Re: PATCH: tests for MI commands

[Nick Roberts <nickrob at snap dot net dot nz>]

 > Nick, I don't fully agree that the mi2- tests serve no useful purpose.
 > The mi2- tests make sure that the MI2 protocol is completly tested (MI2 
 > works no less then what the mi2- tests say they do), and the mi- tests 
 > make sure that the current development MI protocol is being tested. I 
 > think it's important to keep both tests for this reason. Do you still
 > think they should be removed after this explanation?

I think that we should either remove them (my preference) or rewrite parts of
GDB so that the different MI levels act as independent interpreters (which
currently I am certainly not going to offer to do).  If they were independent
then clearly changing the behaviour of the current level, as I have just done,
wouldn't break the tests for level 2.

 > For instance, it's possible that a command works differently between the
 > 2 versions (don't know if this currently even happens).

There is very little difference between -i=mi and -i=mi2.  The format of the
prologue that GDB prints out is one example.  This is a nonsense to pretend
that we are supporting different levels.  MI has been a project in progress
for five years now.  While companies might support backend development to port
GDB to different architectures, presumably frontend development gets less
support.  I think we need to be realistic, supporting one level properly is an
ambitious target.

Nick