How to keep vars

[Andre Poenitz]

On Thu, Jun 11, 2009 at 09:35:14PM +0200, Andreas Pakulat wrote:
> > It is surely possible with any debugger as long as the GUI remembers
> > what is open and re-create it as needed.
> 
> Well, the problem here is to know wether an item in the tree still exists
> in the actual application. Which means we'd have to ask gdb for each
> variable on all open subtree's if it still exists.

One can just assume it's gone and delete/recreate it. That's wasteful
but robust, and from what I can tell pretty much the best approach the
current MI implementation "offers". The MI variables are not very well
suited to handle C++ at all. (Who is _really_ interested in what MI
considers a "child" of a std::vector? - not to mention that intermediate
public/ protected/private level that MI spits out).

> And as far as I know the last GDB MI version that KDevelop3 was
> adjusted to use was anything but fast when retrieving tons of
> variables.

The expensive part does not really seem to be the number of commands,
but the number of roundtrips through gdb, i.e. the times where you need
to wait for a result of a response before formulating the next question.
I've never seen a roundtrip below 30ms, even on the fastest machines,
and things start feeling sluggish at ~150ms per step...

> Hence doing that can be really expensive, especially if the
> debugger doesn't tell you when a variable goes out of scope, you have
> to do this refetching on each single step.

Yes, would be nice if MI variables did their job, but at least from
my perspective they don't..

Andre'