Re: RFC: fix PR c++/14999

[Tom Tromey <tromey@redhat.com>]

>>>>> "Joel" == Joel Brobecker <brobecker@adacore.com> writes:

Joel> I looked at the PR, and it seems to me that the problem comes
Joel> from the fact that the ax stack was missing the "reg 7" operation.
Joel> I don't really understand the code well enough to be sure about
Joel> my fix, in particular what the "loc" parameter is about, but
Joel> the attached patch seems to restore the origin behavior while
Joel> still keeping your new testcase happy.

Can you try the appended instead?
I am testing it here as well.

I think calling ax_reg and setting kind==axs_lvalue_register is wrong,
as it may result in a second ax_reg call if require_rvalue is called.

Instead I think we should mimic what dwarf2eval does for the fbreg case,
and only derferences lvalue_register values.

Tom

diff --git a/gdb/dwarf2loc.c b/gdb/dwarf2loc.c
index 2282feb..3688425 100644
--- a/gdb/dwarf2loc.c
+++ b/gdb/dwarf2loc.c
@@ -2878,7 +2878,8 @@ dwarf2_compile_expr_to_ax (struct agent_expr *expr, struct axs_value *loc,
 	    op_ptr = safe_read_sleb128 (op_ptr, op_end, &offset);
 	    dwarf2_compile_expr_to_ax (expr, loc, arch, addr_size, datastart,
 				       datastart + datalen, per_cu);
-	    require_rvalue (expr, loc);
+	    if (loc->kind == axs_lvalue_register)
+	      require_rvalue (expr, loc);
 
 	    if (offset != 0)
 	      {