Re: Fix lto-symtab ICE during Ada LTO bootstrap

[Eric Botcazou <ebotcazou at adacore dot com>]

> BTW for the LTO type merging issues one could probably just drop those types
> and all derivations to alias set 0. But indeed rewriting them to pointers
> would be better, especially for ABI compatibility.
> 
> The Ada ICE I get is:
> Continuing.
> +===========================GNAT BUG DETECTED==============================+
> | 6.0.0 20151122 (experimental) (x86_64-pc-linux-gnu) Assert_Failure
> | atree.adb:6776| Error detected at system.ads:107:4                       

It's apparently another bug in the DCE pass.  From:

  <bb 4063>:
  # DEBUG id => e_186
  _11422 = atree__unchecked_access__node4.localalias.3007 (e_186);

  <bb 4064>:
  # DEBUG id => NULL
  # DEBUG n => _11422
  # DEBUG n => NULL
  if (_11422 == 0)
    goto <bb 4097> (<L255>);
  else
    goto <bb 4065>;

in phicprop2, we have in cddce2:

Deleting : if (_11422 == 0)

and then:

  <bb 4094>:
  # DEBUG id => e_186
  _11422 = atree__unchecked_access__node4.localalias.3007 (e_186);

  <bb 4095>:
  # DEBUG id => NULL
  # DEBUG n => _11422
  # DEBUG n => NULL
  # DEBUG id => e_186
  goto <bb 4076>;

which is wrong since it's the only exit of the loop in sem_prag.adb:22878:

                           loop
                              Set_Warnings_Off
                                (E, (Chars (Get_Pragma_Arg (Arg1)) =
                                      Name_Off));

                              --  For OFF case, make entry in warnings off
                              --  pragma table for later processing. But we do
                              --  not do that within an instance, since these
                              --  warnings are about what is needed in the
                              --  template, not an instance of it.

                              if Chars (Get_Pragma_Arg (Arg1)) = Name_Off
                                and then Warn_On_Warnings_Off
                                and then not In_Instance
                              then
                                 Warnings_Off_Pragmas.Append ((N, E, Reason));
                              end if;

                              if Is_Enumeration_Type (E) then
                                 declare
                                    Lit : Entity_Id;
                                 begin
                                    Lit := First_Literal (E);
                                    while Present (Lit) loop
                                       Set_Warnings_Off (Lit);
                                       Next_Literal (Lit);
                                    end loop;
                                 end;
                              end if;

                              exit when No (Homonym (E));
                              E := Homonym (E);
                           end loop;

Note that the change you installed is not exactly the patch you had posted, it 
contains an additional hunk:

@@ -134,7 +137,7 @@ mark_stmt_necessary (gimple *stmt, bool
   gimple_set_plf (stmt, STMT_NECESSARY, true);
   if (add_to_worklist)
     worklist.safe_push (stmt);
-  if (bb_contains_live_stmts && !is_gimple_debug (stmt))
+  if (add_to_worklist && bb_contains_live_stmts && !is_gimple_debug (stmt))
     bitmap_set_bit (bb_contains_live_stmts, gimple_bb (stmt)->index);
 }
 
which is not documented in the installed ChangeLog either.

-- 
Eric Botcazou