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."""
22
22
# make GzipFile faster:
24
from gzip import U32, LOWU32, FEXTRA, FCOMMENT, FNAME, FHCRC
24
from gzip import FEXTRA, FCOMMENT, FNAME, FHCRC
29
29
# we want a \n preserved, break on \n only splitlines.
32
__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)
35
98
class GzipFile(gzip.GzipFile):
69
132
"""A tuned version of gzip._write_gzip_header
71
134
We have some extra constrains that plain Gzip does not.
72
1) We want to write the whole blob at once. rather than multiple
135
1) We want to write the whole blob at once. rather than multiple
73
136
calls to fileobj.write().
74
137
2) We never have a filename
75
138
3) We don't care about the time
92
155
def _read(self, size=1024):
93
156
# various optimisations:
94
# reduces lsprof count from 2500 to
157
# reduces lsprof count from 2500 to
95
158
# 8337 calls in 1272, 365 internal
96
159
if self.fileobj is None:
97
160
raise EOFError, "Reached EOF"
121
184
self._add_read_data(self.decompress.flush())
122
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?")
123
187
self._gzip_tail = self.decompress.unused_data[0:8]
125
189
# tell the driving read() call we have stuffed all the data
145
209
self._gzip_tail = self.decompress.unused_data[0:8]
146
210
elif seek_length < 0:
147
211
# we haven't read enough to check the checksum.
148
assert -8 < seek_length, "too great a seek."
212
if not (-8 < seek_length):
213
raise AssertionError("too great a seek")
149
214
buf = self.fileobj.read(-seek_length)
150
215
self._gzip_tail = self.decompress.unused_data + buf
160
225
"""tuned to reduce function calls and eliminate file seeking:
162
227
reduces lsprof count from 800 to 288
164
229
avoid U32 call by using struct format L
167
# 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
168
233
# unused data in the decompressor. If we don't, there is a corrupt file.
169
234
# We use these 8 bytes to calculate the CRC and the recorded file size.
170
235
# We then check the that the computed CRC and size of the
171
236
# uncompressed data matches the stored values. Note that the size
172
237
# stored is the true file size mod 2**32.
173
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.")
174
240
crc32, isize = struct.unpack("<LL", self._gzip_tail)
175
241
# note that isize is unsigned - it can exceed 2GB
176
242
if crc32 != U32(self.crc):
224
290
def readline(self, size=-1):
225
291
"""Tuned to remove buffer length calls in _unread and...
227
293
also removes multiple len(c) calls, inlines _unread,
228
294
total savings - lsprof 5800 to 5300
295
361
def _unread(self, buf, len_buf=None):
296
362
"""tuned to remove unneeded len calls.
298
364
because this is such an inner routine in readline, and readline is
299
365
in many inner loops, this has been inlined into readline().
301
367
The len_buf parameter combined with the reduction in len calls dropped
302
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 -
305
371
if len_buf is None:
323
389
self.offset += data_len
325
391
def writelines(self, lines):
326
# profiling indicated a significant overhead
392
# profiling indicated a significant overhead
327
393
# calling write for each line.
328
394
# this batch call is a lot faster :).
329
395
# (4 seconds to 1 seconds for the sample upgrades I was testing).