1
# Copyright (C) 2005, 2006 by 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., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
16
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
18
18
"""Bzrlib specific gzip tunings. We plan to feed these to the upstream gzip."""
20
from cStringIO import StringIO
20
22
# make GzipFile faster:
22
from gzip import U32, LOWU32, FEXTRA, FCOMMENT, FNAME, FHCRC
24
from gzip import FEXTRA, FCOMMENT, FNAME, FHCRC
27
29
# we want a \n preserved, break on \n only splitlines.
30
__all__ = ["GzipFile"]
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)
33
98
class GzipFile(gzip.GzipFile):
63
128
self.extrasize += len_data
64
129
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')
66
155
def _read(self, size=1024):
67
156
# various optimisations:
68
# reduces lsprof count from 2500 to
157
# reduces lsprof count from 2500 to
69
158
# 8337 calls in 1272, 365 internal
70
159
if self.fileobj is None:
71
160
raise EOFError, "Reached EOF"
95
184
self._add_read_data(self.decompress.flush())
96
assert len(self.decompress.unused_data) >= 8, "what does flush do?"
185
if len(self.decompress.unused_data) < 8:
186
raise AssertionError("what does flush do?")
97
187
self._gzip_tail = self.decompress.unused_data[0:8]
99
189
# tell the driving read() call we have stuffed all the data
119
209
self._gzip_tail = self.decompress.unused_data[0:8]
120
210
elif seek_length < 0:
121
211
# we haven't read enough to check the checksum.
122
assert -8 < seek_length, "too great a seek."
212
if not (-8 < seek_length):
213
raise AssertionError("too great a seek")
123
214
buf = self.fileobj.read(-seek_length)
124
215
self._gzip_tail = self.decompress.unused_data + buf
134
225
"""tuned to reduce function calls and eliminate file seeking:
136
227
reduces lsprof count from 800 to 288
138
229
avoid U32 call by using struct format L
141
# We've read to the end of the file, so we should have 8 bytes of
232
# We've read to the end of the file, so we should have 8 bytes of
142
233
# unused data in the decompressor. If we don't, there is a corrupt file.
143
234
# We use these 8 bytes to calculate the CRC and the recorded file size.
144
235
# We then check the that the computed CRC and size of the
145
236
# uncompressed data matches the stored values. Note that the size
146
237
# stored is the true file size mod 2**32.
147
assert len(self._gzip_tail) == 8, "gzip trailer is incorrect length."
238
if not (len(self._gzip_tail) == 8):
239
raise AssertionError("gzip trailer is incorrect length.")
148
240
crc32, isize = struct.unpack("<LL", self._gzip_tail)
149
241
# note that isize is unsigned - it can exceed 2GB
150
242
if crc32 != U32(self.crc):
198
290
def readline(self, size=-1):
199
291
"""Tuned to remove buffer length calls in _unread and...
201
293
also removes multiple len(c) calls, inlines _unread,
202
294
total savings - lsprof 5800 to 5300
256
348
# 4168 calls in 417.
257
349
# Negative numbers result in reading all the lines
260
content = self.read(sizehint)
261
return bzrlib.osutils.split_lines(content)
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()
263
361
def _unread(self, buf, len_buf=None):
264
362
"""tuned to remove unneeded len calls.
266
364
because this is such an inner routine in readline, and readline is
267
365
in many inner loops, this has been inlined into readline().
269
367
The len_buf parameter combined with the reduction in len calls dropped
270
the lsprof ms count for this routine on my test data from 800 to 200 -
368
the lsprof ms count for this routine on my test data from 800 to 200 -
273
371
if len_buf is None:
291
389
self.offset += data_len
293
391
def writelines(self, lines):
294
# profiling indicated a significant overhead
392
# profiling indicated a significant overhead
295
393
# calling write for each line.
296
394
# this batch call is a lot faster :).
297
395
# (4 seconds to 1 seconds for the sample upgrades I was testing).