Re: [RFA] checkpoint / restart, and multi-fork debugging

[Michael Snyder <msnyder@redhat.com>]

Tom Tromey wrote:
>>>>>>"Michael" == Michael Snyder <msnyder@redhat.com> writes:
> 
> 
> Michael> OK, ready for submission.
> Michael> This patch adds two new functionalities for linux native debugging:
> Michael>    * Multiple fork debugging, and
> Michael>    * Checkpoint / Restart.
> 
> I know nothing about the patch of course, but this looks like an
> excellent feature.  One thing that amazes me about it is that it is
> conceptually simple but I don't recall ever hearing it as an idea
> until recently.

It's an idea that Stan Shebs and I have been kicking around
for years.  By the way, Stan is the one who thought of using
fork to capture the state -- I've been very remiss about giving
him credit for that.

> One thing I didn't see mentioned in the docs is the effect on gdb of a
> 'restart'.  I assume that new breakpoints, watchpoints, etc, are kept
> instead of reverting to the state at the checkpoint.  Maybe this is
> worth mentioning.

Yes, thanks -- I'll work it in.  You're correct, checkpoint only
preserves the inferior's state, not the debugger's.  All breakpoints
and watchpoints are shared between checkpoints and/or forks.

> A nice problem this solves is handling watchpoints when the system has
> address space randomization.  I run into this all the time when
> debugging -- back before this feature was added to the OS I would put
> a fair amount of effort into finding some address to watch in one
> debug session, then restart the inferior with a watchpoint set.
> Randomization made this impossible; but with this patch I could just
> make a checkpoint at 'main'.

True that.