~bzr-pqm/bzr/bzr.dev

2052.3.2 by John Arbash Meinel
Change Copyright .. by Canonical to Copyright ... Canonical
1
# Copyright (C) 2005, 2006 Canonical Ltd
1887.1.1 by Adeodato Simó
Do not separate paragraphs in the copyright statement with blank lines,
2
#
846 by Martin Pool
- start adding refactored/simplified hash cache
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
1887.1.1 by Adeodato Simó
Do not separate paragraphs in the copyright statement with blank lines,
7
#
846 by Martin Pool
- start adding refactored/simplified hash cache
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
11
# GNU General Public License for more details.
1887.1.1 by Adeodato Simó
Do not separate paragraphs in the copyright statement with blank lines,
12
#
846 by Martin Pool
- start adding refactored/simplified hash cache
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
4183.7.1 by Sabin Iacob
update FSF mailing address
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
846 by Martin Pool
- start adding refactored/simplified hash cache
16
3943.8.1 by Marius Kruger
remove all trailing whitespace from bzr source
17
# TODO: Up-front, stat all files in order and remove those which are deleted or
18
# out-of-date.  Don't actually re-read them until they're needed.  That ought
19
# to bring all the inodes into core so that future stats to them are fast, and
953 by Martin Pool
- refactor imports and stats for hashcache
20
# it preserves the nice property that any caller will always get up-to-date
21
# data except in unavoidable cases.
864 by Martin Pool
doc
22
23
# TODO: Perhaps return more details on the file to avoid statting it
24
# again: nonexistent, file type, size, etc
25
1213 by Martin Pool
- move import in hashcache
26
# TODO: Perhaps use a Python pickle instead of a text file; might be faster.
27
864 by Martin Pool
doc
28
29
866 by Martin Pool
- use new path-based hashcache for WorkingTree- squash mtime/ctime to whole seconds- update and if necessary write out hashcache when WorkingTree object is created.
30
CACHE_HEADER = "### bzr hashcache v5\n"
859 by Martin Pool
- add HashCache.write and a simple test for it
31
4241.14.15 by Vincent Ladeuil
Fix one unicode readlink related test failure.
32
import os
33
import stat
34
import time
953 by Martin Pool
- refactor imports and stats for hashcache
35
4241.14.15 by Vincent Ladeuil
Fix one unicode readlink related test failure.
36
from bzrlib import (
37
    atomicfile,
38
    errors,
39
    filters as _mod_filters,
40
    osutils,
41
    trace,
42
    )
1540.1.1 by Martin Pool
[patch] stat-cache fixes from Denys
43
44
1185.59.10 by Denys Duchier
hashcache: new constants and improved comment
45
FP_MTIME_COLUMN = 1
46
FP_CTIME_COLUMN = 2
1092.2.6 by Robert Collins
symlink support updated to work
47
FP_MODE_COLUMN = 5
859 by Martin Pool
- add HashCache.write and a simple test for it
48
846 by Martin Pool
- start adding refactored/simplified hash cache
49
50
51
class HashCache(object):
52
    """Cache for looking up file SHA-1.
53
54
    Files are considered to match the cached value if the fingerprint
55
    of the file has not changed.  This includes its mtime, ctime,
56
    device number, inode number, and size.  This should catch
57
    modifications or replacement of the file by a new one.
58
59
    This may not catch modifications that do not change the file's
60
    size and that occur within the resolution window of the
61
    timestamps.  To handle this we specifically do not cache files
62
    which have changed since the start of the present second, since
63
    they could undetectably change again.
64
65
    This scheme may fail if the machine's clock steps backwards.
66
    Don't do that.
67
68
    This does not canonicalize the paths passed in; that should be
69
    done by the caller.
70
860 by Martin Pool
- refactor hashcache to use just one dictionary
71
    _cache
72
        Indexed by path, points to a two-tuple of the SHA-1 of the file.
73
        and its fingerprint.
