import datetime import string import random import re import sys from django.core.management.color import no_style from django.db import transaction, models from django.db.backends.util import truncate_name from django.db.models.fields import NOT_PROVIDED from django.dispatch import dispatcher from django.conf import settings from django.utils.datastructures import SortedDict from south.logger import get_logger def alias(attrname): """ Returns a function which calls 'attrname' - for function aliasing. We can't just use foo = bar, as this breaks subclassing. """ def func(self, *args, **kwds): return getattr(self, attrname)(*args, **kwds) return func class DatabaseOperations(object): """ Generic SQL implementation of the DatabaseOperations. Some of this code comes from Django Evolution. """ # We assume the generic DB can handle DDL transactions. MySQL wil change this. has_ddl_transactions = True alter_string_set_type = 'ALTER COLUMN %(column)s TYPE %(type)s' alter_string_set_null = 'ALTER COLUMN %(column)s DROP NOT NULL' alter_string_drop_null = 'ALTER COLUMN %(column)s SET NOT NULL' has_check_constraints = True delete_check_sql = 'ALTER TABLE %(table)s DROP CONSTRAINT %(constraint)s' allows_combined_alters = True add_column_string = 'ALTER TABLE %s ADD COLUMN %s;' delete_unique_sql = "ALTER TABLE %s DROP CONSTRAINT %s" delete_foreign_key_sql = 'ALTER TABLE %(table)s DROP CONSTRAINT %(constraint)s' supports_foreign_keys = True max_index_name_length = 63 drop_index_string = 'DROP INDEX %(index_name)s' delete_column_string = 'ALTER TABLE %s DROP COLUMN %s CASCADE;' create_primary_key_string = "ALTER TABLE %(table)s ADD CONSTRAINT %(constraint)s PRIMARY KEY (%(columns)s)" delete_primary_key_sql = "ALTER TABLE %(table)s DROP CONSTRAINT %(constraint)s" backend_name = None default_schema_name = "public" def __init__(self, db_alias): self.debug = False self.deferred_sql = [] self.dry_run = False self.pending_transactions = 0 self.pending_create_signals = [] self.db_alias = db_alias self._initialised = False def _is_multidb(self): try: from django.db import connections except ImportError: return False else: return True def _get_connection(self): """ Returns a django connection for a given DB Alias """ if self._is_multidb(): from django.db import connections return connections[self.db_alias] else: from django.db import connection return connection def _get_setting(self, setting_name): """ Allows code to get a setting (like, for example, STORAGE_ENGINE) """ setting_name = setting_name.upper() connection = self._get_connection() if self._is_multidb(): # Django 1.2 and above return connection.settings_dict[setting_name] else: # Django 1.1 and below return getattr(settings, "DATABASE_%s" % setting_name) def _has_setting(self, setting_name): """ Existence-checking version of _get_setting. """ try: self._get_setting(setting_name) except (KeyError, AttributeError): return False else: return True def _get_schema_name(self): try: return self._get_setting('schema') except (KeyError, AttributeError): return self.default_schema_name def _possibly_initialise(self): if not self._initialised: self.connection_init() self._initialised = True def connection_init(self): """ Run before any SQL to let database-specific config be sent as a command, e.g. which storage engine (MySQL) or transaction serialisability level. """ pass def quote_name(self, name): """ Uses the database backend to quote the given table/column name. """ return self._get_connection().ops.quote_name(name) def execute(self, sql, params=[]): """ Executes the given SQL statement, with optional parameters. If the instance's debug attribute is True, prints out what it executes. """ self._possibly_initialise() cursor = self._get_connection().cursor() if self.debug: print " = %s" % sql, params get_logger().debug('south execute "%s" with params "%s"' % (sql, params)) if self.dry_run: return [] cursor.execute(sql, params) try: return cursor.fetchall() except: return [] def execute_many(self, sql, regex=r"(?mx) ([^';]* (?:'[^']*'[^';]*)*)", comment_regex=r"(?mx) (?:^\s*$)|(?:--.