segfault on some java class

[mgudemann]

there's a SIGSEV on some Java classes, bisecting points to merge commit cbmc@509cf25338f9ff8d25bd7262fd96296be378a350 as begin the first bad one, both parent commits do not segfault

[smowton]

@forejtv I don't have rights to assign here, could you assign me please?

[smowton]

The immediate parent of that commit only doesn't segfault because it throws due to a different bug. The cause is 50d161c

[smowton]

At that point (when initialising a pointer-typed field), class_identifier gives the type of the structure whose field we're currently initialising (so if we're initialising a B* that's part of a class A, class_identifier is A). It only becomes B once we've allocated a B and deref'd it (the expr.id()==ID_struct case below).

The given test case has mutually recursive types (class A { B* b; } and class B { A* a; }), and since the recursion set has been effectively disabled it tries to initialise an infinite series of objects.

Since the problem commit belonged to @mgudemann and I don't know what the check was trying to achieve, I'll assign this back to him (@peterschrammel would you mind doing that?)

[tautschnig]

Assignments done; is there a test case that can be used as regression test for any PR fixing this issue?

[mgudemann]

I'll try to distill one from one of the classes that triggered this error.

[mgudemann]

done in #824