openmedialibrary_platform/Darwin/lib/python2.7/site-packages/alembic/ddl/impl.py
2014-05-16 01:20:41 +02:00

279 lines
10 KiB
Python

from sqlalchemy.sql.expression import _BindParamClause
from sqlalchemy.ext.compiler import compiles
from sqlalchemy import schema, text
from sqlalchemy import types as sqltypes
from ..compat import string_types, text_type, with_metaclass
from .. import util
from . import base
class ImplMeta(type):
def __init__(cls, classname, bases, dict_):
newtype = type.__init__(cls, classname, bases, dict_)
if '__dialect__' in dict_:
_impls[dict_['__dialect__']] = cls
return newtype
_impls = {}
class DefaultImpl(with_metaclass(ImplMeta)):
"""Provide the entrypoint for major migration operations,
including database-specific behavioral variances.
While individual SQL/DDL constructs already provide
for database-specific implementations, variances here
allow for entirely different sequences of operations
to take place for a particular migration, such as
SQL Server's special 'IDENTITY INSERT' step for
bulk inserts.
"""
__dialect__ = 'default'
transactional_ddl = False
command_terminator = ";"
def __init__(self, dialect, connection, as_sql,
transactional_ddl, output_buffer,
context_opts):
self.dialect = dialect
self.connection = connection
self.as_sql = as_sql
self.output_buffer = output_buffer
self.memo = {}
self.context_opts = context_opts
if transactional_ddl is not None:
self.transactional_ddl = transactional_ddl
@classmethod
def get_by_dialect(cls, dialect):
return _impls[dialect.name]
def static_output(self, text):
self.output_buffer.write(text_type(text + "\n\n"))
self.output_buffer.flush()
@property
def bind(self):
return self.connection
def _exec(self, construct, execution_options=None,
multiparams=(),
params=util.immutabledict()):
if isinstance(construct, string_types):
construct = text(construct)
if self.as_sql:
if multiparams or params:
# TODO: coverage
raise Exception("Execution arguments not allowed with as_sql")
self.static_output(text_type(
construct.compile(dialect=self.dialect)
).replace("\t", " ").strip() + self.command_terminator)
else:
conn = self.connection
if execution_options:
conn = conn.execution_options(**execution_options)
conn.execute(construct, *multiparams, **params)
def execute(self, sql, execution_options=None):
self._exec(sql, execution_options)
def alter_column(self, table_name, column_name,
nullable=None,
server_default=False,
name=None,
type_=None,
schema=None,
autoincrement=None,
existing_type=None,
existing_server_default=None,
existing_nullable=None,
existing_autoincrement=None
):
if autoincrement is not None or existing_autoincrement is not None:
util.warn("nautoincrement and existing_autoincrement only make sense for MySQL")
if nullable is not None:
self._exec(base.ColumnNullable(table_name, column_name,
nullable, schema=schema,
existing_type=existing_type,
existing_server_default=existing_server_default,
existing_nullable=existing_nullable,
))
if server_default is not False:
self._exec(base.ColumnDefault(
table_name, column_name, server_default,
schema=schema,
existing_type=existing_type,
existing_server_default=existing_server_default,
existing_nullable=existing_nullable,
))
if type_ is not None:
self._exec(base.ColumnType(
table_name, column_name, type_, schema=schema,
existing_type=existing_type,
existing_server_default=existing_server_default,
existing_nullable=existing_nullable,
))
# do the new name last ;)
if name is not None:
self._exec(base.ColumnName(
table_name, column_name, name, schema=schema,
existing_type=existing_type,
existing_server_default=existing_server_default,
existing_nullable=existing_nullable,
))
def add_column(self, table_name, column, schema=None):
self._exec(base.AddColumn(table_name, column, schema=schema))
def drop_column(self, table_name, column, schema=None, **kw):
