[Bug default/26987] GCC 11 Ada binaries contain unknown tags

["tromey at sourceware dot org" <sourceware-bugzilla@sourceware.org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=26987

--- Comment #20 from Tom Tromey <tromey at sourceware dot org> ---
(In reply to Jakub Jelinek from comment #18)

> Strict reading of the spec says that constant class is
> DW_FORM_data{1,2,4,8,16},
> DW_FORM_{s,u}data and DW_FORM_implicit_const, at least in DWARF5.

I'd rather we not do strict readings of the spec, though,
unless the strict-dwarf setting is enabled.  DWARF extensibility
is useful.  In this case it is just for backward compatibility
of some kind, since the problem is solved in DWARF 5.

The problem with using sdata/udata is that it's unpredictable
when a 128-bit value might be needed.  This means all the
readers will have to handle this at the lowest level.  Using a
block form avoids this, because gdb can decide at certain points
to just drop things it doesn't understand.

Here's the internal test case that shows the issue.
A similar, but slightly different, test case is in gdb, see
gdb/testsuite/gdb.ada/fixed_points/fixed_points.adb.

Anyway save this as "test_case.adb" and then do:
$ gnatmake -g test_case

I see the subrange DIE in the test_case.abd CU.
I used today's gcc trunk.

with System;

procedure Test_Case is

   type Base_Fixed_Point_Type is
     delta 1.0 / 16.0
       range (System.Min_Int / 2) * 1.0 / 16.0 ..
       (System.Max_Int / 2) * 1.0 / 16.0;

     subtype Fixed_Point_Subtype is
       Base_Fixed_Point_Type range -50.0 .. 50.0;

     type New_Fixed_Point_Type is
       new Base_Fixed_Point_Type range -50.0 .. 50.0;

     Base_Object            : Base_Fixed_Point_Type := -50.0;
     Subtype_Object         : Fixed_Point_Subtype := -50.0;
     New_Type_Object        : New_Fixed_Point_Type := -50.0;
begin
   Base_Object := 1.0/16.0; -- BREAK
   Subtype_Object := 1.0/16.0;
end Test_Case;

-- 
You are receiving this mail because:
You are on the CC list for the bug.