Collecting such map of sets should help to clean up the incremental loading (can remove obsolete unknowns corresponding to functions without total iteration and string-based comparison) and make it easier (and more performant) to restart unknowns related to certain syntactic program features.
For example, a single program global variable g has many different unknowns related to it since #473:

• From base privatization (depending on chosen privatization).
• Protecting lockets from mutex analysis.
• Accesses from access analysis.

TODO

• Program variables.
• Program functions.
• Program nodes.