[Bug c/97882] [8/9/10/11 Regression] Segmentation Fault on improper redeclaration of function

[jsm28 at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97882

Joseph S. Myers <jsm28 at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2020-11-17
   Target Milestone|---                         |8.5
             Status|UNCONFIRMED                 |NEW
            Summary|Segmentation Fault on       |[8/9/10/11 Regression]
                   |improper redeclaration of   |Segmentation Fault on
                   |function                    |improper redeclaration of
                   |                            |function
     Ever confirmed|0                           |1

--- Comment #1 from Joseph S. Myers <jsm28 at gcc dot gnu.org> ---
Looks like a regression in GCC 7 relative to GCC 6.  Whether this code is valid
or invalid GNU C is a tricky question, but I'm inclined to say that when the
incomplete enum type extension is used, we should *not* count such enums as
compatible with unsigned int since we don't know what members the enum will
have once completed.  So, for example, I think we ought to reject

extern enum foo *x;
unsigned int *x;

and certainly ought to reject

extern enum foo *x;
unsigned int *x;
enum foo { A = -1 };

where the enum ends up compatible with int rather than unsigned int.  (But a
patch involving rejecting such code might not be such a good idea for
backporting, given the risk of breaking user programs that build OK with the
release branch compilers.)

(A C11 defect fix incorporated in C17 means a qualifier on a function return
type is ignored, so whether the code is valid is not affected by the "const" in
the original test case.)