##// END OF EJS Templates
fix(backend): fixed broken backends function after python3 migration
fix(backend): fixed broken backends function after python3 migration

File last commit:

r5165:e6085588 default
r5523:112c3403 stable
Show More
schema.py
721 lines | 24.3 KiB | text/x-python | PythonLexer
"""
Schema module providing common schema operations.
"""
import abc
try: # Python 3
from collections.abc import MutableMapping as DictMixin
except ImportError: # Python 2
from UserDict import DictMixin
import warnings
import sqlalchemy
from sqlalchemy.schema import ForeignKeyConstraint
from sqlalchemy.schema import UniqueConstraint
from rhodecode.lib.dbmigrate.migrate.exceptions import *
from rhodecode.lib.dbmigrate.migrate.changeset import SQLA_07, SQLA_08
from rhodecode.lib.dbmigrate.migrate.changeset import util
from rhodecode.lib.dbmigrate.migrate.changeset.databases.visitor import (
get_engine_visitor, run_single_visitor)
__all__ = [
'create_column',
'drop_column',
'alter_column',
'rename_table',
'rename_index',
'ChangesetTable',
'ChangesetColumn',
'ChangesetIndex',
'ChangesetDefaultClause',
'ColumnDelta',
]
def create_column(column, table=None, *p, **kw):
"""Create a column, given the table.
API to :meth:`ChangesetColumn.create`.
"""
if table is not None:
return table.create_column(column, *p, **kw)
return column.create(*p, **kw)
def drop_column(column, table=None, *p, **kw):
"""Drop a column, given the table.
API to :meth:`ChangesetColumn.drop`.
"""
if table is not None:
return table.drop_column(column, *p, **kw)
return column.drop(*p, **kw)
def rename_table(table, name, engine=None, **kw):
"""Rename a table.
If Table instance is given, engine is not used.
API to :meth:`ChangesetTable.rename`.
:param table: Table to be renamed.
:param name: New name for Table.
:param engine: Engine instance.
:type table: string or Table instance
:type name: string
:type engine: obj
"""
table = _to_table(table, engine)
table.rename(name, **kw)
def rename_index(index, name, table=None, engine=None, **kw):
"""Rename an index.
If Index instance is given,
table and engine are not used.
API to :meth:`ChangesetIndex.rename`.
:param index: Index to be renamed.
:param name: New name for index.
:param table: Table to which Index is reffered.
:param engine: Engine instance.
:type index: string or Index instance
:type name: string
:type table: string or Table instance
:type engine: obj
"""
index = _to_index(index, table, engine)
index.rename(name, **kw)
def alter_column(*p, **k):
"""Alter a column.
This is a helper function that creates a :class:`ColumnDelta` and
runs it.
:argument column:
The name of the column to be altered or a
:class:`ChangesetColumn` column representing it.
:param table:
A :class:`~sqlalchemy.schema.Table` or table name to
for the table where the column will be changed.
:param engine:
The :class:`~sqlalchemy.engine.base.Engine` to use for table
reflection and schema alterations.
:returns: A :class:`ColumnDelta` instance representing the change.
"""
if 'table' not in k and isinstance(p[0], sqlalchemy.Column):
k['table'] = p[0].table
if 'engine' not in k:
k['engine'] = k['table'].bind
# deprecation
if len(p) >= 2 and isinstance(p[1], sqlalchemy.Column):
warnings.warn(
"Passing a Column object to alter_column is deprecated."
