RE: Who uses gdbreplay?

[Michael Snyder <msnyder at specifix dot com>]

On Thu, 2008-06-12 at 15:28 -0400, Marc Khouzam wrote:
> > > This very morning, I was asked if GDB had any kind of 
> > foundation for reverse
> > > debugging on a target.  Is gdbreplay what I am looking for?
> > 
> > Not yet...   ;-)
> 
> A couple of years ago, you had posted a proposal about a gdb interface
> for reverse debugging
> http://sourceware.org/ml/gdb/2006-01/msg00013.html
> 
> I guess this did not make it to GDB?

Not to the main trunk, but it's preserved on a branch.
I'm currently creating a new updated branch, with the
eventual idea of working it into the trunk.

> Am I correct to understand that the efforts for GDB Reverse debugging
> are currently focused on teawater's GDB Record patch, which you guys 
> are helping with?

Not quite... Teawater's excellent efforts are prompting me
to revive this older work, and attempt to merge the two,
preserving the best elements of each.

But I would like to keep this thread focused on gdbreplay
as much as possible.

> To be honnest, I didn't look at the details of that patch (much too low
> level for me).  However, I was wondering if it considers reverse debugging 
> with a target (gdbserver or other stub?)  From what I understand, 
> GDB Record, will record register and memory changes as they are made.  
> For some targets, storing all this may be prohibitive.
> I wondered if the recording is done by GDB on the host instead?

That is in fact how gdbreplay works.  The log is recorded
by gdb on the host side.  Playback is done entirely on the
host side, with no participation by the target (although 
to gdb it looks like the playback is coming from the remote
target over a socket).