Done.

Looks good to merge. As always, we also want a change log entry.

I changed the pull request.

I'd prefer to have the different cases consistent, i.e., don't replace the extension with ".pddl" in one case but preserve it in others. I'll add a concrete suggestion to the pull request.

I implemented the suggestion: https://github.com/aibasel/downward/pull/63

Note, however, that the code also assumes that if "domain" is not a prefix but a suffix, then it also ends on .pddl (to be more precise, the suffix has to be "-domain.pddl"). This is totally fine for me, I just wanted to bring this up because this was the argument for not hard-coding [:-5] for task files. So I'm happy to merge if there are no objections or comments.

Agreed to not use [:-5] to split off an extension that may or may not be there. Some IPC domains had filenames like "pfile20" for example, without extension. (At some point, we added ".pddl" unilaterally in our repository.)

I would use os.path.splitext if we currently use os.path and pathlib if we currently use pathlib, but not start mixing in related code. Conversion of more stuff to pathlib is of course also OK.

I agree that this is even cleaner. But then I would suggest to switch the modern pathlib which comes with a method for obtaining a file name without its extension.

I agree, this should be changed.

Why do you prefer basename[:-5] over os.path.splitext?
Yours is obviously shorter, but it implies the knowledge that all relevant files end with ".pddl". The common Python method makes it explicit that we want to split the file extension.

    dirname, basename = os.path.split(task_filename)
    # The Python documentation uses the variables root and ext for the return values
    basename_root, _ = os.path.splitext(basename)

    domain_basenames = [
        "domain.pddl",
        basename_root + "-domain.pddl",
        "domain_" + basename,
        "domain-" + basename,
    ]

I like the suggestion.

These are the current rules for finding a domain file for a given task file in the driver: 

    dirname, basename = os.path.split(task_filename)

    domain_basenames = [
        "domain.pddl",
        basename[:3] + "-domain.pddl",
        "domain_" + basename,
        "domain-" + basename,
    ]

The second rule only works for tasks of the name "xxx.pddl" and the corresponding domain files "xxx-domain.pddl". This works for names like p01 but is, in my opinion, a very arbitrary rule. I would suggest to change this line to

basename[:-5] + "-domain.pddl"

which works for all task files called <anything>.pddl and corresponding domain files <anything>-domain.pddl, which is probably still not a perfect rule, but much better in my opinion. I also checked the downward benchmarks and all files there end on ".pddl". Any objections?