14.9 Ancestry-recoverable hard failure

An ancestry-recoverable hard failure is a type of partially recoverable hard failure. An ancestry-recoverable failure allows a superset of the application behaviors allowed by a library-recoverable hard failure. More precisely, let the “failure object” be the object that detected the ancestry-recoverable hard failure. After an ancestry-recoverable hard failure, the following behaviors are no longer application behaviors:

• Libmarpa mutator and constructor method calls where the object is the failure object, or one of its descendants.

Recall that any use of a behavior that is not an application behavior is an irrecoverable failure.

The application behaviors remaining after a ancestry-recoverable hard failure are the following:

• All non-Libmarpa interfaces, including calls to libraries in the application and in the application environment.
• All Libmarpa static method calls.
• All Libmarpa accessor and destructor method calls.
• All Libmarpa mutator and constructor method calls for time objects that are not the failure object, or one of its descendants.

Note that all Libmarpa destructors remain available after an ancestry-recoverable failure. An application will often want to destroy the failure object and all of its descendants, in order to clear memory of problematic objects.

As an example, users calling marpa_g_precompute() will often want to treat a MARPA_EVENT_EARLEY_ITEM_THRESHOLD event as if it were an ancestry-recoverable hard failure. See marpa_g_precompute().

Library-recoverable failure is a special case of ancestry-recoverable failure. When the failure object is a grammar, ancestry-recoverable failure is synonymous with library-recoverable failure.