1
# Copyright (C) 2005, 2006 Canonical Ltd
1
# Copyright (C) 2005, 2006 by 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."""
20
from cStringIO import StringIO
22
20
# make GzipFile faster:
24
from gzip import FEXTRA, FCOMMENT, FNAME, FHCRC
22
from gzip import U32, LOWU32, FEXTRA, FCOMMENT, FNAME, FHCRC
29
27
# we want a \n preserved, break on \n only splitlines.
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
def bytes_to_gzip(bytes, factory=zlib.compressobj,
51
level=zlib.Z_DEFAULT_COMPRESSION, method=zlib.DEFLATED,
52
width=-zlib.MAX_WBITS, mem=zlib.DEF_MEM_LEVEL,
54
"""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
'\037\213' # self.fileobj.write('\037\213') # magic header
69
'\010' # self.fileobj.write('\010') # compression method
70
# fname = self.filename[:-3]
74
'\x00' # self.fileobj.write(chr(flags))
75
'\0\0\0\0' # write32u(self.fileobj, long(time.time()))
76
'\002' # self.fileobj.write('\002')
77
'\377' # self.fileobj.write('\377')
79
'' # self.fileobj.write(fname + '\000')
81
# using a compressobj avoids a small header and trailer that the compress()
82
# utility function adds.
83
compress = factory(level, method, width, mem, 0)
87
crc = crc32(chunk, crc)
88
total_len += len(chunk)
89
zbytes = compress.compress(chunk)
92
result.append(compress.flush())
93
# size may exceed 2GB, or even 4GB
94
result.append(struct.pack("<LL", LOWU32(crc), LOWU32(total_len)))
95
return ''.join(result)
30
__all__ = ["GzipFile"]
98
33
class GzipFile(gzip.GzipFile):
128
63
self.extrasize += len_data
129
64
self.size += len_data
131
def _write_gzip_header(self):
132
"""A tuned version of gzip._write_gzip_header
134
We have some extra constrains that plain Gzip does not.
135
1) We want to write the whole blob at once. rather than multiple
136
calls to fileobj.write().
137
2) We never have a filename
138
3) We don't care about the time
141
'\037\213' # self.fileobj.write('\037\213') # magic header
142
'\010' # self.fileobj.write('\010') # compression method
143
# fname = self.filename[:-3]
147
'\x00' # self.fileobj.write(chr(flags))
148
'\0\0\0\0' # write32u(self.fileobj, long(time.time()))
149
'\002' # self.fileobj.write('\002')
150
'\377' # self.fileobj.write('\377')
152
'' # self.fileobj.write(fname + '\000')
155
66
def _read(self, size=1024):
156
67
# various optimisations:
157
# reduces lsprof count from 2500 to
68
# reduces lsprof count from 2500 to
158
69
# 8337 calls in 1272, 365 internal
159
70
if self.fileobj is None:
160
71
raise EOFError, "Reached EOF"
184
95
self._add_read_data(self.decompress.flush())
185
if len(self.decompress.unused_data) < 8:
186
raise AssertionError("what does flush do?")
96
assert len(self.decompress.unused_data) >= 8, "what does flush do?"
187
97
self._gzip_tail = self.decompress.unused_data[0:8]
189
99
# tell the driving read() call we have stuffed all the data
209
119
self._gzip_tail = self.decompress.unused_data[0:8]
210
120
elif seek_length < 0:
211
121
# we haven't read enough to check the checksum.
212
if not (-8 < seek_length):
213
raise AssertionError("too great a seek")
122
assert -8 < seek_length, "too great a seek."
214
123
buf = self.fileobj.read(-seek_length)
215
124
self._gzip_tail = self.decompress.unused_data + buf
225
134
"""tuned to reduce function calls and eliminate file seeking:
227
136
reduces lsprof count from 800 to 288
229
138
avoid U32 call by using struct format L
232
# We've read to the end of the file, so we should have 8 bytes of
141
# We've read to the end of the file, so we should have 8 bytes of
233
142
# unused data in the decompressor. If we don't, there is a corrupt file.
234
143
# We use these 8 bytes to calculate the CRC and the recorded file size.
235
144
# We then check the that the computed CRC and size of the
236
145
# uncompressed data matches the stored values. Note that the size
237
146
# stored is the true file size mod 2**32.
238
if not (len(self._gzip_tail) == 8):
239
raise AssertionError("gzip trailer is incorrect length.")
147
assert len(self._gzip_tail) == 8, "gzip trailer is incorrect length."
240
148
crc32, isize = struct.unpack("<LL", self._gzip_tail)
241
149
# note that isize is unsigned - it can exceed 2GB
242
150
if crc32 != U32(self.crc):
290
198
def readline(self, size=-1):
291
199
"""Tuned to remove buffer length calls in _unread and...
293
201
also removes multiple len(c) calls, inlines _unread,
294
202
total savings - lsprof 5800 to 5300
348
256
# 4168 calls in 417.
349
257
# Negative numbers result in reading all the lines
351
# python's gzip routine uses sizehint. This is a more efficient way
352
# than python uses to honor it. But it is even more efficient to
353
# just read the entire thing and use cStringIO to split into lines.
356
# content = self.read(sizehint)
357
# return bzrlib.osutils.split_lines(content)
358
content = StringIO(self.read(-1))
359
return content.readlines()
260
content = self.read(sizehint)
261
return bzrlib.osutils.split_lines(content)
361
263
def _unread(self, buf, len_buf=None):
362
264
"""tuned to remove unneeded len calls.
364
266
because this is such an inner routine in readline, and readline is
365
267
in many inner loops, this has been inlined into readline().
367
269
The len_buf parameter combined with the reduction in len calls dropped
368
the lsprof ms count for this routine on my test data from 800 to 200 -
270
the lsprof ms count for this routine on my test data from 800 to 200 -
371
273
if len_buf is None:
389
291
self.offset += data_len
391
293
def writelines(self, lines):
392
# profiling indicated a significant overhead
294
# profiling indicated a significant overhead
393
295
# calling write for each line.
394
296
# this batch call is a lot faster :).
395
297
# (4 seconds to 1 seconds for the sample upgrades I was testing).