The Coq mailing list

[Théo Zimmermann <theo.zimmi AT gmail.com>]

Hi Jonathan,

This is actually a FAQ. In a previous Coq-Club thread I referenced the
abstract I wrote in 2016 explaining this technique (which wasn't novel
at this time since I was attributing it to an e-mail of Michael
Soegtrop in a previous Coq-Club thread):
https://hal.archives-ouvertes.fr/hal-01671994/document

Théo

Le sam. 9 mai 2020 à 07:03, 
jonikelee AT gmail.com
 
<jonikelee AT gmail.com>
 a écrit :
>
> On Fri, 8 May 2020 21:33:32 -0400
> Jason Gross 
> <jasongross9 AT gmail.com>
>  wrote:
>
> > I believe you can add `shelve` as a hint, and this will allow the
> > search to succeed with a partial solution.  You can then use
> > `unshelve eauto` to get back the left-over goals
>
> That sounds familiar, Jason.  Didn't you once advise me to try this?  I
> recall trying something like "Hint Extern 999 _ => shelve : db".  I
> don't recall it working out that well, though.
>
> Eddy, can you make the rules into rewrites, and use autorewrite with db
> instead?
>
> >
> > On Fri, May 8, 2020, 20:06 Eddy Westbrook 
> > <westbrook AT galois.com>
> > wrote:
> >
> > > Hey coq-club list-ers,
> > >
> > > Is there any way to partially solve a goal using the auto hint
> > > databases? That is, is there a way to do something like "eauto with
> > > <my_db>” but for it to make as much progress as it can and then
> > > leave the remaining sub-goals for me to prove manually?
> > >
> > > The use case here is that I have monadic programs I’m trying to
> > > prove properties of, and I have a set of rules to prove those
> > > properties modulo properties of the values produced in those
> > > computations. So, for instance, I have a monadic program that
> > > manipulates bitvectors, and my rules can reduce (in a rote and
> > > automatable way) all the properties about my monadic program to
> > > properties about bitvectors that I have to prove by hand. The hope
> > > is that I could use the hint databases to define how to apply my
> > > rules, and then I could apply these hints and end up with my goals
> > > about bitvectors to prove by hand. The only other option is to
> > > write a monolithic tactic that pattern-matches on the form of the
> > > goal to apply my rules, and that isn’t as nice and extensible.
> > >
> > > Any ideas?
> > >
> > > Thanks,
> > > -Eddy
>