Re: [PATCH][gdb/testsuite] Reimplement gdb.gdb/python-interrupts.exp as unittest

[Tom de Vries via Gdb-patches <gdb-patches@sourceware.org>]

On 9/11/21 2:02 PM, Tom de Vries wrote:
> +#if GDB_SELF_TEST
> +namespace selftests {
> +extern void (*hook_set_active_ext_lang) (void);
> +void (*hook_set_active_ext_lang) (void) = nullptr;
> +}
> +#endif
> +
>  /* Set the currently active extension language to NOW_ACTIVE.
>     The result is a pointer to a malloc'd block of memory to pass to
>     restore_active_ext_lang.
> @@ -708,6 +715,11 @@ install_gdb_sigint_handler (struct signal_handler *previous)
>  struct active_ext_lang_state *
>  set_active_ext_lang (const struct extension_language_defn *now_active)
>  {
> +#if GDB_SELF_TEST
> +  if (selftests::hook_set_active_ext_lang)
> +    selftests::hook_set_active_ext_lang ();
> +#endif

In particular, I'd like feedback on whether this is acceptable.

F.i., I could do instead the less intrusive (but also less flexible):
...
#if GDB_SELF_TEST
namespace selftests {
extern int hook_raise_at_entry_active_ext_lang;
int hook_raise_at_entry_active_ext_lang = 0;
#endif

struct active_ext_lang_state *
set_active_ext_lang (const struct extension_language_defn *now_active)
{
#if GDB_SELF_TEST
  if (selftests::hook_raise_at_entry_active_ext_lang)
    raise (hook_raise_at_entry_active_ext_lang);
#endif
...

The hook that is a function pointer is the most flexible, but possibly
less safe.  OTOH, it is in the selftest code, which we disable by
default in release branches, so by default this is not included in
releases.  OTOH, it's still possible to enable it in releases using
--enable-unit-tests. Perhaps this could be handled using a
DEVEL_GDB_SELF_TEST, which is disabled in release branches independent
of --enable-unit-tests?

Any comment appreciated.

Thanks,
- Tom