*$)"): """ Takes a SQL file and executes it as many separate statements. (Some backends, such as Postgres, don't work otherwise.) """ # Be warned: This function is full of dark magic. Make sure you really # know regexes before trying to edit it. # First, strip comments sql = "\n".join([x.strip().replace("%", "%%") for x in re.split(comment_regex, sql) if x.strip()]) # Now execute each statement for st in re.split(regex, sql)[1:][::2]: self.execute(st) def add_deferred_sql(self, sql): """ Add a SQL statement to the deferred list, that won't be executed until this instance's execute_deferred_sql method is run. """ self.deferred_sql.append(sql) def execute_deferred_sql(self): """ Executes all deferred SQL, resetting the deferred_sql list """ for sql in self.deferred_sql: self.execute(sql) self.deferred_sql = [] def clear_deferred_sql(self): """ Resets the deferred_sql list to empty. """ self.deferred_sql = [] def clear_run_data(self, pending_creates = None): """ Resets variables to how they should be before a run. Used for dry runs. If you want, pass in an old panding_creates to reset to. """ self.clear_deferred_sql() self.pending_create_signals = pending_creates or [] def get_pending_creates(self): return self.pending_create_signals def create_table(self, table_name, fields): """ Creates the table 'table_name'. 'fields' is a tuple of fields, each repsented by a 2-part tuple of field name and a django.db.models.fields.Field object """ if len(table_name) > 63: print " ! WARNING: You have a table name longer than 63 characters; this will not fully work on PostgreSQL or MySQL." columns = [ self.column_sql(table_name, field_name, field) for field_name, field in fields ] self.execute('CREATE TABLE %s (%s);' % ( self.quote_name(table_name), ', '.join([col for col in columns if col]), )) add_table = alias('create_table') # Alias for consistency's sake def rename_table(self, old_table_name, table_name): """ Renames the table 'old_table_name' to 'table_name'. """ if old_table_name == table_name: # Short-circuit out. return params = (self.quote_name(old_table_name), self.quote_name(table_name)) self.execute('ALTER TABLE %s RENAME TO %s;' % params) def delete_table(self, table_name, cascade=True): """ Deletes the table 'table_name'. """ params = (self.quote_name(table_name), ) if cascade: self.execute('DROP TABLE %s CASCADE;' % params) else: self.execute('DROP TABLE %s;' % params) drop_table = alias('delete_table') def clear_table(self, table_name): """ Deletes all rows from 'table_name'. """ params = (self.quote_name(table_name), ) self.execute('DELETE FROM %s;' % params) def add_column(self, table_name, name, field, keep_default=True): """ Adds the column 'name' to the table 'table_name'. Uses the 'field' paramater, a django.db.models.fields.Field instance, to generate the necessary sql @param table_name: The name of the table to add the column to @param name: The name of the column to add @param field: The field to use """ sql = self.column_sql(table_name, name, field) if sql: params = ( self.quote_name(table_name), sql, ) sql = self.add_column_string % params self.execute(sql) # Now, drop the default if we need to if not keep_default and field.default is not None: field.default = NOT_PROVIDED self.alter_column(table_name, name, field, explicit_name=False, ignore_constraints=True) def _db_type_for_alter_column(self, field): """ Returns a field's type suitable for ALTER COLUMN. By default it just returns field.db_type(). To be overriden by backend specific subclasses @param field: The field to generate type for """ try: return field.db_type(connection=self._get_connection()) except TypeError: return field.db_type() def _alter_set_defaults(self, field, name, params, sqls): "Subcommand of alter_column that sets default values (overrideable)" # Next, set any default if not field.null and field.has_default(): default = field.get_default() sqls.append(('ALTER COLUMN %s SET DEFAULT %%s ' % (self.quote_name(name),), [default])) else: sqls.append(('ALTER COLUMN %s DROP DEFAULT' % (self.quote_name(name),), [])) def alter_column(self, table_name, name, field, explicit_name=True, ignore_constraints=False): """ Alters the given column name so it will match the given field. Note that conversion between the two by the database must be possible. Will not automatically add _id by default; to have this behavour, pass explicit_name=False. @param table_name: The name of the table to add the column to @param name: The name of the column to alter @param field: The new field