django1/django/db/backends/oracle/schema.py

166 lines
7.2 KiB
Python

import binascii
import copy
import datetime
import re
from django.db.backends.base.schema import BaseDatabaseSchemaEditor
from django.db.utils import DatabaseError
from django.utils.encoding import force_text
class DatabaseSchemaEditor(BaseDatabaseSchemaEditor):
sql_create_column = "ALTER TABLE %(table)s ADD %(column)s %(definition)s"
sql_alter_column_type = "MODIFY %(column)s %(type)s"
sql_alter_column_null = "MODIFY %(column)s NULL"
sql_alter_column_not_null = "MODIFY %(column)s NOT NULL"
sql_alter_column_default = "MODIFY %(column)s DEFAULT %(default)s"
sql_alter_column_no_default = "MODIFY %(column)s DEFAULT NULL"
sql_delete_column = "ALTER TABLE %(table)s DROP COLUMN %(column)s"
sql_delete_table = "DROP TABLE %(table)s CASCADE CONSTRAINTS"
def quote_value(self, value):
if isinstance(value, (datetime.date, datetime.time, datetime.datetime)):
return "'%s'" % value
elif isinstance(value, str):
return "'%s'" % value.replace("\'", "\'\'")
elif isinstance(value, (bytes, bytearray, memoryview)):
return "'%s'" % force_text(binascii.hexlify(value))
elif isinstance(value, bool):
return "1" if value else "0"
else:
return str(value)
def remove_field(self, model, field):
# If the column is an identity column, drop the identity before
# removing the field.
if self._is_identity_column(model._meta.db_table, field.column):
self._drop_identity(model._meta.db_table, field.column)
super().remove_field(model, field)
def delete_model(self, model):
# Run superclass action
super().delete_model(model)
# Clean up manually created sequence.
self.execute("""
DECLARE
i INTEGER;
BEGIN
SELECT COUNT(1) INTO i FROM USER_SEQUENCES
WHERE SEQUENCE_NAME = '%(sq_name)s';
IF i = 1 THEN
EXECUTE IMMEDIATE 'DROP SEQUENCE "%(sq_name)s"';
END IF;
END;
/""" % {'sq_name': self.connection.ops._get_no_autofield_sequence_name(model._meta.db_table)})
def alter_field(self, model, old_field, new_field, strict=False):
try:
super().alter_field(model, old_field, new_field, strict)
except DatabaseError as e:
description = str(e)
# If we're changing type to an unsupported type we need a
# SQLite-ish workaround
if 'ORA-22858' in description or 'ORA-22859' in description:
self._alter_field_type_workaround(model, old_field, new_field)
# If an identity column is changing to a non-numeric type, drop the
# identity first.
elif 'ORA-30675' in description:
self._drop_identity(model._meta.db_table, old_field.column)
self.alter_field(model, old_field, new_field, strict)
# If a primary key column is changing to an identity column, drop
# the primary key first.
elif 'ORA-30673' in description and old_field.primary_key:
self._delete_primary_key(model, strict=True)
self._alter_field_type_workaround(model, old_field, new_field)
else:
raise
def _alter_field_type_workaround(self, model, old_field, new_field):
"""
Oracle refuses to change from some type to other type.
What we need to do instead is:
- Add a nullable version of the desired field with a temporary name. If
the new column is an auto field, then the temporary column can't be
nullable.
- Update the table to transfer values from old to new
- Drop old column
- Rename the new column and possibly drop the nullable property
"""
# Make a new field that's like the new one but with a temporary
# column name.
new_temp_field = copy.deepcopy(new_field)
new_temp_field.null = (new_field.get_internal_type() not in ('AutoField', 'BigAutoField'))
new_temp_field.column = self._generate_temp_name(new_field.column)
# Add it
self.add_field(model, new_temp_field)
# Explicit data type conversion
# https://docs.oracle.com/database/121/SQLRF/sql_elements002.htm#SQLRF51054
new_value = self.quote_name(old_field.column)
old_type = old_field.db_type(self.connection)
if re.match('^N?CLOB', old_type):
new_value = "TO_CHAR(%s)" % new_value
old_type = 'VARCHAR2'
if re.match('^N?VARCHAR2', old_type):
new_internal_type = new_field.get_internal_type()
if new_internal_type == 'DateField':
new_value = "TO_DATE(%s, 'YYYY-MM-DD')" % new_value
elif new_internal_type == 'DateTimeField':
new_value = "TO_TIMESTAMP(%s, 'YYYY-MM-DD HH24:MI:SS.FF')" % new_value
elif new_internal_type == 'TimeField':
# TimeField are stored as TIMESTAMP with a 1900-01-01 date part.
new_value = "TO_TIMESTAMP(CONCAT('1900-01-01 ', %s), 'YYYY-MM-DD HH24:MI:SS.FF')" % new_value
# Transfer values across
self.execute("UPDATE %s set %s=%s" % (
self.quote_name(model._meta.db_table),
self.quote_name(new_temp_field.column),
new_value,
))
# Drop the old field
self.remove_field(model, old_field)
# Rename and possibly make the new field NOT NULL
super().alter_field(model, new_temp_field, new_field)
def normalize_name(self, name):
"""
Get the properly shortened and uppercased identifier as returned by
quote_name() but without the quotes.
"""
nn = self.quote_name(name)
if nn[0] == '"' and nn[-1] == '"':
nn = nn[1:-1]
return nn
def _generate_temp_name(self, for_name):
"""Generate temporary names for workarounds that need temp columns."""
suffix = hex(hash(for_name)).upper()[1:]
return self.normalize_name(for_name + "_" + suffix)
def prepare_default(self, value):
return self.quote_value(value)
def _field_should_be_indexed(self, model, field):
create_index = super()._field_should_be_indexed(model, field)
db_type = field.db_type(self.connection)
if db_type is not None and db_type.lower() in self.connection._limited_data_types:
return False
return create_index
def _is_identity_column(self, table_name, column_name):
with self.connection.cursor() as cursor:
cursor.execute("""
SELECT
CASE WHEN identity_column = 'YES' THEN 1 ELSE 0 END
FROM user_tab_cols
WHERE table_name = %s AND
column_name = %s
""", [self.normalize_name(table_name), self.normalize_name(column_name)])
row = cursor.fetchone()
return row[0] if row else False
def _drop_identity(self, table_name, column_name):
self.execute('ALTER TABLE %(table)s MODIFY %(column)s DROP IDENTITY' % {
'table': self.quote_name(table_name),
'column': self.quote_name(column_name),
})