" Just pass in keyword parameters instead.",
MigrateDeprecationWarning
)
engine = k['engine']
# enough tests seem to break when metadata is always altered
# that this crutch has to be left in until they can be sorted
# out
k['alter_metadata']=True
delta = ColumnDelta(*p, **k)
visitorcallable = get_engine_visitor(engine, 'schemachanger')
_run_visitor(engine, visitorcallable, delta)
return delta
def _to_table(table, engine=None):
"""Return if instance of Table, else construct new with metadata"""
if isinstance(table, sqlalchemy.Table):
return table
# Given: table name, maybe an engine
meta = sqlalchemy.MetaData()
if engine is not None:
meta.bind = engine
return sqlalchemy.Table(table, meta)
def _to_index(index, table=None, engine=None):
"""Return if instance of Index, else construct new with metadata"""
if isinstance(index, sqlalchemy.Index):
return index
# Given: index name; table name required
table = _to_table(table, engine)
ret = sqlalchemy.Index(index)
ret.table = table
return ret
def _run_visitor(
connectable, visitorcallable, element, connection=None, **kwargs
):
if connection is not None:
visitorcallable(
connection.dialect, connection, **kwargs).traverse_single(element)
else:
conn = connectable.connect()
try:
visitorcallable(
conn.dialect, conn, **kwargs).traverse_single(element)
finally:
conn.close()
# Python3: if we just use:
#
# class ColumnDelta(DictMixin, sqlalchemy.schema.SchemaItem):
# ...
#
# We get the following error:
# TypeError: metaclass conflict: the metaclass of a derived class must be a
# (non-strict) subclass of the metaclasses of all its bases.
#
# The complete inheritance/metaclass relationship list of ColumnDelta can be
# summarized by this following dot file:
#
# digraph test123 {
# ColumnDelta -> MutableMapping;
# MutableMapping -> Mapping;
# Mapping -> {Sized Iterable Container};
# {Sized Iterable Container} -> ABCMeta[style=dashed];
#
# ColumnDelta -> SchemaItem;
# SchemaItem -> {SchemaEventTarget Visitable};
# SchemaEventTarget -> object;
# Visitable -> {VisitableType object} [style=dashed];
# VisitableType -> type;
# }
#
# We need to use a metaclass that inherits from all the metaclasses of
# DictMixin and sqlalchemy.schema.SchemaItem. Let's call it "MyMeta".
class MyMeta(sqlalchemy.sql.visitors.VisitableType, abc.ABCMeta, object):
pass
class ColumnDelta(DictMixin, sqlalchemy.schema.SchemaItem, metaclass=MyMeta):
"""Extracts the differences between two columns/column-parameters
May receive parameters arranged in several different ways:
* **current_column, new_column, \*p, \*\*kw**
Additional parameters can be specified to override column
differences.
* **current_column, \*p, \*\*kw**
Additional parameters alter current_column. Table name is extracted
from current_column object.
Name is changed to current_column.name from current_name,
if current_name is specified.
* **current_col_name, \*p, \*\*kw**
Table kw must specified.
:param table: Table at which current Column should be bound to.\
If table name is given, reflection will be used.
:type table: string or Table instance
:param metadata: A :class:`MetaData` instance to store
reflected table names
:param engine: When reflecting tables, either engine or metadata must \
be specified to acquire engine object.
:type engine: :class:`Engine` instance
:returns: :class:`ColumnDelta` instance provides interface for altered attributes to \
`result_column` through :func:`dict` alike object.
* :class:`ColumnDelta`.result_column is altered column with new attributes
* :class:`ColumnDelta`.current_name is current name of column in db
"""
# Column attributes that can be altered
diff_keys = ('name', 'type', 'primary_key', 'nullable',
'server_onupdate', 'server_default', 'autoincrement')
diffs = dict()
__visit_name__ = 'column'
def __init__(self, *p, **kw):
# 'alter_metadata' is not a public api. It exists purely
# as a crutch until the tests that fail when 'alter_metadata'
# behaviour always happens can be sorted out
self.alter_metadata = kw.pop("alter_metadata", False)
self.meta = kw.pop("metadata", None)
self.engine = kw.pop("engine", None)
