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.
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)
32
__all__ = ["GzipFile"]
98
35
class GzipFile(gzip.GzipFile):
132
69
"""A tuned version of gzip._write_gzip_header
134
71
We have some extra constrains that plain Gzip does not.
135
1) We want to write the whole blob at once. rather than multiple
72
1) We want to write the whole blob at once. rather than multiple
136
73
calls to fileobj.write().
137
74
2) We never have a filename
138
75
3) We don't care about the time
155
92
def _read(self, size=1024):
156
93
# various optimisations:
157
# reduces lsprof count from 2500 to
94
# reduces lsprof count from 2500 to
158
95
# 8337 calls in 1272, 365 internal
159
96
if self.fileobj is None:
160
97
raise EOFError, "Reached EOF"
184
121
self._add_read_data(self.decompress.flush())
185
if len(self.decompress.unused_data) < 8:
186
raise AssertionError("what does flush do?")
122
assert len(self.decompress.unused_data) >= 8, "what does flush do?"
187
123
self._gzip_tail = self.decompress.unused_data[0:8]
189
125
# tell the driving read() call we have stuffed all the data
209
145
self._gzip_tail = self.decompress.unused_data[0:8]
210
146
elif seek_length < 0:
211
147
# we haven't read enough to check the checksum.
212
if not (-8 < seek_length):
213
raise AssertionError("too great a seek")
148
assert -8 < seek_length, "too great a seek."
214
149
buf = self.fileobj.read(-seek_length)
215
150
self._gzip_tail = self.decompress.unused_data + buf
225
160
"""tuned to reduce function calls and eliminate file seeking:
227
162
reduces lsprof count from 800 to 288
229
164
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
167
# We've read to the end of the file, so we should have 8 bytes of
233
168
# unused data in the decompressor. If we don't, there is a corrupt file.
234
169
# We use these 8 bytes to calculate the CRC and the recorded file size.
235
170
# We then check the that the computed CRC and size of the
236
171
# uncompressed data matches the stored values. Note that the size
237
172
# 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.")
173
assert len(self._gzip_tail) == 8, "gzip trailer is incorrect length."
240
174
crc32, isize = struct.unpack("<LL", self._gzip_tail)
241
175
# note that isize is unsigned - it can exceed 2GB
242
176
if crc32 != U32(self.crc):
290
224
def readline(self, size=-1):
291
225
"""Tuned to remove buffer length calls in _unread and...
293
227
also removes multiple len(c) calls, inlines _unread,
294
228
total savings - lsprof 5800 to 5300
361
295
def _unread(self, buf, len_buf=None):
362
296
"""tuned to remove unneeded len calls.
364
298
because this is such an inner routine in readline, and readline is
365
299
in many inner loops, this has been inlined into readline().
367
301
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 -
302
the lsprof ms count for this routine on my test data from 800 to 200 -
371
305
if len_buf is None:
389
323
self.offset += data_len
391
325
def writelines(self, lines):
392
# profiling indicated a significant overhead
326
# profiling indicated a significant overhead
393
327
# calling write for each line.
394
328
# this batch call is a lot faster :).
395
329
# (4 seconds to 1 seconds for the sample upgrades I was testing).