self._exec(base.DropColumn(table_name, column, schema=schema))
def add_constraint(self, const):
if const._create_rule is None or \
const._create_rule(self):
self._exec(schema.AddConstraint(const))
def drop_constraint(self, const):
self._exec(schema.DropConstraint(const))
def rename_table(self, old_table_name, new_table_name, schema=None):
self._exec(base.RenameTable(old_table_name,
new_table_name, schema=schema))
def create_table(self, table):
if util.sqla_07:
table.dispatch.before_create(table, self.connection,
checkfirst=False,
_ddl_runner=self)
self._exec(schema.CreateTable(table))
if util.sqla_07:
table.dispatch.after_create(table, self.connection,
checkfirst=False,
_ddl_runner=self)
for index in table.indexes:
self._exec(schema.CreateIndex(index))
def drop_table(self, table):
self._exec(schema.DropTable(table))
def create_index(self, index):
self._exec(schema.CreateIndex(index))
def drop_index(self, index):
self._exec(schema.DropIndex(index))
def bulk_insert(self, table, rows, multiinsert=True):
if not isinstance(rows, list):
raise TypeError("List expected")
elif rows and not isinstance(rows[0], dict):
raise TypeError("List of dictionaries expected")
if self.as_sql:
for row in rows:
self._exec(table.insert(inline=True).values(**dict(
(k,
_literal_bindparam(k, v, type_=table.c[k].type)
if not isinstance(v, _literal_bindparam) else v)
for k, v in row.items()
)))
else:
# work around http://www.sqlalchemy.org/trac/ticket/2461
if not hasattr(table, '_autoincrement_column'):
table._autoincrement_column = None
if rows:
if multiinsert:
self._exec(table.insert(inline=True), multiparams=rows)
else:
for row in rows:
self._exec(table.insert(inline=True).values(**row))
def compare_type(self, inspector_column, metadata_column):
conn_type = inspector_column.type
metadata_type = metadata_column.type
metadata_impl = metadata_type.dialect_impl(self.dialect)
# work around SQLAlchemy bug "stale value for type affinity"
# fixed in 0.7.4
metadata_impl.__dict__.pop('_type_affinity', None)
if conn_type._compare_type_affinity(
metadata_impl
):
comparator = _type_comparators.get(conn_type._type_affinity, None)
return comparator and comparator(metadata_type, conn_type)
else:
return True
def compare_server_default(self, inspector_column,
metadata_column,
rendered_metadata_default,
rendered_inspector_default):
return rendered_inspector_default != rendered_metadata_default
def correct_for_autogen_constraints(self, conn_uniques, conn_indexes,
metadata_unique_constraints,
metadata_indexes):
pass
def start_migrations(self):
"""A hook called when :meth:`.EnvironmentContext.run_migrations`
is called.
Implementations can set up per-migration-run state here.
"""
def emit_begin(self):
"""Emit the string ``BEGIN``, or the backend-specific
equivalent, on the current connection context.
This is used in offline mode and typically
via :meth:`.EnvironmentContext.begin_transaction`.
"""
self.static_output("BEGIN" + self.command_terminator)
def emit_commit(self):
"""Emit the string ``COMMIT``, or the backend-specific
equivalent, on the current connection context.
This is used in offline mode and typically
via :meth:`.EnvironmentContext.begin_transaction`.
"""
self.static_output("COMMIT" + self.command_terminator)
class _literal_bindparam(_BindParamClause):
pass
@compiles(_literal_bindparam)
def _render_literal_bindparam(element, compiler, **kw):
return compiler.render_literal_bindparam(element, **kw)
def _string_compare(t1, t2):
return \
t1.length is not None and \
t1.length != t2.length
def _numeric_compare(t1, t2):
return \
(
t1.precision is not None and \
t1.precision != t2.precision
) or \
(
t1.scale is not None and \
t1.scale != t2.scale
)
_type_comparators = {
sqltypes.String:_string_compare,
sqltypes.Numeric:_numeric_compare
}