MI floating varobj of invalid value

Presently, floating varobjs are reparsed and reevaluated at
each stop. Reevaluation
can change the value, or can change type, or even can result
in error, in which case
varobj will be marked out of scope. The only transition that
is not supposed
now is from invalid state into valid state.

Suppose I've created varobj for watching for expression
'foo' and after some
debugging ended the debug session. When I start debug again,
I still want to
see the value of 'foo'. However, if frontend tries to create
varobj for foo
at the first stop, and 'foo' is not in scope, then varobj
will not be created.
This means that the frontend has to try to create varobj on
each stop, until
that succeeds. This sounds unnecessary -- we can make
creation of floating
varobj always succeed, and then frontend can use -var-update
to monitor changes
in varobj state. 

Does making creation of floating varobj always succeed seem
a good idea?

- Volodya

> From: Vladimir Prus <vladimircodesourcery.com>
> Date: Sun, 27 Apr 2008 01:27:48 +0400
> 
> However, if frontend tries to create varobj for foo
> at the first stop, and 'foo' is not in scope, then
varobj will not be created.

Why is that?  Since a floating varobj's expression is
re-parsed every
time, you should be able to delay parsing and not fail
creation
because some of its terms are not in scope, right?

> Does making creation of floating varobj always succeed
seem a good idea?

Yes, I think so.

> From: Vladimir Prus <vladimircodesourcery.com>
> Date: Sun, 27 Apr 2008 01:27:48 +0400
> 
> However, if frontend tries to create varobj for foo
> at the first stop, and 'foo' is not in scope, then
varobj will not be created.

Why is that?  Since a floating varobj's expression is
re-parsed every
time, you should be able to delay parsing and not fail
creation
because some of its terms are not in scope, right?

> Does making creation of floating varobj always succeed
seem a good idea?

Yes, I think so.

> From: gdb-ownersourceware.org on behalf of Vladimir Prus
> Sent: Sat 4/26/2008 5:27 PM
> 
> Does making creation of floating varobj always succeed
seem a good idea?

Sounds good to me too.