tip-org / benchmarks Goto Github PK

This benchmark has been wrongly translated from the "Productive use of failure paper". It should be an OR instead of an AND in the condition.

Hello, I can't see a license anywhere for how to license my contributions. Could you add a LICENSE or COPYING file somewhere please?

For instance list_z_elem.
https://github.com/tip-org/benchmarks/blob/master/benchmarks/tip2015/list_z_elem.smt2

It seems to me that it is a bit inconvenient to use implicit polymorphism (à la ML) instead of explicit type parameters, which I think is already present in ad-hoc ways in SMTlib2 (for array types and bitvectors sizes). The as construct feels a bit strange (nil as (list a)) could be simply written (nil a). Of course I understand this is probably too late to change :)

According to the README the problems in the false/ subdir should all be false propositions for counterexample finding, but there seem to be a few true ones in there as well. In particular the one that surprised me was productive_use_of_failure_app_inj1 as a straightforward translation of it into Coq produced a statement which is trivial to prove by induction on xs. The lemma as stated also seems intuitively true to me: with the first argument fixed, append should be injective. I admit a lack of experience in the SMT tools which TIP normally targets, so perhaps I'm seriously misunderstanding something here. Could someone with more experience on this project care to comment?

The algebraic datatypes don't seem to comply with SMT-LIB 2.6 format. Do you have any plans to make it SMT-LIB compliant to enable checking your benchmarks between tools? The "prove" command seems extra, but is easier to add.

Citing from http://hydra.cryp.to/build/926186/log/raw:

Building tip-lib-0.1...
Preprocessing library tip-lib-0.1...
Setup: The program 'alex' is required but it could not be found

It would be great if a build-tools stanza listing alex could be added to the Cabal file.

The SMTLIB2 specification fixes top-down order of pattern comparison in match terms (see sec. 3.6.1, after Remark 4).
In the generated benchmark however we obtain:

(match x
    ((_ false)
     (ESC true)
     (P true)
     (Q true)
     (R true)))

(taken from benchmark).
This forces pattern matching to follow the first _ clause.

Some problems, e.g. benchmarks/isaplanner/prop_12.smt2, contain the symbol @, which is not declared. I am not very familiar with smtlib2 but I don't think it is a special keyword.

I don't know where discussions about the TIP format itself should go (some specific mailing list? or the tip-org.github.io page repository?), but in the mean time, I'd like to propose a toplevel statement (lemma t) (where t is a term of type Bool). The statement differs from (assert t) in that it must be proved before being used. This seems really useful in the context of inductive theorem proving, which is not cut free. Tools like quickspec could then parse a TIP problem and return a new TIP problem that is similar, except it contains additional lemma statements containing conjectures.

There a lot of duplicates in benchmarks, which probably should be eliminated.
They can be detected with rdfind (rdfind -dryrun true TIP).
Consider the full list of duplicates:
results.txt

Citing from http://hydra.cryp.to/build/926212/nixlog/2/raw:

src/Tip/Pretty/Isabelle.hs:154:34:
    parse error on input ‘-- $\ "::" $\ ppType 0 (exprType e)’