definition to use """ if self.dry_run: return # hook for the field to do any resolution prior to it's attributes being queried if hasattr(field, 'south_init'): field.south_init() # Add _id or whatever if we need to field.set_attributes_from_name(name) if not explicit_name: name = field.column else: field.column = name if not ignore_constraints: # Drop all check constraints. TODO: Add the right ones back. if self.has_check_constraints: check_constraints = self._constraints_affecting_columns(table_name, [name], "CHECK") for constraint in check_constraints: self.execute(self.delete_check_sql % { 'table': self.quote_name(table_name), 'constraint': self.quote_name(constraint), }) # Drop all foreign key constraints try: self.delete_foreign_key(table_name, name) except ValueError: # There weren't any pass # First, change the type params = { "column": self.quote_name(name), "type": self._db_type_for_alter_column(field), "table_name": table_name } # SQLs is a list of (SQL, values) pairs. sqls = [] # Only alter the column if it has a type (Geometry ones sometimes don't) if params["type"] is not None: sqls.append((self.alter_string_set_type % params, [])) # Next, nullity if field.null: sqls.append((self.alter_string_set_null % params, [])) else: sqls.append((self.alter_string_drop_null % params, [])) # Next, set any default self._alter_set_defaults(field, name, params, sqls) # Finally, actually change the column if self.allows_combined_alters: sqls, values = zip(*sqls) self.execute( "ALTER TABLE %s %s;" % (self.quote_name(table_name), ", ".join(sqls)), flatten(values), ) else: # Databases like e.g. MySQL don't like more than one alter at once. for sql, values in sqls: self.execute("ALTER TABLE %s %s;" % (self.quote_name(table_name), sql), values) if not ignore_constraints: # Add back FK constraints if needed if field.rel and self.supports_foreign_keys: self.execute( self.foreign_key_sql( table_name, field.column, field.rel.to._meta.db_table, field.rel.to._meta.get_field(field.rel.field_name).column ) ) def _constraints_affecting_columns(self, table_name, columns, type="UNIQUE"): """ Gets the names of the constraints affecting the given columns. If columns is None, returns all constraints of the type on the table. """ if self.dry_run: raise ValueError("Cannot get constraints for columns during a dry run.") if columns is not None: columns = set(columns) if type == "CHECK": ifsc_table = "constraint_column_usage" else: ifsc_table = "key_column_usage" schema = self._get_schema_name() # First, load all constraint->col mappings for this table. rows = self.execute(""" SELECT kc.constraint_name, kc.column_name FROM information_schema.%s AS kc JOIN information_schema.table_constraints AS c ON kc.table_schema = c.table_schema AND kc.table_name = c.table_name AND kc.constraint_name = c.constraint_name WHERE kc.table_schema = %%s AND kc.table_name = %%s AND c.constraint_type = %%s """ % ifsc_table, [schema, table_name, type]) # Load into a dict mapping = {} for constraint, column in rows: mapping.setdefault(constraint, set()) mapping[constraint].add(column) # Find ones affecting these columns for constraint, itscols in mapping.items(): # If columns is None we definitely want this field! (see docstring) if itscols == columns or columns is None: yield constraint def create_unique(self, table_name, columns): """ Creates a UNIQUE constraint on the columns on the given table. """ if not isinstance(columns, (list, tuple)): columns = [columns] name = self.create_index_name(table_name, columns, suffix="_uniq") cols = ", ".join(map(self.quote_name, columns)) self.execute("ALTER TABLE %s ADD CONSTRAINT %s UNIQUE (%s)" % ( self.quote_name(table_name), self.quote_name(name), cols, )) return name def delete_unique(self, table_name, columns): """ Deletes a UNIQUE constraint on precisely the columns on the given table. """ if not isinstance(columns, (list, tuple)): columns = [columns] # Dry runs mean we can't do anything. if self.dry_run: return constraints = list(self._constraints_affecting_columns(table_name, columns)) if not constraints: raise ValueError("Cannot find a UNIQUE constraint on table %s, columns %r" % (table_name, columns)) for constraint in constraints: self.execute(self.delete_unique_sql % ( self.quote_name(table_name), self.quote_name(constraint), )) def