ENHSP-2020 result different from JAR on ENHSP website

[nvitucci]

I've run ENHSP-2020 with a simple domain and problem, and I get a different result than what I get by running the ENHSP-2020 JAR file from the project website. With planutils the problem is deemed to be unsolvable, while with the JAR file it is solved easily.

If nothing is different in the code, it may just be a different default -planner and/or -h(euristic) across the two versions. This may be relevant to @hstairs too.

Domain file:

(define
    (domain maze)
    (:requirements :strips :typing)
    (:types cell - object)
    (:constants goal - cell)
    (:predicates
        (is-at ?x - cell)
        (is-open ?x - cell)
        (connected ?x - cell ?y - cell)
    )
    (:action move
        :parameters (?from - cell ?to - cell)
        :precondition (and
            (is-at ?from)
            (is-open ?to)
            (connected ?from ?to)
        )
        :effect (and
            (is-at ?to)
            (not (is-at ?from))
        )
    )
)

Problem file:

(define
    (problem solve-maze)
    (:domain maze)
    (:objects
        a - cell
        b - cell
        c - cell
        d - cell
    )
    (:init
        (is-open b)
        (is-open c)
        (is-open d)
        (is-open goal)

        (connected a b)
        (connected b c)
        (connected c d)
        (connected b goal)
        (connected d goal)

        (is-at a)
    )
    (:goal
        (is-at goal)
    )
)

[haz]

@hstairs ?

[nvitucci]

Also, I am not sure it is related but enhsp-2020 runs correctly as a command in planutils, while enhsp does not (it fails with a "-o": executable file not found in $PATH error).

[hstairs]

Sorry but i can't easily reproduce what you have because I am on a m1 architecture and had some issue with planutils. Will try to look into this as soon as I manage to run everything under docker...