# Things are initialized differently depending on how many column
# parameters are given. Figure out how many and call the appropriate
# method.
if len(p) >= 1 and isinstance(p[0], sqlalchemy.Column):
# At least one column specified
if len(p) >= 2 and isinstance(p[1], sqlalchemy.Column):
# Two columns specified
diffs = self.compare_2_columns(*p, **kw)
else:
# Exactly one column specified
diffs = self.compare_1_column(*p, **kw)
else:
# Zero columns specified
if not len(p) or not isinstance(p[0], str):
raise ValueError("First argument must be column name")
diffs = self.compare_parameters(*p, **kw)
self.apply_diffs(diffs)
def __repr__(self):
return '<ColumnDelta altermetadata=%r, %s>' % (
self.alter_metadata,
super(ColumnDelta, self).__repr__()
)
def __getitem__(self, key):
if key not in list(self.keys()):
raise KeyError("No such diff key, available: %s" % self.diffs )
return getattr(self.result_column, key)
def __setitem__(self, key, value):
if key not in list(self.keys()):
raise KeyError("No such diff key, available: %s" % self.diffs )
setattr(self.result_column, key, value)
def __delitem__(self, key):
raise NotImplementedError
def __len__(self):
raise NotImplementedError
def __iter__(self):
raise NotImplementedError
def keys(self):
return list(self.diffs.keys())
def compare_parameters(self, current_name, *p, **k):
"""Compares Column objects with reflection"""
self.table = k.pop('table')
self.result_column = self._table.c.get(current_name)
if len(p):
k = self._extract_parameters(p, k, self.result_column)
return k
def compare_1_column(self, col, *p, **k):
"""Compares one Column object"""
self.table = k.pop('table', None)
if self.table is None:
self.table = col.table
self.result_column = col
if len(p):
k = self._extract_parameters(p, k, self.result_column)
return k
def compare_2_columns(self, old_col, new_col, *p, **k):
"""Compares two Column objects"""
self.process_column(new_col)
self.table = k.pop('table', None)
# we cannot use bool() on table in SA06
if self.table is None:
self.table = old_col.table
if self.table is None:
new_col.table
self.result_column = old_col
# set differences
# leave out some stuff for later comp
for key in (set(self.diff_keys) - set(('type',))):
val = getattr(new_col, key, None)
if getattr(self.result_column, key, None) != val:
k.setdefault(key, val)
# inspect types
if not self.are_column_types_eq(self.result_column.type, new_col.type):
k.setdefault('type', new_col.type)
if len(p):
k = self._extract_parameters(p, k, self.result_column)
return k
def apply_diffs(self, diffs):
"""Populate dict and column object with new values"""
self.diffs = diffs
for key in self.diff_keys:
if key in diffs:
setattr(self.result_column, key, diffs[key])
self.process_column(self.result_column)
# create an instance of class type if not yet
if 'type' in diffs:
if callable(self.result_column.type):
self.result_column.type = self.result_column.type()
if self.result_column.autoincrement and \
not issubclass(
self.result_column.type._type_affinity,
sqlalchemy.Integer):
self.result_column.autoincrement = False
# add column to the table
if self.table is not None and self.alter_metadata:
self.result_column.add_to_table(self.table)
def are_column_types_eq(self, old_type, new_type):
"""Compares two types to be equal"""
ret = old_type.__class__ == new_type.__class__
# String length is a special case
if ret and isinstance(new_type, sqlalchemy.types.String):
ret = (getattr(old_type, 'length', None) == \
getattr(new_type, 'length', None))
return ret
def _extract_parameters(self, p, k, column):
"""Extracts data from p and modifies diffs"""
p = list(p)
while len(p):
if isinstance(p[0], str):
k.setdefault('name', p.pop(0))
elif isinstance(p[0], sqlalchemy.types.TypeEngine):
k.setdefault('type', p.pop(0))
elif callable(p[0]):
p[0] = p[0]()
else:
break
if len(p):
new_col = column.copy_fixed()
new_col._init_items(*p)
k = self.compare_2_columns(column, new_col, **k)
return k
def process_column(self, column):
"""Processes default values for column"""
# XXX: this is a snippet from SA processing of positional parameters
toinit = list()
if column.server_default is not None:
if isinstance(column.server_default, sqlalchemy.FetchedValue):
toinit.append(column.server_default)
else:
toinit.append(sqlalchemy.DefaultClause(column.server_default))
if column.server_onupdate is not None:
if isinstance(column.server_onupdate, FetchedValue):
toinit.append(column.server_default)
else:
toinit.append(sqlalchemy.DefaultClause(column.server_onupdate,
for_update=True))
if toinit:
column._init_items(*toinit)
def _get_table(self):
return getattr(self, '_table', None)
def _set_table(self, table):
if isinstance(table, str):
if self.alter_metadata:
if not self.meta:
raise ValueError("metadata must be specified for table"
" reflection when using alter_metadata")
meta = self.meta
if self.engine:
meta.bind = self.engine
else:
if not self.engine and not self.meta:
raise ValueError("engine or metadata must be specified"
" to reflect tables")
if not self.engine:
self.engine = self.meta.bind
meta = sqlalchemy.MetaData(bind=self.engine)
self._table = sqlalchemy.Table(table, meta, autoload=True)
elif isinstance(table, sqlalchemy.Table):
self._table = table
if not self.alter_metadata:
self._table.meta = sqlalchemy.MetaData(bind=self._table.bind)
def _get_result_column(self):
return getattr(self, '_result_column', None)
def _set_result_column(self, column):
"""Set Column to Table based on alter_metadata evaluation."""
