to 288
236
194
avoid U32 call by using struct format L
239
# We've read to the end of the file, so we should have 8 bytes of
197
# We've read to the end of the file, so we should have 8 bytes of
240
198
# unused data in the decompressor. If we don't, there is a corrupt file.
241
199
# We use these 8 bytes to calculate the CRC and the recorded file size.
242
200
# We then check the that the computed CRC and size of the
243
201
# uncompressed data matches the stored values. Note that the size
244
202
# stored is the true file size mod 2**32.
245
if not (len(self._gzip_tail) == 8):
246
raise AssertionError("gzip trailer is incorrect length.")
203
assert len(self._gzip_tail) == 8, "gzip trailer is incorrect length."
247
204
crc32, isize = struct.unpack("<LL", self._gzip_tail)
248
205
# note that isize is unsigned - it can exceed 2GB
249
206
if crc32 != U32(self.crc):
254
211
def _read_gzip_header(self, bytes=None):
255
212
"""Supply bytes if the minimum header size is already read.
257
214
:param bytes: 10 bytes of header data.
259
216
"""starting cost: 300 in 3998
297
254
def readline(self, size=-1):
298
255
"""Tuned to remove buffer length calls in _unread and...
300
257
also removes multiple len(c) calls, inlines _unread,
301
258
total savings - lsprof 5800 to 5300
355
312
# 4168 calls in 417.
356
313
# Negative numbers result in reading all the lines
358
315
# python's gzip routine uses sizehint. This is a more efficient way
359
316
# than python uses to honor it. But it is even more efficient to
360
317
# just read the entire thing and use cStringIO to split into lines.
368
325
def _unread(self, buf, len_buf=None):
369
326
"""tuned to remove unneeded len calls.
371
328
because this is such an inner routine in readline, and readline is
372
329
in many inner loops, this has been inlined into readline().
374
331
The len_buf parameter combined with the reduction in len calls dropped
375
the lsprof ms count for this routine on my test data from 800 to 200 -
332
the lsprof ms count for this routine on my test data from 800 to 200 -
378
335
if len_buf is None:
396
353
self.offset += data_len
398
355
def writelines(self, lines):
399
# profiling indicated a significant overhead
356
# profiling indicated a significant overhead
400
357
# calling write for each line.
401
358
# this batch call is a lot faster :).
402
359
# (4 seconds to 1 seconds for the sample upgrades I was testing).
403
360
self.write(''.join(lines))
405
if sys.version_info > (2, 7):
406
# As of Python 2.7 the crc32 must be positive when close is called
408
if self.fileobj is None:
410
if self.mode == gzip.WRITE:
411
self.crc &= 0xFFFFFFFFL
412
gzip.GzipFile.close(self)