column_sql(self, table_name, field_name, field, tablespace='', with_name=True, field_prepared=False): """ Creates the SQL snippet for a column. Used by add_column and add_table. """ # If the field hasn't already been told its attribute name, do so. if not field_prepared: field.set_attributes_from_name(field_name) # hook for the field to do any resolution prior to it's attributes being queried if hasattr(field, 'south_init'): field.south_init() # Possible hook to fiddle with the fields (e.g. defaults & TEXT on MySQL) field = self._field_sanity(field) try: sql = field.db_type(connection=self._get_connection()) except TypeError: sql = field.db_type() if sql: # Some callers, like the sqlite stuff, just want the extended type. if with_name: field_output = [self.quote_name(field.column), sql] else: field_output = [sql] field_output.append('%sNULL' % (not field.null and 'NOT ' or '')) if field.primary_key: field_output.append('PRIMARY KEY') elif field.unique: # Just use UNIQUE (no indexes any more, we have delete_unique) field_output.append('UNIQUE') tablespace = field.db_tablespace or tablespace if tablespace and self._get_connection().features.supports_tablespaces and field.unique: # We must specify the index tablespace inline, because we # won't be generating a CREATE INDEX statement for this field. field_output.append(self._get_connection().ops.tablespace_sql(tablespace, inline=True)) sql = ' '.join(field_output) sqlparams = () # if the field is "NOT NULL" and a default value is provided, create the column with it # this allows the addition of a NOT NULL field to a table with existing rows if not getattr(field, '_suppress_default', False): if field.has_default(): default = field.get_default() # If the default is actually None, don't add a default term if default is not None: # If the default is a callable, then call it! if callable(default): default = default() # Now do some very cheap quoting. TODO: Redesign return values to avoid this. if isinstance(default, basestring): default = "'%s'" % default.replace("'", "''") elif isinstance(default, (datetime.date, datetime.time, datetime.datetime)): default = "'%s'" % default # Escape any % signs in the output (bug #317) if isinstance(default, basestring): default = default.replace("%", "%%") # Add it in sql += " DEFAULT %s" sqlparams = (default) elif (not field.null and field.blank) or (field.get_default() == ''): if field.empty_strings_allowed and self._get_connection().features.interprets_empty_strings_as_nulls: sql += " DEFAULT ''" # Error here would be nice, but doesn't seem to play fair. #else: # raise ValueError("Attempting to add a non null column that isn't character based without an explicit default value.") if field.rel and self.supports_foreign_keys: self.add_deferred_sql( self.foreign_key_sql( table_name, field.column, field.rel.to._meta.db_table, field.rel.to._meta.get_field(field.rel.field_name).column ) ) # Things like the contrib.gis module fields have this in 1.1 and below if hasattr(field, 'post_create_sql'): for stmt in field.post_create_sql(no_style(), table_name): self.add_deferred_sql(stmt) # In 1.2 and above, you have to ask the DatabaseCreation stuff for it. # This also creates normal indexes in 1.1. if hasattr(self._get_connection().creation, "sql_indexes_for_field"): # Make a fake model to pass in, with only db_table model = self.mock_model("FakeModelForGISCreation", table_name) for stmt in self._get_connection().creation.sql_indexes_for_field(model, field, no_style()): self.add_deferred_sql(stmt) if sql: return sql % sqlparams else: return None def _field_sanity(self, field): """ Placeholder for DBMS-specific field alterations (some combos aren't valid, e.g. DEFAULT and TEXT on MySQL) """ return field def foreign_key_sql(self, from_table_name, from_column_name, to_table_name, to_column_name): """ Generates a full SQL statement to add a foreign key constraint """ constraint_name = '%s_refs_%s_%x' % (from_column_name, to_column_name, abs(hash((from_table_name, to_table_name)))) return 'ALTER TABLE %s ADD CONSTRAINT %s FOREIGN KEY (%s) REFERENCES %s (%s)%s;' % ( self.quote_name(from_table_name), self.quote_name(truncate_name(constraint_name, self._get_connection().ops.max_name_length())), self.quote_name(from_column_name), self.quote_name(to_table_name), self.quote_name(to_column_name), self._get_connection().ops.deferrable_sql() # Django