This improves type safety and prepares the way for being able to access the location number as a location number (rather than an opaque string) without having to check the types whenever it is accessed.
This will provide a better basis for other extensions that will be required to dynamic object in SSS and be useful when doing other operations on dynamic objects as part of Marius' later work.

I'm all in for the encapsulation part of this patch, but how is the renaming set_instance -> set_location_number justified?

All previous usage of this field for stuffing any other information in has been removed and replaced with better typed storage. It now exclusively holds the location number of the malloc that creates this DO. A more descriptive name is hence justified.

It now exclusively holds the location number of the malloc that creates this DO. A more descriptive name is hence justified.

Unless I'm misreading the code: pointer-analysis/value_set_fi*.cpp seem to be encoding a different bit of information via bit manipulation?

Worth checking with Daniel, but I don't see why the _fi variants can't assume that location_number is a sufficient identifier?

(the from_index variable is always set from a goto_programt::instructiont::location_number, which are globally unique, so I suspect this was written before that was true, and/or by someone who wasn't sure if it was true)

@smowton Note that these analyses are context-sensitive (as opposed to the plain value_sett), and from_function encodes the function that is the caller (one level of context). The use of instance thus seems more appropriate as is generalises the concept of a location-bound expression. I would thus maintain that renaming is not appropriate.

@kroening, that looks ready to go.