Sebastian Nozzi wrote: 
2012/11/12 Damien Cassou <damien.cassou@gmail.com>:
  
it's a backup! You probably never need any backup... until you do :-).
    

Exactly! ;-)

  
In a perfect world Pharo would be perfect, never crash and never loose
any data.
    

To be fair, I have not had a VM crash in quite some time. What I do
manage to archieve is to make the image unresponsive (either by
halting where I am not supposed to, or introduce
delays/infinite-recursions in the UI thread, etc.). Then I have to
kill the VM. It is sad having to kill the VM, not being able to rescue
your unsaved work.
I've never had a VM crash on me, but many times I've hung the image in this way (must be a slow learner) and the interrupt key doesn't work.
I notice that when I then kill the VM it brings up a dialog "Quit without saving?"�� An interesting feature at this point might be some additional recovery options, such as...
+ Provide a list of Smalltalk windows and/or running processes and give a chance to kill them.
+ Save to new 'recovery.image' and set some flag in that image that at next startup limits which code starts running, and/or provides an emergency editor prior to executing the normal startup.