Commit 165434ea authored by Jérome Perrin's avatar Jérome Perrin Committed by Rafael Monnerat

persistent_migration: convert object states with zodbupdate when loading

See https://zope.readthedocs.io/en/latest/migrations/zope4/zodb.html

The approach here is to have a ZODB patch to apply zodbupdate on object
load, decoding with UTF-8 encoding because it will be correct most of
the time with ERP5
parent 1ce9ae40
......@@ -198,6 +198,11 @@ def runwsgi():
help='Set soft limit of file descriptors erp5 can open to hard limit',
action="store_true")
parser.add_argument('--enable-xml-rpc', help='Enable XML-RPC interface', action='store_true')
parser.add_argument(
'--zodbupdate',
help='Use zodbupdate to convert object states pickles on the fly. '
'Only makes sense when using a database created on python2 with python3',
action="store_true")
args = parser.parse_args()
if not sys.warnoptions:
......@@ -244,6 +249,10 @@ def runwsgi():
f.write('%s\n' % os.getpid())
atexit.register(os.unlink, args.pidfile)
if args.zodbupdate:
from Products.ERP5Type.dynamic.persistent_migration import enable_zodbupdate_load_monkey_patch
enable_zodbupdate_load_monkey_patch()
startup = os.path.dirname(Zope2.Startup.__file__)
if os.path.isfile(os.path.join(startup, 'wsgischema.xml')):
schema = ZConfig.loadSchema(os.path.join(startup, 'wsgischema.xml'))
......
......@@ -182,7 +182,10 @@ class String(Scalar):
# This is used when strings represent references which need to
# be converted.
encoding = 'base64'
v = base64_encodebytes(self._v)[:-1]
v = self._v
if not isinstance(v, bytes):
v = v.encode()
v = base64_encodebytes(v)[:-1]
self._v = bytes2str(self.mapping.convertBase64(v))
else:
encoding, self._v = convert(self._v)
......
......@@ -220,3 +220,113 @@ if 1:
else:
__setstate__ = None
def enable_zodbupdate_load_monkey_patch():
import six
assert six.PY3
from _compat_pickle import NAME_MAPPING
from _compat_pickle import IMPORT_MAPPING
from io import BytesIO
from ZODB._compat import Unpickler
import zodbpickle.pickle
# Make ZODB._compat.Unpickler use ascii / bytes by defaut, which is
# fine for persistent ids, that are supposed to be bytes or ascii str.
# ObjectReader will use utf-8, which is correct most of the time
# and simplify the job of zodbupdate coverters
def _Unpickler__init__(self, f, encoding='ascii', errors='bytes'):
zodbpickle.pickle.Unpickler.__init__(self, f, encoding=encoding, errors=errors)
Unpickler.__init__ = _Unpickler__init__
# Unpickler converts the class names from python2 names to python3 names
# when loading a pickle from protocol < 3, with the assumtion that python2
# can not have produced pickle protocol 3, since it was never supported on
# python3, but zodbpickle backported protocol 3 support for python2, so we
# have prococl 3 pickle which needs to be decoded. This patch just applies
# the conversion regardless of the pickle protocol.
_orig_Unpickler_find_class = Unpickler.find_class
def _Unpickler_find_class(self, modulename, name):
if (modulename, name) in NAME_MAPPING:
modulename, name = NAME_MAPPING[(modulename, name)]
if modulename in IMPORT_MAPPING:
modulename = IMPORT_MAPPING[modulename]
return _orig_Unpickler_find_class(self, modulename, name)
Unpickler.find_class = _Unpickler_find_class
import zodbupdate.convert
_zodbupdate_convert_decoders = zodbupdate.convert.load_decoders()
from ZODB.serialize import ObjectReader
from ZODB._compat import PersistentUnpickler
def _ObjectReader_get_unpickler(self, pickle):
file = BytesIO(pickle)
factory = self._factory
conn = self._conn
def find_global(modulename, name):
if (modulename, name) in NAME_MAPPING:
modulename, name = NAME_MAPPING[(modulename, name)]
if modulename in IMPORT_MAPPING:
modulename = IMPORT_MAPPING[modulename]
return factory(conn, modulename, name)
def persistent_load(ooid):
if isinstance(ooid, str):
ooid = ooid.encode()
elif isinstance(ooid, tuple) and isinstance(ooid[0], str):
assert len(ooid) == 2
ooid = (ooid[0].encode(), ooid[1])
return self._persistent_load(ooid)
class PersistentUnpicklerWithMigration:
"""A wrapper around PeristentUnpickler that converts the object
state using zodbupdate while loading.
This only support being called with `load` twice: once for the klass
and then once for the state. If the class is known by zodbupdate, the
state is returned convered.
"""
def __init__(self):
self._unpickler = PersistentUnpickler(
find_global,
persistent_load,
file,
encoding='utf-8',
)
self._klass_modulename = None
self._klass_name = None
def load(self):
if self._klass_modulename is None:
loaded_klass = self._unpickler.load()
# first load the class and remember the module name
# and class name. See ObjectWriter.serialize for the
# three formats.
if isinstance(loaded_klass, type):
self._klass_modulename = loaded_klass.__module__
self._klass_name = loaded_klass.__name__
else:
(klass, _newargs) = loaded_klass
if isinstance(klass, type):
self._klass_modulename = klass.__module__
self._klass_name = klass.__name__
else:
self._klass_modulename, self._klass_name = klass
return loaded_klass
# second, load state and convert it using zodbupdate
state = self._unpickler.load()
for decoder in _zodbupdate_convert_decoders.get(
(self._klass_modulename, self._klass_name), ()):
decoder(state)
return state
return PersistentUnpicklerWithMigration()
ObjectReader._get_unpickler = _ObjectReader_get_unpickler
......@@ -30,12 +30,23 @@ import io
import six.moves.urllib as urllib
import six.moves.http_client
from DateTime import DateTime
import six
if six.PY3:
try:
"str" + b"bytes" # type:ignore
except TypeError:
from Products.ERP5Type.dynamic.persistent_migration import enable_zodbupdate_load_monkey_patch
enable_zodbupdate_load_monkey_patch()
else:
print("Running with pygolang bstr+ustr")
class OldDataFsSetup(ERP5TypeTestCase):
"""Set up of the "old" site, executed when saving.
"""
def setUpOnce(self):
assert six.PY2
self.tic()
self.configure_security()
self.create_person_with_login()
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment