Re: pining for the good old days of yore

On 31/03/2014, at 8:30 PM, Jan Wielemaker wrote:
> 
> I'm not so sure.  Typically, you'd like your program to evolve a bit
> more independently
> from your data and you want only some of your data to be persistent.
> Classical
> saved states also provide no safety against failure of Prolog or the
> computer it
> runs on.  Even if you wanted it, it is more or less not doable on modern
> systems ...
> That indeed, is a pity :-(

Classical saved states in DEC-10 Prolog
 - only worked on the same kind of machine
 - did not work in the presence of Address Space Layout Randomisation
 - could not be read by human beings.

A real classical saved state includes a snapshot of the
state of the running code, so that when you restored
one, it resumed execution where it was when it was saved.

That's actually a *wonderful* thing to have (think
checkpoint/restart) but it has major semantic problems
if open files have changed (which didn't actually work)
and especially with things like open network connections
(which DEC-10 Prolog did not support).

The first major blow against classical saved states was
the Apollo DOMAIN operating system, where you could not
rely on a program being loaded at the same address twice
running.  These days this is done to protect against
malware; Apollo had other reasons back in the early 80s.

Changes to the underlying Prolog system could also be a
problem.

Saving stuff out as readable text has a lot of advantages.
I once developed what I fondly hoped might become a
"portable Prolog term binary format", only to discover
that plain text with minimal spaces was nearly as good.
Never underestimate the power and debuggability of plain text.