Re: [PATCH] DWARF: add DW_AT_count to zero-length arrays

[Omar Sandoval <osandov@osandov.com>]

On Thu, Aug 16, 2018 at 10:27:48PM -0700, Andrew Pinski wrote:
> On Thu, Aug 16, 2018 at 9:29 PM Omar Sandoval <osandov@osandov.com> wrote:
> >
> > Hi,
> >
> > This fixes the issue that it is impossible to distinguish a zero-length array
> > type from a flexible array type given the DWARF produced by GCC (which I
> > reported here [1]). We do so by adding a DW_AT_count attribute with a value of
> > zero only for zero-length arrays (this is what clang does in this case, too).
> >
> > 1: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86985
> >
> > The reproducer from the PR now produces the expected output:
> >
> > $ ~/gcc-build/gcc/xgcc -B ~/gcc-build/gcc -g -c zero_length.c
> > $ ~/gcc-build/gcc/xgcc -B ~/gcc-build/gcc -g -c flexible.c
> > $ gdb -batch -ex 'ptype baz' zero_length.o
> > type = struct {
> >     int foo;
> >     int bar[0];
> > }
> > $ gdb -batch -ex 'ptype baz' flexible.o
> > type = struct {
> >     int foo;
> >     int bar[];
> > }
> >
> > This was bootstrapped and tested on x86_64-pc-linux-gnu.
> >
> > I don't have commit rights (first time contributor), so if this change is okay
> > could it please be applied?
> 
> I don't think is really required.  Zero-sized arrays are a GCC
> extension which was introduced before flexible array types were part
> of C.  They should be interchangable in all places.  Can you give an
> example of where they are not?

See https://gcc.gnu.org/onlinedocs/gcc/Zero-Length.html. All of the
following cases are only valid for zero-length arrays and not flexible
arrays:

sizeof(int [0]);

struct foo {
	int bar[0];
};

struct foo {
	int bar[0];
	int baz;
};

union foo {
	int bar[0];
	int baz;
};

int arr[2][0];

The following is only valid for flexible arrays and not zero-length
arrays:

struct {
	int foo;
	int bar[];
} baz = {1, {2, 3}};

> A comment about the patch below.
> 
> >
> > Thanks!
> >
> > 2018-08-16  Omar Sandoval  <osandov@osandov.com>
> >
> >         * dwarf2out.c (is_c_family): New.
> >         (add_subscript_info): Add DW_AT_count of 0 for C zero-length arrays.
> >
> > diff --git a/gcc/dwarf2out.c b/gcc/dwarf2out.c
> > index 5a74131d332..b638942c156 100644
> > --- a/gcc/dwarf2out.c
> > +++ b/gcc/dwarf2out.c
> > @@ -3671,6 +3671,7 @@ static const char *get_AT_string (dw_die_ref, enum dwarf_attribute);
> >  static int get_AT_flag (dw_die_ref, enum dwarf_attribute);
> >  static unsigned get_AT_unsigned (dw_die_ref, enum dwarf_attribute);
> >  static inline dw_die_ref get_AT_ref (dw_die_ref, enum dwarf_attribute);
> > +static bool is_c_family (void);
> >  static bool is_cxx (void);
> >  static bool is_cxx (const_tree);
> >  static bool is_fortran (void);
> > @@ -5434,6 +5435,21 @@ get_AT_file (dw_die_ref die, enum dwarf_attribute attr_kind)
> >    return a ? AT_file (a) : NULL;
> >  }
> >
> > +/* Return TRUE if the language is C or C++.  */
> > +
> > +static inline bool
> > +is_c_family (void)
> > +{
> > +  unsigned int lang = get_AT_unsigned (comp_unit_die (), DW_AT_language);
> > +
> > +  return (lang == DW_LANG_C || lang == DW_LANG_C89 || lang == DW_LANG_C99
> > +         || lang == DW_LANG_C11 || lang == DW_LANG_C_plus_plus
> > +         || lang == DW_LANG_ObjC_plus_plus || lang == DW_LANG_C_plus_plus_11
> > +         || lang == DW_LANG_C_plus_plus_14);
> > +
> > +
> > +}
> 
> I think you should just "is_cxx () || is_c ()" and factor out the
> C/Objective-C parts (C++ is already done).
> This is will make it easier to maintain so if c++17 or c++20 comes
> along, only one place needs to be changed.

Okay, will do.

Thanks for taking a look!

> > +
> >  /* Return TRUE if the language is C++.  */
> >
> >  static inline bool
> > @@ -20918,12 +20934,24 @@ add_subscript_info (dw_die_ref type_die, tree type, bool collapse_p)
> >                dimension arr(N:*)
> >              Since the debugger is definitely going to need to know N
> >              to produce useful results, go ahead and output the lower
> > -            bound solo, and hope the debugger can cope.  */
> > +            bound solo, and hope the debugger can cope.
> > +
> > +            For C and C++, if upper is NULL, this may be a zero-length array
> > +            or a flexible array; we'd like to be able to distinguish between
> > +            the two.  Set DW_AT_count to 0 for the former.  TYPE_SIZE is NULL
> > +            for the latter.  */
> >
> >           if (!get_AT (subrange_die, DW_AT_lower_bound))
> >             add_bound_info (subrange_die, DW_AT_lower_bound, lower, NULL);
> > -         if (upper && !get_AT (subrange_die, DW_AT_upper_bound))
> > -           add_bound_info (subrange_die, DW_AT_upper_bound, upper, NULL);
> > +         if (!get_AT (subrange_die, DW_AT_upper_bound)
> > +             && !get_AT (subrange_die, DW_AT_count))
> > +           {
> > +             if (upper)
> > +               add_bound_info (subrange_die, DW_AT_upper_bound, upper, NULL);
> > +             else if (is_c_family () && TYPE_SIZE (type))
> > +               add_bound_info (subrange_die, DW_AT_count,
> > +                               build_int_cst (TREE_TYPE (lower), 0), NULL);
> > +           }
> >         }
> >
> >        /* Otherwise we have an array type with an unspecified length.  The