846 by Martin Pool
- start adding refactored/simplified hash cache
74
75
    stat_count
76
        number of times files have been statted
77
78
    hit_count
79
        number of times files have been retrieved from the cache, avoiding a
80
        re-read
3943.8.1 by Marius Kruger
remove all trailing whitespace from bzr source
81
846 by Martin Pool
- start adding refactored/simplified hash cache
82
    miss_count
83
        number of misses (times files have been completely re-read)
84
    """
866 by Martin Pool
- use new path-based hashcache for WorkingTree- squash mtime/ctime to whole seconds- update and if necessary write out hashcache when WorkingTree object is created.
85
    needs_write = False
86
3368.2.4 by Ian Clatworthy
make content filter lookup a tree responsibility
87
    def __init__(self, root, cache_file_name, mode=None,
88
            content_filter_stack_provider=None):
89
        """Create a hash cache in base dir, and set the file mode to mode.
90
3368.2.5 by Ian Clatworthy
incorporate jameinel's review feedback
91
        :param content_filter_stack_provider: a function that takes a
92
            path (relative to the top of the tree) and a file-id as
3368.2.19 by Ian Clatworthy
first round of changes from abentley's review
93
            parameters and returns a stack of ContentFilters.
3368.2.5 by Ian Clatworthy
incorporate jameinel's review feedback
94
            If None, no content filtering is performed.
3368.2.4 by Ian Clatworthy
make content filter lookup a tree responsibility
95
        """
4241.14.15 by Vincent Ladeuil
Fix one unicode readlink related test failure.
96
        self.root = osutils.safe_unicode(root)
2255.2.149 by Robert Collins
Crufty but existing _iter_changes implementation for WorkingTreeFormat4.
97
        self.root_utf8 = self.root.encode('utf8') # where is the filesystem encoding ?
846 by Martin Pool
- start adding refactored/simplified hash cache
98
        self.hit_count = 0
99
        self.miss_count = 0
100
        self.stat_count = 0
101
        self.danger_count = 0
953 by Martin Pool
- refactor imports and stats for hashcache
102
        self.removed_count = 0
954 by Martin Pool
- separate out code that just scans the hash cache to find files that are possibly
103
        self.update_count = 0
860 by Martin Pool
- refactor hashcache to use just one dictionary
104
        self._cache = {}
1534.4.50 by Robert Collins
Got the bzrdir api straightened out, plenty of refactoring to use it pending, but the api is up and running.
105
        self._mode = mode
4241.14.15 by Vincent Ladeuil
Fix one unicode readlink related test failure.
106
        self._cache_file_name = osutils.safe_unicode(cache_file_name)
3368.2.23 by Ian Clatworthy
cleanup some method names
107
        self._filter_provider = content_filter_stack_provider
846 by Martin Pool
- start adding refactored/simplified hash cache
108
866 by Martin Pool
- use new path-based hashcache for WorkingTree- squash mtime/ctime to whole seconds- update and if necessary write out hashcache when WorkingTree object is created.
109
    def cache_file_name(self):
1534.4.51 by Robert Collins
Test the disk layout of format3 working trees.
110
        return self._cache_file_name
866 by Martin Pool
- use new path-based hashcache for WorkingTree- squash mtime/ctime to whole seconds- update and if necessary write out hashcache when WorkingTree object is created.
111
846 by Martin Pool
- start adding refactored/simplified hash cache
112
    def clear(self):
860 by Martin Pool
- refactor hashcache to use just one dictionary
113
        """Discard all cached information.
114
115
        This does not reset the counters."""
866 by Martin Pool
- use new path-based hashcache for WorkingTree- squash mtime/ctime to whole seconds- update and if necessary write out hashcache when WorkingTree object is created.
116
        if self._cache:
117
            self.needs_write = True
118
            self._cache = {}
846 by Martin Pool
- start adding refactored/simplified hash cache
119
954 by Martin Pool
- separate out code that just scans the hash cache to find files that are possibly
120
    def scan(self):
121
        """Scan all files and remove entries where the cache entry is obsolete.
3943.8.1 by Marius Kruger
remove all trailing whitespace from bzr source
122
954 by Martin Pool
- separate out code that just scans the hash cache to find files that are possibly
123
        Obsolete entries are those where the file has been modified or deleted
3943.8.1 by Marius Kruger
remove all trailing whitespace from bzr source
124
        since the entry was inserted.
