concurrent popper on the same thread, inspired by rpgoldman

[DARKMOONlite]

Issue:

running popper in the same process doesn't work due to conflict between the popper instances. to run even using a ProcessPoolThread requires the max_tasks_per_child parameter set to 1 which caused some issues with running Popper for me. since then I've been using @rpgoldman's branch to get around this issue, but this branch is out of date, and is missing out on a lot of modern improvements to popper since the big refactoring.
e.g.

        with ProcessPoolExecutor(max_workers=args.threads,max_tasks_per_child=1) as executor:
            directory_results = executor.map(directory_runner, directories)
            for directory, csv_row in tqdm(directory_results, total=len(directories), desc="Running Popper tasks"):
                csv_path = directory / f"{'maxsynth' if args.maxsynth else 'popper'}_results.csv" if not args.file_name else directory / Path(args.file_name)
                write_results_to_csv(csv_row, csv_path)

Fix

• every time the Tester class is created it creates a unique uuid to create a unique and identifiable popper module.

        self.module_name = 'popper_tester_module_' + str(uuid.uuid4().int)

• all calls to janus functions are then wrapped in the query_once function which loads all relevant facts into that specific module name (the original method is loaded as janus_query_once) this is done to reduce coding footprint, other naming conventions are usable idk.

def query_once(query,inputs={},keep=False,truth_vals=TruthVal.PLAIN_TRUTHVALS,module_name=None):
    """A wrapper function for janus_swi.query_once that formats the query string and passes the module name.
    """
    if module_name is None: # if no module name is provided, just call the original query_once function
        return janus_query_once(query, inputs=inputs, keep=keep, truth_vals=truth_vals)
    return janus_query_once(f"{module_name}:({query.strip().rstrip('.')})", inputs=inputs, keep=keep, truth_vals=truth_vals)

• the test.pl file that is shared between experiments cannot be loaded into the module because it must be shared to all ongoing experiments. to get around this I create a temporary file, fill it with the same contents then pass the temp file to janus.

        # create temporary test_pl_file 
        with tempfile.NamedTemporaryFile(delete=False, suffix=".pl") as tmp:
            tmp.close()
            shutil.copyfile(test_pl_path, tmp.name)
            
            for x in [exs_pl_path, bk_pl_path, tmp.name]:
                if os.name == 'nt': # if on Windows, SWI requires escaped directory separators
                    x = x.replace('\\', '\\\\')
                logger.info(f'Consulting {x}')
                consult(x,module=self.module_name) # this shouldn't be loaded into a module as otherwise other modules cant read these shared files

Result:

the system can now run multiple threads using a ThreadPoolExecutor without background bleed.
@andrewcropper what do you think?

[andrewcropper]

I revised the code a little to make it a bit simpler. I did not know how to edit this PR (or even if it is possible) so I made a new branch:

https://github.com/logic-and-learning-lab/Popper/tree/fix/tester-module-consult

I think it works as you expect. Can you check?

[DARKMOONlite]

sorry, had some pressing home matters to address.
The branch looks good. I've tested them on my system and they seem to work fine.
It should be noted that there is a small performance difference between multiple instances of popper across multiple threads vs multiple processes, so the default should still be a process pool, but now this should allow some more flexibility in how its used.
merged your commit into the pull request.