remove_vector and remove_complex break all irep sharing even when the program doesn't use those features

[smowton]

Suggested patch smowton@274e2ea checks whether the features are in use at all before using subtype, Forall_operands and other functions that break sharing. Processing Sakai (a Java program that obviously can't use either feature) this saved at least 20% of goto-analyzer's memory usage (maybe more, it exhausted by available memory before completing).

One caveat: in programs that do make heavy use of complex and vector types and which have highly nested expressions this may be costly due to repeated re-checking of sub-expressions.

[martin-cs]

On Thu, 2016-10-27 at 03:25 -0700, Chris Smowton wrote:

Suggested patch smowton@274e2ea checks whether the features are in use at all before using subtype, Forall_operands and other functions that break sharing. Processing Sakai (a Java program that obviously can't use either feature) this saved at least 20% of goto-analyzer's memory usage (maybe more, it exhausted by available memory before completing).

From what you've said it seems like this would also affect C and C++.

One caveat: in programs that do make heavy use of complex and vector types and which have highly nested expressions this may be costly due to repeated re-checking of sub-expressions.

More generally, we should be performing sharing-aware traversal and
modification of expressions. There was a move to use a visitor for
this; I think the infrastructure should still be there.

[tautschnig]

@smowton We should work out to what extent your proposed patch and #150 do exactly the same thing.

[smowton]

100%, looks like.

[smowton]

Duplicate of #150