32
32
import os, stat, time
35
from bzrlib.osutils import sha_file
35
from bzrlib.osutils import sha_file, pathjoin, safe_unicode
36
36
from bzrlib.trace import mutter, warning
37
37
from bzrlib.atomicfile import AtomicFile
38
from bzrlib.errors import BzrError
42
def _fingerprint(abspath):
44
fs = os.lstat(abspath)
46
# might be missing, etc
49
if stat.S_ISDIR(fs.st_mode):
52
# we discard any high precision because it's not reliable; perhaps we
53
# could do better on some systems?
54
return (fs.st_size, long(fs.st_mtime),
55
long(fs.st_ctime), fs.st_ino, fs.st_dev, fs.st_mode)
58
47
class HashCache(object):
116
104
self.needs_write = True
121
108
"""Scan all files and remove entries where the cache entry is obsolete.
123
110
Obsolete entries are those where the file has been modified or deleted
124
111
since the entry was inserted.
113
# FIXME optimisation opportunity, on linux [and check other oses]:
114
# rather than iteritems order, stat in inode order.
126
115
prep = [(ce[1][3], path, ce) for (path, ce) in self._cache.iteritems()]
129
118
for inum, path, cache_entry in prep:
130
abspath = os.sep.join([self.basedir, path])
131
fp = _fingerprint(abspath)
119
abspath = pathjoin(self.root, path)
120
fp = self._fingerprint(abspath)
132
121
self.stat_count += 1
134
123
cache_fp = cache_entry[1]
139
128
self.needs_write = True
140
129
del self._cache[path]
143
131
def get_sha1(self, path):
144
132
"""Return the sha1 of a file.
146
abspath = os.sep.join([self.basedir, path])
134
abspath = pathjoin(self.root, path)
147
135
self.stat_count += 1
148
file_fp = _fingerprint(abspath)
136
file_fp = self._fingerprint(abspath)
151
139
# not a regular file or not existing
161
149
cache_sha1, cache_fp = None, None
163
151
if cache_fp == file_fp:
152
## mutter("hashcache hit for %s %r -> %s", path, file_fp, cache_sha1)
153
## mutter("now = %s", time.time())
164
154
self.hit_count += 1
165
155
return cache_sha1
167
157
self.miss_count += 1
170
159
mode = file_fp[FP_MODE_COLUMN]
171
160
if stat.S_ISREG(mode):
172
digest = sha_file(file(abspath, 'rb', buffering=65000))
161
digest = self._really_sha1_file(abspath)
173
162
elif stat.S_ISLNK(mode):
174
link_target = os.readlink(abspath)
175
163
digest = sha.new(os.readlink(abspath)).hexdigest()
177
165
raise BzrError("file %r: unknown file stat mode: %o"%(abspath,mode))
179
now = int(time.time())
180
if file_fp[1] >= now or file_fp[2] >= now:
167
# window of 3 seconds to allow for 2s resolution on windows,
168
# unsynchronized file servers, etc.
169
cutoff = self._cutoff_time()
170
if file_fp[FP_MTIME_COLUMN] >= cutoff \
171
or file_fp[FP_CTIME_COLUMN] >= cutoff:
181
172
# changed too recently; can't be cached. we can
182
173
# return the result and it could possibly be cached
184
self.danger_count += 1
176
# the point is that we only want to cache when we are sure that any
177
# subsequent modifications of the file can be detected. If a
178
# modification neither changes the inode, the device, the size, nor
179
# the mode, then we can only distinguish it by time; therefore we
180
# need to let sufficient time elapse before we may cache this entry
181
# again. If we didn't do this, then, for example, a very quick 1
182
# byte replacement in the file might go undetected.
183
## mutter('%r modified too recently; not caching', path)
184
self.danger_count += 1
186
186
self.removed_count += 1
187
187
self.needs_write = True
188
188
del self._cache[path]
190
## mutter('%r added to cache: now=%f, mtime=%d, ctime=%d',
191
## path, time.time(), file_fp[FP_MTIME_COLUMN],
192
## file_fp[FP_CTIME_COLUMN])
190
193
self.update_count += 1
191
194
self.needs_write = True
192
195
self._cache[path] = (digest, file_fp)
198
def _really_sha1_file(self, abspath):
199
"""Calculate the SHA1 of a file by reading the full text"""
200
return sha_file(file(abspath, 'rb', buffering=65000))
196
203
"""Write contents of cache to file."""
197
outf = AtomicFile(self.cache_file_name(), 'wb')
204
outf = AtomicFile(self.cache_file_name(), 'wb', new_mode=self._mode)
199
206
print >>outf, CACHE_HEADER,
207
214
print >>outf, "%d" % fld,
211
217
self.needs_write = False
218
## mutter("write hash cache: %s hits=%d misses=%d stat=%d recent=%d updates=%d",
219
## self.cache_file_name(), self.hit_count, self.miss_count,
221
## self.danger_count, self.update_count)
213
223
if not outf.closed:
219
227
"""Reinstate cache from file.
229
237
inf = file(fn, 'rb', buffering=65000)
230
238
except IOError, e:
231
mutter("failed to open %s: %s" % (fn, e))
239
mutter("failed to open %s: %s", fn, e)
232
240
# better write it now so it is valid
233
241
self.needs_write = True
237
244
hdr = inf.readline()
238
245
if hdr != CACHE_HEADER:
239
mutter('cache header marker not found at top of %s; discarding cache'
246
mutter('cache header marker not found at top of %s;'
247
' discarding cache', fn)
241
248
self.needs_write = True
264
271
self._cache[path] = (sha1, fp)
266
273
self.needs_write = False
275
def _cutoff_time(self):
276
"""Return cutoff time.
278
Files modified more recently than this time are at risk of being
279
undetectably modified and so can't be cached.
281
return int(time.time()) - 3
283
def _fingerprint(self, abspath):
285
fs = os.lstat(abspath)
287
# might be missing, etc
289
if stat.S_ISDIR(fs.st_mode):
291
# we discard any high precision because it's not reliable; perhaps we
292
# could do better on some systems?
293
return (fs.st_size, long(fs.st_mtime),
294
long(fs.st_ctime), fs.st_ino, fs.st_dev, fs.st_mode)