954 by Martin Pool
- separate out code that just scans the hash cache to find files that are possibly
125
        """
1534.4.50 by Robert Collins
Got the bzrdir api straightened out, plenty of refactoring to use it pending, but the api is up and running.
126
        # FIXME optimisation opportunity, on linux [and check other oses]:
127
        # rather than iteritems order, stat in inode order.
954 by Martin Pool
- separate out code that just scans the hash cache to find files that are possibly
128
        prep = [(ce[1][3], path, ce) for (path, ce) in self._cache.iteritems()]
953 by Martin Pool
- refactor imports and stats for hashcache
129
        prep.sort()
3943.8.1 by Marius Kruger
remove all trailing whitespace from bzr source
130
954 by Martin Pool
- separate out code that just scans the hash cache to find files that are possibly
131
        for inum, path, cache_entry in prep:
4241.14.15 by Vincent Ladeuil
Fix one unicode readlink related test failure.
132
            abspath = osutils.pathjoin(self.root, path)
1845.1.3 by Martin Pool
Improvements to hashcache testing:
133
            fp = self._fingerprint(abspath)
954 by Martin Pool
- separate out code that just scans the hash cache to find files that are possibly
134
            self.stat_count += 1
3943.8.1 by Marius Kruger
remove all trailing whitespace from bzr source
135
954 by Martin Pool
- separate out code that just scans the hash cache to find files that are possibly
136
            cache_fp = cache_entry[1]
3943.8.1 by Marius Kruger
remove all trailing whitespace from bzr source
137
954 by Martin Pool
- separate out code that just scans the hash cache to find files that are possibly
138
            if (not fp) or (cache_fp != fp):
139
                # not here or not a regular file anymore
140
                self.removed_count += 1
141
                self.needs_write = True
142
                del self._cache[path]
143
2012.1.7 by Aaron Bentley
Get tree._iter_changed down to ~ 1 stat per file
144
    def get_sha1(self, path, stat_value=None):
953 by Martin Pool
- refactor imports and stats for hashcache
145
        """Return the sha1 of a file.
846 by Martin Pool
- start adding refactored/simplified hash cache
146
        """
2255.2.149 by Robert Collins
Crufty but existing _iter_changes implementation for WorkingTreeFormat4.
147
        if path.__class__ is str:
4241.14.15 by Vincent Ladeuil
Fix one unicode readlink related test failure.
148
            abspath = osutils.pathjoin(self.root_utf8, path)
2255.2.149 by Robert Collins
Crufty but existing _iter_changes implementation for WorkingTreeFormat4.
149
        else:
4241.14.15 by Vincent Ladeuil
Fix one unicode readlink related test failure.
150
            abspath = osutils.pathjoin(self.root, path)
954 by Martin Pool
- separate out code that just scans the hash cache to find files that are possibly
151
        self.stat_count += 1
2012.1.7 by Aaron Bentley
Get tree._iter_changed down to ~ 1 stat per file
152
        file_fp = self._fingerprint(abspath, stat_value)
3943.8.1 by Marius Kruger
remove all trailing whitespace from bzr source
153
954 by Martin Pool
- separate out code that just scans the hash cache to find files that are possibly
154
        if not file_fp:
155
            # not a regular file or not existing
156
            if path in self._cache:
157
                self.removed_count += 1
158
                self.needs_write = True
159
                del self._cache[path]
3943.8.1 by Marius Kruger
remove all trailing whitespace from bzr source
160
            return None
953 by Martin Pool
- refactor imports and stats for hashcache
161
954 by Martin Pool
- separate out code that just scans the hash cache to find files that are possibly
162
        if path in self._cache:
163
            cache_sha1, cache_fp = self._cache[path]
860 by Martin Pool
- refactor hashcache to use just one dictionary
164
        else:
165
            cache_sha1, cache_fp = None, None
846 by Martin Pool
- start adding refactored/simplified hash cache
166
954 by Martin Pool
- separate out code that just scans the hash cache to find files that are possibly
167
        if cache_fp == file_fp:
1845.1.2 by mbp at sourcefrog
Use larger time window on hashcache to be safe with fractional times
168
            ## mutter("hashcache hit for %s %r -> %s", path, file_fp, cache_sha1)
169
            ## mutter("now = %s", time.time())
846 by Martin Pool
- start adding refactored/simplified hash cache
170
            self.hit_count += 1
860 by Martin Pool
- refactor hashcache to use just one dictionary
171
            return cache_sha1
