https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107952

            Bug ID: 107952
           Summary: tree-object-size: inconsistent size for flexible
                    arrays nested in structs
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: siddhesh at gcc dot gnu.org
  Target Milestone: ---

With -fstrict-flex-arrays, there is a clearer definition of what constitutes
flexible array members, controlled by the frontend.  tree-object-size however
doesn't seem to handle this well enough when the flex array is nested.  e.g.:

typedef struct {
  char pad;
  char data[8];
} F2;

typedef struct {
  unsigned pad;
  F2 flex;
} S2;

#define NULL (void *) 0

__SIZE_TYPE__
nested_flexarray (__SIZE_TYPE__ n)
{
  S2 *p = (S2 *) __builtin_malloc (n);

  return __builtin_dynamic_object_size (p->flex.data, 1);
}

__SIZE_TYPE__
nested_flexarray2 (S2 *p)
{
  return __builtin_dynamic_object_size (p->flex.data, 1);
}

The current behaviour is to treat data as a flex array and nested_flexarray
returns the size as if it were a flex array.  nested_flexarray2 however returns
the subscripted size, thinking of it as a fixed array, which should not be the
expected behaviour with -fstrict-flex-arrays=0.  Instead it should return -1
for maximum size and perhaps 8 as minimum size.

If F2 is changed to:

typedef struct {
  char pad;
  char data[0];
} F2;

the current behaviour ends up returning 0 in both cases, which is incorrect. 
Again, it should return the size as if it were a flex array in nested_flexarray
and -1 for maximum size for nested_flexarray2.

I have a patch that does this, but I need to fix up tests that currently expect
older behaviour and wanted to get some consensus before I fixed them up.

Reply via email to