self.process_column(column)
if not hasattr(self, 'current_name'):
self.current_name = column.name
if self.alter_metadata:
self._result_column = column
else:
self._result_column = column.copy_fixed()
table = property(_get_table, _set_table)
result_column = property(_get_result_column, _set_result_column)
class ChangesetTable(object):
"""Changeset extensions to SQLAlchemy tables."""
def create_column(self, column, *p, **kw):
"""Creates a column.
The column parameter may be a column definition or the name of
a column in this table.
API to :meth:`ChangesetColumn.create`
:param column: Column to be created
:type column: Column instance or string
"""
if not isinstance(column, sqlalchemy.Column):
# It's a column name
column = getattr(self.c, str(column))
column.create(table=self, *p, **kw)
def drop_column(self, column, *p, **kw):
"""Drop a column, given its name or definition.
API to :meth:`ChangesetColumn.drop`
:param column: Column to be droped
:type column: Column instance or string
"""
if not isinstance(column, sqlalchemy.Column):
# It's a column name
try:
column = getattr(self.c, str(column))
except AttributeError:
# That column isn't part of the table. We don't need
# its entire definition to drop the column, just its
# name, so create a dummy column with the same name.
column = sqlalchemy.Column(str(column), sqlalchemy.Integer())
column.drop(table=self, *p, **kw)
def rename(self, name, connection=None, **kwargs):
"""Rename this table.
:param name: New name of the table.
:type name: string
:param connection: reuse connection istead of creating new one.
:type connection: :class:`sqlalchemy.engine.base.Connection` instance
"""
engine = self.bind
self.new_name = name
visitorcallable = get_engine_visitor(engine, 'schemachanger')
run_single_visitor(engine, visitorcallable, self, connection, **kwargs)
# Fix metadata registration
self.name = name
self.deregister()
self._set_parent(self.metadata)
def _meta_key(self):
"""Get the meta key for this table."""
return sqlalchemy.schema._get_table_key(self.name, self.schema)
def deregister(self):
"""Remove this table from its metadata"""
if SQLA_07:
self.metadata._remove_table(self.name, self.schema)
else:
key = self._meta_key()
meta = self.metadata
if key in meta.tables:
del meta.tables[key]
class ChangesetColumn(object):
"""Changeset extensions to SQLAlchemy columns."""
def alter(self, *p, **k):
"""Makes a call to :func:`alter_column` for the column this
method is called on.
"""
if 'table' not in k:
k['table'] = self.table
if 'engine' not in k:
k['engine'] = k['table'].bind
return alter_column(self, *p, **k)
def create(self, table=None, index_name=None, unique_name=None,
primary_key_name=None, populate_default=True, connection=None, **kwargs):
"""Create this column in the database.
Assumes the given table exists. ``ALTER TABLE ADD COLUMN``,
for most databases.
:param table: Table instance to create on.
:param index_name: Creates :class:`ChangesetIndex` on this column.
:param unique_name: Creates :class:\
`~migrate.changeset.constraint.UniqueConstraint` on this column.
:param primary_key_name: Creates :class:\
`~migrate.changeset.constraint.PrimaryKeyConstraint` on this column.
:param populate_default: If True, created column will be \
populated with defaults
:param connection: reuse connection istead of creating new one.