3943.8.1 by Marius Kruger
remove all trailing whitespace from bzr source
172
954 by Martin Pool
- separate out code that just scans the hash cache to find files that are possibly
173
        self.miss_count += 1
1092.2.6 by Robert Collins
symlink support updated to work
174
175
        mode = file_fp[FP_MODE_COLUMN]
176
        if stat.S_ISREG(mode):
3368.2.23 by Ian Clatworthy
cleanup some method names
177
            if self._filter_provider is None:
3368.2.5 by Ian Clatworthy
incorporate jameinel's review feedback
178
                filters = []
3368.2.4 by Ian Clatworthy
make content filter lookup a tree responsibility
179
            else:
3368.2.23 by Ian Clatworthy
cleanup some method names
180
                filters = self._filter_provider(path=path, file_id=None)
3368.2.4 by Ian Clatworthy
make content filter lookup a tree responsibility
181
            digest = self._really_sha1_file(abspath, filters)
1092.2.6 by Robert Collins
symlink support updated to work
182
        elif stat.S_ISLNK(mode):
4241.14.15 by Vincent Ladeuil
Fix one unicode readlink related test failure.
183
            target = osutils.readlink(osutils.safe_unicode(abspath))
184
            digest = osutils.sha_string(target.encode('UTF-8'))
1092.2.6 by Robert Collins
symlink support updated to work
185
        else:
4241.14.15 by Vincent Ladeuil
Fix one unicode readlink related test failure.
186
            raise errors.BzrError("file %r: unknown file stat mode: %o"
187
                                  % (abspath, mode))
954 by Martin Pool
- separate out code that just scans the hash cache to find files that are possibly
188
1845.1.2 by mbp at sourcefrog
Use larger time window on hashcache to be safe with fractional times
189
        # window of 3 seconds to allow for 2s resolution on windows,
190
        # unsynchronized file servers, etc.
1845.1.3 by Martin Pool
Improvements to hashcache testing:
191
        cutoff = self._cutoff_time()
1845.1.2 by mbp at sourcefrog
Use larger time window on hashcache to be safe with fractional times
192
        if file_fp[FP_MTIME_COLUMN] >= cutoff \
193
                or file_fp[FP_CTIME_COLUMN] >= cutoff:
954 by Martin Pool
- separate out code that just scans the hash cache to find files that are possibly
194
            # changed too recently; can't be cached.  we can
195
            # return the result and it could possibly be cached
196
            # next time.
1185.59.10 by Denys Duchier
hashcache: new constants and improved comment
197
            #
198
            # the point is that we only want to cache when we are sure that any
199
            # subsequent modifications of the file can be detected.  If a
200
            # modification neither changes the inode, the device, the size, nor
201
            # the mode, then we can only distinguish it by time; therefore we
202
            # need to let sufficient time elapse before we may cache this entry
203
            # again.  If we didn't do this, then, for example, a very quick 1
204
            # byte replacement in the file might go undetected.
1845.1.2 by mbp at sourcefrog
Use larger time window on hashcache to be safe with fractional times
205
            ## mutter('%r modified too recently; not caching', path)
206
            self.danger_count += 1
954 by Martin Pool
- separate out code that just scans the hash cache to find files that are possibly
207
            if cache_fp:
208
                self.removed_count += 1
209
                self.needs_write = True
210
                del self._cache[path]
846 by Martin Pool
- start adding refactored/simplified hash cache
211
        else:
1845.1.2 by mbp at sourcefrog
Use larger time window on hashcache to be safe with fractional times
212
            ## mutter('%r added to cache: now=%f, mtime=%d, ctime=%d',
213
            ##        path, time.time(), file_fp[FP_MTIME_COLUMN],
214
            ##        file_fp[FP_CTIME_COLUMN])
954 by Martin Pool
- separate out code that just scans the hash cache to find files that are possibly
215
            self.update_count += 1
216
            self.needs_write = True
217
            self._cache[path] = (digest, file_fp)
218
        return digest
1845.1.3 by Martin Pool
Improvements to hashcache testing:
219
3368.2.4 by Ian Clatworthy
make content filter lookup a tree responsibility
220
    def _really_sha1_file(self, abspath, filters):
1845.1.3 by Martin Pool
Improvements to hashcache testing:
221
        """Calculate the SHA1 of a file by reading the full text"""
4241.14.15 by Vincent Ladeuil
Fix one unicode readlink related test failure.
