20
a cache for chewed-up "change" data structures, which are basically just a
21
different way of storing a revision. the cache improves lookup times 10x
22
over bazaar's xml revision structure, though, so currently still worth doing.
20
a cache for chewed-up 'file change' data structures, which are basically just
21
a different way of storing a revision delta. the cache improves lookup times
22
10x over bazaar's xml revision structure, though, so currently still worth
24
25
once a revision is committed in bazaar, it never changes, so once we have
25
26
cached a change, it's good forever.
31
from loggerhead import util
32
from loggerhead.lockfile import LockFile
34
with_lock = util.with_lock('_lock', 'ChangeCache')
36
SQLITE_INTERFACE = os.environ.get('SQLITE_INTERFACE', 'sqlite3')
38
if SQLITE_INTERFACE == 'sqlite3':
39
36
from sqlite3 import dbapi2
42
_select_stmt = ("select data from revisiondata where revid = ?"
43
).replace('?', _param_marker)
44
_insert_stmt = ("insert into revisiondata (revid, data) "
45
"values (?, ?)").replace('?', _param_marker)
38
from pysqlite2 import dbapi2
40
# We take an optimistic approach to concurrency here: we might do work twice
41
# in the case of races, but not crash or corrupt data.
43
def safe_init_db(filename, init_sql):
44
# To avoid races around creating the database, we create the db in
45
# a temporary file and rename it into the ultimate location.
46
fd, temp_path = tempfile.mkstemp(dir=os.path.dirname(filename))
48
con = dbapi2.connect(temp_path)
53
os.rename(temp_path, filename)
50
55
class FakeShelf(object):
51
57
def __init__(self, filename):
52
58
create_table = not os.path.exists(filename)
61
filename, "create table RevisionData "
62
"(revid binary primary key, data binary)")
53
63
self.connection = dbapi2.connect(filename)
54
64
self.cursor = self.connection.cursor()
57
def _create_table(self):
66
def _create_table(self, filename):
67
con = dbapi2.connect(filename)
59
70
"create table RevisionData "
60
71
"(revid binary primary key, data binary)")
61
self.connection.commit()
62
75
def _serialize(self, obj):
63
r = dbapi2.Binary(cPickle.dumps(obj, protocol=2))
76
return dbapi2.Binary(cPickle.dumps(obj, protocol=2))
65
78
def _unserialize(self, data):
66
79
return cPickle.loads(str(data))
67
81
def get(self, revid):
68
self.cursor.execute(_select_stmt, (revid,))
83
"select data from revisiondata where revid = ?", (revid, ))
69
84
filechange = self.cursor.fetchone()
70
85
if filechange is None:
73
88
return self._unserialize(filechange[0])
74
def add(self, revid_obj_pairs):
75
for (r, d) in revid_obj_pairs:
76
self.cursor.execute(_insert_stmt, (r, self._serialize(d)))
77
self.connection.commit()
90
def add(self, revid, object):
93
"insert into revisiondata (revid, data) values (?, ?)",
94
(revid, self._serialize(object)))
95
self.connection.commit()
96
except dbapi2.IntegrityError:
97
# If another thread or process attempted to set the same key, we
98
# assume it set it to the same value and carry on with our day.
80
102
class FileChangeCache(object):
81
def __init__(self, history, cache_path):
82
self.history = history
104
def __init__(self, cache_path):
84
106
if not os.path.exists(cache_path):
85
107
os.mkdir(cache_path)
87
109
self._changes_filename = os.path.join(cache_path, 'filechanges.sql')
89
# use a lockfile since the cache folder could be shared across
90
# different processes.
91
self._lock = LockFile(os.path.join(cache_path, 'filechange-lock'))
94
def get_file_changes(self, entries):
97
missing_entry_indices = []
111
def get_file_changes(self, entry):
98
112
cache = FakeShelf(self._changes_filename)
100
changes = cache.get(entry.revid)
101
if changes is not None:
104
missing_entries.append(entry)
105
missing_entry_indices.append(len(out))
108
missing_changes = self.history.get_file_changes_uncached(missing_entries)
109
revid_changes_pairs = []
110
for i, entry, changes in zip(
111
missing_entry_indices, missing_entries, missing_changes):
112
revid_changes_pairs.append((entry.revid, changes))
114
cache.add(revid_changes_pairs)
113
changes = cache.get(entry.revid)
115
changes = self.history.get_file_changes_uncached(entry)
116
cache.add(entry.revid, changes)
120
class RevInfoDiskCache(object):
121
"""Like `RevInfoMemoryCache` but backed in a sqlite DB."""
123
def __init__(self, cache_path):
124
if not os.path.exists(cache_path):
126
filename = os.path.join(cache_path, 'revinfo.sql')
127
create_table = not os.path.exists(filename)
130
filename, "create table Data "
131
"(key binary primary key, revid binary, data binary)")
132
self.connection = dbapi2.connect(filename)
133
self.cursor = self.connection.cursor()
135
def get(self, key, revid):
137
"select revid, data from data where key = ?", (dbapi2.Binary(key),))
138
row = self.cursor.fetchone()
141
elif str(row[0]) != revid:
144
return marshal.loads(zlib.decompress(row[1]))
146
def set(self, key, revid, data):
149
'delete from data where key = ?', (dbapi2.Binary(key), ))
150
blob = zlib.compress(marshal.dumps(data))
152
"insert into data (key, revid, data) values (?, ?, ?)",
153
map(dbapi2.Binary, [key, revid, blob]))
154
self.connection.commit()
155
except dbapi2.IntegrityError:
156
# If another thread or process attempted to set the same key, we
157
# don't care too much -- it's only a cache after all!