vafeiadis / hahn Goto Github PK

From hahn Require Import Hahn.

Section aa.
  Variables A B C D : Prop.
  
  Hypothesis HH : A /\ B.
  
  Lemma aaa (OO : C \/ D) : False.
  Proof.
    desf.
  Admitted.
End aa.

In the code above, desf doesn't destruct OO. However, if you put clear HH before desf, then it does.

This is not an actual "issue", but rather a question about the design of the library.

I am a bit confused about the relationship between (co)dom_rel and doma/domb.

dom_rel is defined in HahnRelationsBasic.v as follows:

Definition dom_rel := fun x => exists y, r x y.

doma is defined in HahnDom.v as follows:

Definition doma := forall x y (REL: r x y), d x.

Clearly there is a connection between the two.
If we fix relation r then all sets d satisfying doma r d will form partially ordered set (ordered by set inclusion) with dom_rel r as a minimal element.

However, I cannot find any lemma in Hahn that relate these two definitions (please, correct me if I am wrong). Moreover some lemmas about doma duplicate similar lemmas for dom_rel.
For example

Lemma doma_rewrite : doma r d -> r ⊆ ⦗d⦘ ⨾ r.

and

Lemma dom_rel_helper (IN : dom_rel r ⊆₁ d) : r ≡ ⦗d⦘ ⨾ r.

So my question is what doma/domb are useful for?
Are they more convenient to work with in some situations compared to dom_rel/codom_rel?
Why there are no lemmas relating the two?

BTW, just greped in weakestmoToImm and IMM repos.
It seems that doma/domb are not used in weakestmo at all (except two usages related to IMM),
and used rarely in IMM (and I suppose all these usages can be refactored to use dom_rel instead).

Dear Viktor,

I am migrating the promising semantics Coq model (https://github.com/snu-sf/promising-coq) from Coq 8.5 to 8.6. As it depends on hahn, I wonder if you have a plan to migrate hahn to 8.6. I tried myself but I couldn't leave the proof style untouched, and cannot help making the proof longer..

Installation from opam on Coq 8.9.1 fails on building HahnCountable. The problem is due to the seq_app lemma which seems to be added later.
Also, it's unclear to me how opam resolves version requirements because currently opam configuration file states that Coq version should be less than 8.9. However, this check only occurs on attempt to pin hahn to another repository and not on regular opam upgrade. Also, there are commit messages saying that hahn works on 8.10 as well.

It seems that the index given to the LTS_final predicate is incorrect in HahnTrace, line 733.

If I am not mistaken, LTS_final is used to denote the terminal state. However LTS_complete_trace requires state 0 to be terminal. It is not obvious why it should be this way.

In the following code, desc terminates in the proof of ff but doesn't in the proof of ff'.

From hahn Require Import Hahn.

Theorem ff (HH : << AA : True >> /\ << BB : True >>) : False.
Proof.
  desc.
Admitted.

Section aa.
  Variable HH : << AA : True >> /\ << BB : True >>.
  
  Theorem ff' : False.
  Proof.
    desc.
  Admitted.
End aa.

The same problem is presented if we replace desc with its part:

repeat match goal with
  | H : ?p /\ ?q |- _ =>
    let x' := match p with | NW (fun z => _) => fresh z | _ => H end in
    let y' := match q with | NW (fun z => _) => fresh z | _ => fresh H end in
    destruct H as [x' y'];
    match p with | NW _ => red in x' | _ => idtac end;
    match q with | NW _ => red in y' | _ => idtac end
end.

That is, the problem appears since destruct HH doesn't remove HH from hypotheses if HH is a section variable.

Replacing destruct H as [x' y'] with destruct H as [x' y']; clear H in the tactic solves the problem.
I propose to update the des* family of tactics in that way.

It seems some lemmas are incompatible with 8.15.1, or something is wrong on my end.

The problems are in HahnMinPath and HahnWf.

the problem in HahnWf seems to be related to a mismatch between the internal name of the binder (y) and what is shown on screen (y0). It can be fixed like this:

    rename y into y'. (* not necessary, but makes sure that the names stay the same *)
    apply In_split in IN; desf.
    eapply H with (y := l1 ++ l2); ins. (* was y0:= *)
      by rewrite !app_length; simpl; lia.

However it's strange that the names are not the same and could point to the problem being in my installation.