222
        return _mod_filters.internal_size_sha_file_byname(abspath, filters)[1]
3943.8.1 by Marius Kruger
remove all trailing whitespace from bzr source
223
866 by Martin Pool
- use new path-based hashcache for WorkingTree- squash mtime/ctime to whole seconds- update and if necessary write out hashcache when WorkingTree object is created.
224
    def write(self):
859 by Martin Pool
- add HashCache.write and a simple test for it
225
        """Write contents of cache to file."""
4241.14.15 by Vincent Ladeuil
Fix one unicode readlink related test failure.
226
        outf = atomicfile.AtomicFile(self.cache_file_name(), 'wb',
227
                                     new_mode=self._mode)
859 by Martin Pool
- add HashCache.write and a simple test for it
228
        try:
1908.4.8 by John Arbash Meinel
Small tweak to hashcache to make it write out faster
229
            outf.write(CACHE_HEADER)
859 by Martin Pool
- add HashCache.write and a simple test for it
230
860 by Martin Pool
- refactor hashcache to use just one dictionary
231
            for path, c  in self._cache.iteritems():
1908.4.8 by John Arbash Meinel
Small tweak to hashcache to make it write out faster
232
                line_info = [path.encode('utf-8'), '// ', c[0], ' ']
233
                line_info.append(' '.join([str(fld) for fld in c[1]]))
234
                line_info.append('\n')
235
                outf.write(''.join(line_info))
859 by Martin Pool
- add HashCache.write and a simple test for it
236
            outf.commit()
866 by Martin Pool
- use new path-based hashcache for WorkingTree- squash mtime/ctime to whole seconds- update and if necessary write out hashcache when WorkingTree object is created.
237
            self.needs_write = False
1845.1.1 by mbp at sourcefrog
Refactor and improve hashcache tests
238
            ## mutter("write hash cache: %s hits=%d misses=%d stat=%d recent=%d updates=%d",
239
            ##        self.cache_file_name(), self.hit_count, self.miss_count,
240
            ##        self.stat_count,
241
            ##        self.danger_count, self.update_count)
859 by Martin Pool
- add HashCache.write and a simple test for it
242
        finally:
1755.3.1 by Robert Collins
Tune the time to build our kernel_like tree : make LocalTransport.put faster, AtomicFile faster, LocalTransport.append faster.
243
            outf.close()
862 by Martin Pool
- code to re-read hashcache from file
244
866 by Martin Pool
- use new path-based hashcache for WorkingTree- squash mtime/ctime to whole seconds- update and if necessary write out hashcache when WorkingTree object is created.
245
    def read(self):
862 by Martin Pool
- code to re-read hashcache from file
246
        """Reinstate cache from file.
247
248
        Overwrites existing cache.
249
3943.8.1 by Marius Kruger
remove all trailing whitespace from bzr source
250
        If the cache file has the wrong version marker, this just clears
862 by Martin Pool
- code to re-read hashcache from file
251
        the cache."""
252
        self._cache = {}
253
866 by Martin Pool
- use new path-based hashcache for WorkingTree- squash mtime/ctime to whole seconds- update and if necessary write out hashcache when WorkingTree object is created.
254
        fn = self.cache_file_name()
255
        try:
948 by Martin Pool
- more buffering when reading/writing hashcache
256
            inf = file(fn, 'rb', buffering=65000)
866 by Martin Pool
- use new path-based hashcache for WorkingTree- squash mtime/ctime to whole seconds- update and if necessary write out hashcache when WorkingTree object is created.
