Re: MI-related testsuite regressions

[Bob Rossi <bob at brasko dot net>]

>  > >  Currently the default MI interpreter *is*
>  > > mi2, so these tests are just duplicated.  It would seem sensible, and save
>  > > effort, to have just one file (mi-stack.exp) until the MI version is bumped
>  > > up one number.  Then mi-stack.exp could be renamed mi2-stack.exp and a new
>  > > file mi-stack.exp could be created.
>  > > 
>  > > In reality GDB doesn't support more than one version of MI (the current
>  > > one).  As has been shown on the mailing list recently, even MI output
>  > > from GDB 6.3 differs from MI output from GDB in CVS.
>  > 
>  > mi2 is a released protocol; the intent is that we not make
>  > backwards-incompatible changes to mi2-*.exp, at least not without
>  > paying close attention to them.  -i=mi sets mi_version to 3 today. 
>  > Until we're ready to declare mi3 usable, we need to continue caring
>  > about mi2.
> 
> -i=mi sets mi_version to 2 in my copy:
> 
>   interp_add (interp_new (INTERP_MI, NULL, mi_out_new (2), &procs));
> 
> and
> 
>   if (current_interp_named_p (INTERP_MI1))
>     deprecated_command_loop_hook = mi1_command_loop;
>   else if (current_interp_named_p (INTERP_MI2))
>     deprecated_command_loop_hook = mi2_command_loop;
>   else if (current_interp_named_p (INTERP_MI3))
>     deprecated_command_loop_hook = mi3_command_loop;
>   else
>     deprecated_command_loop_hook = mi2_command_loop;
> 
> 
> As far as I can see mi3 does nothing that mi2 doesn't do.

I think the ideology is that mi3 is allowed to change in such a way that
mi2 is not. So, all new features go into mi3, while mi2 should remain
stable and unchanged, except for bug fixes and minor new improvements
that are backwards compatible.

Bob Rossi