knows this ) def delete_foreign_key(self, table_name, column): "Drop a foreign key constraint" if self.dry_run: return # We can't look at the DB to get the constraints constraints = list(self._constraints_affecting_columns(table_name, [column], "FOREIGN KEY")) if not constraints: raise ValueError("Cannot find a FOREIGN KEY constraint on table %s, column %s" % (table_name, column)) for constraint_name in constraints: self.execute(self.delete_foreign_key_sql % { "table": self.quote_name(table_name), "constraint": self.quote_name(constraint_name), }) drop_foreign_key = alias('delete_foreign_key') def create_index_name(self, table_name, column_names, suffix=""): """ Generate a unique name for the index """ table_name = table_name.replace('"', '').replace('.', '_') index_unique_name = '' if len(column_names) > 1: index_unique_name = '_%x' % abs(hash((table_name, ','.join(column_names)))) # If the index name is too long, truncate it index_name = ('%s_%s%s%s' % (table_name, column_names[0], index_unique_name, suffix)).replace('"', '').replace('.', '_') if len(index_name) > self.max_index_name_length: part = ('_%s%s%s' % (column_names[0], index_unique_name, suffix)) index_name = '%s%s' % (table_name[:(self.max_index_name_length-len(part))], part) return index_name def create_index_sql(self, table_name, column_names, unique=False, db_tablespace=''): """ Generates a create index statement on 'table_name' for a list of 'column_names' """ if not column_names: print "No column names supplied on which to create an index" return '' connection = self._get_connection() if db_tablespace and connection.features.supports_tablespaces: tablespace_sql = ' ' + connection.ops.tablespace_sql(db_tablespace) else: tablespace_sql = '' index_name = self.create_index_name(table_name, column_names) return 'CREATE %sINDEX %s ON %s (%s)%s;' % ( unique and 'UNIQUE ' or '', self.quote_name(index_name), self.quote_name(table_name), ','.join([self.quote_name(field) for field in column_names]), tablespace_sql ) def create_index(self, table_name, column_names, unique=False, db_tablespace=''): """ Executes a create index statement """ sql = self.create_index_sql(table_name, column_names, unique, db_tablespace) self.execute(sql) def delete_index(self, table_name, column_names, db_tablespace=''): """ Deletes an index created with create_index. This is possible using only columns due to the deterministic index naming function which relies on column names. """ if isinstance(column_names, (str, unicode)): column_names = [column_names] name = self.create_index_name(table_name, column_names) sql = self.drop_index_string % { "index_name": self.quote_name(name), "table_name": self.quote_name(table_name), } self.execute(sql) drop_index = alias('delete_index') def delete_column(self, table_name, name): """ Deletes the column 'column_name' from the table 'table_name'. """ params = (self.quote_name(table_name), self.quote_name(name)) self.execute(self.delete_column_string % params, []) drop_column = alias('delete_column') def rename_column(self, table_name, old, new): """ Renames the column 'old' from the table 'table_name' to 'new'. """ raise NotImplementedError("rename_column has no generic SQL syntax") def delete_primary_key(self, table_name): """ Drops the old primary key. """ # Dry runs mean we can't do anything. if self.dry_run: return constraints = list(self._constraints_affecting_columns(table_name, None, type="PRIMARY KEY")) if not constraints: raise ValueError("Cannot find a PRIMARY KEY constraint on table %s" % (table_name,)) for constraint in constraints: self.execute(self.delete_primary_key_sql % { "table": self.quote_name(table_name), "constraint": self.quote_name(constraint), }) drop_primary_key = alias('delete_primary_key') def create_primary_key(self, table_name, columns): """ Creates a new primary key on the specified columns. """ if not isinstance(columns, (list, tuple)): columns = [columns] self.execute(self.create_primary_key_string % { "table": self.quote_name(table_name), "constraint": self.quote_name(table_name+"_pkey"), "columns": ", ".join(map(self.quote_name, columns)), }) def start_transaction(self): """ Makes sure the following commands are inside a transaction. Must