http://sourceware.org/ml/gdb-patches/2014-08/msg00025.html

On Fri, 01 Aug 2014 09:20:19 +0200, Keven Boell wrote:
> I just tried it on Fedora 20 i686.  Applied the patch, you mentioned, on top of
> the Fortran VLA series and executed your dynamic-other-frame test.  Everything
> is working fine here, I cannot reproduce the crash.

I have it reproducible on Fedora 20 i686 with plain
CFLAGS=-g ./configure;make;cd gdb/testsuite;make site.exp;runtest gdb.fortran/dynamic-other-frame.exp

Besides that I have updated the testcase with
	gdb_test_no_output "set print frame-arguments all"
so that there is no longer needed the patch:
	[patch] Display Fortran strings in backtraces
	https://sourceware.org/ml/gdb-patches/2014-07/msg00709.html

The fix below has no regressions for me.  Unfortunately I do not see why you
cannot reproduce it.

Thanks,
Jan

diff --git a/gdb/dwarf2loc.c b/gdb/dwarf2loc.c
--- a/gdb/dwarf2loc.c
+++ b/gdb/dwarf2loc.c
@@ -42,6 +42,7 @@
 #include <algorithm>
 #include <vector>
 #include <unordered_set>
+#include <functional>
 #include "common/underlying.h"
 #include "common/byte-vector.h"
 
@@ -2348,6 +2349,20 @@ dwarf2_evaluate_loc_desc_full (struct type *type, struct frame_info *frame,
   ctx.per_cu = per_cu;
   ctx.obj_address = 0;
 
+frame_id old_frame_id (get_frame_id (deprecated_safe_get_selected_frame ()));
+class RestoreCall {
+private:
+  const std::function<void ()> func;
+public:
+  RestoreCall(std::function<void ()> func_):func(func_) {}
+  ~RestoreCall() { func(); }
+} restore_frame([=]() {
+  frame_info *old_frame (frame_find_by_id (old_frame_id));
+  if (old_frame != NULL)
+    select_frame (old_frame);
+});
+if (frame != NULL) select_frame (frame);
+
   scoped_value_mark free_values;
 
   ctx.gdbarch = get_objfile_arch (objfile);