257
        except IOError, e:
4241.14.15 by Vincent Ladeuil
Fix one unicode readlink related test failure.
258
            trace.mutter("failed to open %s: %s", fn, e)
1214 by Martin Pool
- hashcache should be written out if it can't be read
259
            # better write it now so it is valid
260
            self.needs_write = True
866 by Martin Pool
- use new path-based hashcache for WorkingTree- squash mtime/ctime to whole seconds- update and if necessary write out hashcache when WorkingTree object is created.
261
            return
262
862 by Martin Pool
- code to re-read hashcache from file
263
        hdr = inf.readline()
264
        if hdr != CACHE_HEADER:
4241.14.15 by Vincent Ladeuil
Fix one unicode readlink related test failure.
265
            trace.mutter('cache header marker not found at top of %s;'
266
                         ' discarding cache', fn)
1214 by Martin Pool
- hashcache should be written out if it can't be read
267
            self.needs_write = True
862 by Martin Pool
- code to re-read hashcache from file
268
            return
269
270
        for l in inf:
271
            pos = l.index('// ')
272
            path = l[:pos].decode('utf-8')
273
            if path in self._cache:
4241.14.15 by Vincent Ladeuil
Fix one unicode readlink related test failure.
274
                trace.warning('duplicated path %r in cache' % path)
862 by Martin Pool
- code to re-read hashcache from file
275
                continue
276
277
            pos += 3
278
            fields = l[pos:].split(' ')
1092.2.6 by Robert Collins
symlink support updated to work
279
            if len(fields) != 7:
4241.14.15 by Vincent Ladeuil
Fix one unicode readlink related test failure.
280
                trace.warning("bad line in hashcache: %r" % l)
862 by Martin Pool
- code to re-read hashcache from file
281
                continue
282
283
            sha1 = fields[0]
284
            if len(sha1) != 40:
4241.14.15 by Vincent Ladeuil
Fix one unicode readlink related test failure.
285
                trace.warning("bad sha1 in hashcache: %r" % sha1)
862 by Martin Pool
- code to re-read hashcache from file
286
                continue
287
288
            fp = tuple(map(long, fields[1:]))
289
290
            self._cache[path] = (sha1, fp)
291
866 by Martin Pool
- use new path-based hashcache for WorkingTree- squash mtime/ctime to whole seconds- update and if necessary write out hashcache when WorkingTree object is created.
292
        self.needs_write = False
1845.1.3 by Martin Pool
Improvements to hashcache testing:
293
294
    def _cutoff_time(self):
295
        """Return cutoff time.
296
297
        Files modified more recently than this time are at risk of being
298
        undetectably modified and so can't be cached.
299
        """
300
        return int(time.time()) - 3
3943.8.1 by Marius Kruger
remove all trailing whitespace from bzr source
301
2012.1.18 by Aaron Bentley
rename fs param to stat_value
302
    def _fingerprint(self, abspath, stat_value=None):
303
        if stat_value is None:
2012.1.7 by Aaron Bentley
Get tree._iter_changed down to ~ 1 stat per file
304
            try:
2012.1.18 by Aaron Bentley
rename fs param to stat_value
305
                stat_value = os.lstat(abspath)
2012.1.7 by Aaron Bentley
Get tree._iter_changed down to ~ 1 stat per file
306
            except OSError:
307
                # might be missing, etc
308
                return None
2012.1.18 by Aaron Bentley
rename fs param to stat_value
309
        if stat.S_ISDIR(stat_value.st_mode):
1845.1.3 by Martin Pool
Improvements to hashcache testing:
310
            return None
311
        # we discard any high precision because it's not reliable; perhaps we
312
        # could do better on some systems?
2012.1.18 by Aaron Bentley
rename fs param to stat_value
313
        return (stat_value.st_size, long(stat_value.st_mtime),
3943.8.1 by Marius Kruger
remove all trailing whitespace from bzr source
314
                long(stat_value.st_ctime), stat_value.st_ino,
2012.1.18 by Aaron Bentley
rename fs param to stat_value
315
                stat_value.st_dev, stat_value.st_mode)