be followed by a (commit|rollback)_transaction call. """ if self.dry_run: self.pending_transactions += 1 transaction.commit_unless_managed() transaction.enter_transaction_management() transaction.managed(True) def commit_transaction(self): """ Commits the current transaction. Must be preceded by a start_transaction call. """ if self.dry_run: return transaction.commit() transaction.leave_transaction_management() def rollback_transaction(self): """ Rolls back the current transaction. Must be preceded by a start_transaction call. """ if self.dry_run: self.pending_transactions -= 1 transaction.rollback() transaction.leave_transaction_management() def rollback_transactions_dry_run(self): """ Rolls back all pending_transactions during this dry run. """ if not self.dry_run: return while self.pending_transactions > 0: self.rollback_transaction() if transaction.is_dirty(): # Force an exception, if we're still in a dirty transaction. # This means we are missing a COMMIT/ROLLBACK. transaction.leave_transaction_management() def send_create_signal(self, app_label, model_names): self.pending_create_signals.append((app_label, model_names)) def send_pending_create_signals(self, verbosity=0, interactive=False): # Group app_labels together signals = SortedDict() for (app_label, model_names) in self.pending_create_signals: try: signals[app_label].extend(model_names) except KeyError: signals[app_label] = list(model_names) # Send only one signal per app. for (app_label, model_names) in signals.iteritems(): self.really_send_create_signal(app_label, list(set(model_names)), verbosity=verbosity, interactive=interactive) self.pending_create_signals = [] def really_send_create_signal(self, app_label, model_names, verbosity=0, interactive=False): """ Sends a post_syncdb signal for the model specified. If the model is not found (perhaps it's been deleted?), no signal is sent. TODO: The behavior of django.contrib.* apps seems flawed in that they don't respect created_models. Rather, they blindly execute over all models within the app sending the signal. This is a patch we should push Django to make For now, this should work. """ if self.debug: print " - Sending post_syncdb signal for %s: %s" % (app_label, model_names) app = models.get_app(app_label) if not app: return created_models = [] for model_name in model_names: model = models.get_model(app_label, model_name) if model: created_models.append(model) if created_models: if hasattr(dispatcher, "send"): # Older djangos dispatcher.send(signal=models.signals.post_syncdb, sender=app, app=app, created_models=created_models, verbosity=verbosity, interactive=interactive) else: if self._is_multidb(): # Django 1.2+ models.signals.post_syncdb.send( sender=app, app=app, created_models=created_models, verbosity=verbosity, interactive=interactive, db=self.db_alias, ) else: # Django 1.1 - 1.0 models.signals.post_syncdb.send( sender=app, app=app, created_models=created_models, verbosity=verbosity, interactive=interactive, ) def mock_model(self, model_name, db_table, db_tablespace='', pk_field_name='id', pk_field_type=models.AutoField, pk_field_args=[], pk_field_kwargs={}): """ Generates a MockModel class that provides enough information to be used by a foreign key/many-to-many relationship. Migrations should prefer to use these rather than actual models as models could get deleted over time, but these can remain in migration files forever. Depreciated. """ class MockOptions(object): def __init__(self): self.db_table = db_table self.db_tablespace = db_tablespace or settings.DEFAULT_TABLESPACE self.object_name = model_name self.module_name = model_name.lower() if pk_field_type == models.AutoField: pk_field_kwargs['primary_key'] = True self.pk = pk_field_type(*pk_field_args, **pk_field_kwargs) self.pk.set_attributes_from_name(pk_field_name) self.abstract = False def get_field_by_name(self, field_name): # we only care about the pk field return (self.pk, self.model, True, False) def get_field(self, name): # we only care about the pk field return self.pk class MockModel(object): _meta = None # We need to return an actual class object here, not an instance MockModel._meta = MockOptions() MockModel._meta.model = MockModel return MockModel # Single-level flattening of lists def flatten(ls): nl = [] for l in ls: nl += l return nl