from django.db import migrations class MigrationOptimizer(object): """ Powers the optimization process, where you provide a list of Operations and you are returned a list of equal or shorter length - operations are merged into one if possible. For example, a CreateModel and an AddField can be optimised into a new CreateModel, and CreateModel and DeleteModel can be optimised into nothing. """ def optimize(self, operations, app_label=None): """ Main optimization entry point. Pass in a list of Operation instances, get out a new list of Operation instances. Unfortunately, due to the scope of the optimisation (two combinable operations might be separated by several hundred others), this can't be done as a peephole optimisation with checks/output implemented on the Operations themselves; instead, the optimizer looks at each individual operation and scans forwards in the list to see if there are any matches, stopping at boundaries - operations which can't be optimized over (RunSQL, operations on the same field/model, etc.) The inner loop is run until the starting list is the same as the result list, and then the result is returned. This means that operation optimization must be stable and always return an equal or shorter list. The app_label argument is optional, but if you pass it you'll get more efficient optimization. """ # Internal tracking variable for test assertions about # of loops self._iterations = 0 while True: result = self.optimize_inner(operations, app_label) self._iterations += 1 if result == operations: return result operations = result def optimize_inner(self, operations, app_label=None): """ Inner optimization loop. """ new_operations = [] for i, operation in enumerate(operations): # Compare it to each operation after it for j, other in enumerate(operations[i+1:]): result = self.reduce(operation, other) if result is not None: # Optimize! Add result, then remaining others, then return new_operations.extend(result) new_operations.extend(operations[i+1:i+1+j]) new_operations.extend(operations[i+j+2:]) return new_operations if not self.can_optimize_through(operation, other, app_label): new_operations.append(operation) break else: new_operations.append(operation) return new_operations #### REDUCTION #### def reduce(self, operation, other): """ Either returns a list of zero, one or two operations, or None, meaning this pair cannot be optimized. """ submethods = [ (migrations.CreateModel, migrations.DeleteModel, self.reduce_model_create_delete), (migrations.AlterModelTable, migrations.DeleteModel, self.reduce_model_alter_delete), (migrations.AlterUniqueTogether, migrations.DeleteModel, self.reduce_model_alter_delete), (migrations.AlterIndexTogether, migrations.DeleteModel, self.reduce_model_alter_delete), ] for ia, ib, om in submethods: if isinstance(operation, ia) and isinstance(other, ib): return om(operation, other) return None def reduce_model_create_delete(self, operation, other): """ Folds a CreateModel and a DeleteModel into nothing. """ if operation.name == other.name: return [] return None def reduce_model_alter_delete(self, operation, other): """ Folds an AlterModelSomething and a DeleteModel into nothing. """ if operation.name == other.name: return [other] return None #### THROUGH CHECKS #### def can_optimize_through(self, operation, other, app_label=None): """ Returns True if it's possible to optimize 'operation' with something the other side of 'other'. This is possible if, for example, they affect different models. """ MODEL_LEVEL_OPERATIONS = ( migrations.CreateModel, migrations.DeleteModel, migrations.AlterModelTable, migrations.AlterUniqueTogether, migrations.AlterIndexTogether, ) # If it's a model level operation, let it through if there's # nothing that looks like a reference to us in 'other'. if isinstance(operation, MODEL_LEVEL_OPERATIONS): if not other.references_model(operation.name, app_label): return True return False