:type table: Table instance
:type index_name: string
:type unique_name: string
:type primary_key_name: string
:type populate_default: bool
:type connection: :class:`sqlalchemy.engine.base.Connection` instance
:returns: self
"""
self.populate_default = populate_default
self.index_name = index_name
self.unique_name = unique_name
self.primary_key_name = primary_key_name
for cons in ('index_name', 'unique_name', 'primary_key_name'):
self._check_sanity_constraints(cons)
self.add_to_table(table)
engine = self.table.bind
visitorcallable = get_engine_visitor(engine, 'columngenerator')
_run_visitor(engine, visitorcallable, self, connection, **kwargs)
# TODO: reuse existing connection
if self.populate_default and self.default is not None:
stmt = table.update().values({self: engine._execute_default(self.default)})
engine.execute(stmt)
return self
def drop(self, table=None, connection=None, **kwargs):
"""Drop this column from the database, leaving its table intact.
``ALTER TABLE DROP COLUMN``, for most databases.
:param connection: reuse connection istead of creating new one.
:type connection: :class:`sqlalchemy.engine.base.Connection` instance
"""
if table is not None:
self.table = table
engine = self.table.bind
visitorcallable = get_engine_visitor(engine, 'columndropper')
_run_visitor(engine, visitorcallable, self, connection, **kwargs)
self.remove_from_table(self.table, unset_table=False)
self.table = None
return self
def add_to_table(self, table):
if table is not None and self.table is None:
if SQLA_07:
table.append_column(self)
else:
self._set_parent(table)
def _col_name_in_constraint(self,cons,name):
return False
def remove_from_table(self, table, unset_table=True):
# TODO: remove primary keys, constraints, etc
if unset_table:
self.table = None
to_drop = set()
for index in table.indexes:
columns = []
for col in index.columns:
if col.name!=self.name:
columns.append(col)
if columns:
index.columns = columns
if SQLA_08:
index.expressions = columns
else:
to_drop.add(index)
table.indexes = table.indexes - to_drop
to_drop = set()
for cons in table.constraints:
# TODO: deal with other types of constraint
if isinstance(cons,(ForeignKeyConstraint,
UniqueConstraint)):
for col_name in cons.columns:
if not isinstance(col_name, str):
col_name = col_name.name
if self.name==col_name:
to_drop.add(cons)
table.constraints = table.constraints - to_drop
if table.c.contains_column(self):
if SQLA_07:
table._columns.remove(self)
else:
table.c.remove(self)
# TODO: this is fixed in 0.6
def copy_fixed(self, **kw):
"""Create a copy of this ``Column``, with all attributes."""
q = util.safe_quote(self)
return sqlalchemy.Column(self.name, self.type, self.default,
key=self.key,
primary_key=self.primary_key,
nullable=self.nullable,
quote=q,
index=self.index,
unique=self.unique,
onupdate=self.onupdate,
autoincrement=self.autoincrement,
server_default=self.server_default,
server_onupdate=self.server_onupdate,
*[c.copy(**kw) for c in self.constraints])
def _check_sanity_constraints(self, name):
"""Check if constraints names are correct"""
obj = getattr(self, name)
if (getattr(self, name[:-5]) and not obj):
raise InvalidConstraintError("Column.create() accepts index_name,"
" primary_key_name and unique_name to generate constraints")
if not isinstance(obj, str) and obj is not None:
raise InvalidConstraintError(
"%s argument for column must be constraint name" % name)
class ChangesetIndex(object):
"""Changeset extensions to SQLAlchemy Indexes."""
__visit_name__ = 'index'
def rename(self, name, connection=None, **kwargs):
"""Change the name of an index.
:param name: New name of the Index.
:type name: string
:param connection: reuse connection istead of creating new one.
:type connection: :class:`sqlalchemy.engine.base.Connection` instance
"""
engine = self.table.bind
self.new_name = name
visitorcallable = get_engine_visitor(engine, 'schemachanger')
engine._run_visitor(visitorcallable, self, connection, **kwargs)
self.name = name
class ChangesetDefaultClause(object):
"""Implements comparison between :class:`DefaultClause` instances"""
def __eq__(self, other):
if isinstance(other, self.__class__):
if self.arg == other.arg:
return True
def __ne__(self, other):
return not self.__eq__(other)