I remember trying to work around some aspect of this tool in a shell tool for a client , and in the end abandoned that, in favour of background tasks and tracking pids, which is kind of astonishing.

I don’t actually remember the problem but possibly something to do with how output is handled.

Almost every time I think I want this tool, I end up with a ad-hoc Python script instead, and Python is objectively speaking a terrible language for multiprocessing tasks.

Huh, I just had the opposite experience. I wanted to parallelize the work of converting a bunch of raw camera images on my laptop to JPGs using ImageMagick, and after looking into what it would take to do it in Ruby I found that `parallel` was the better way to go.

I've used it a bunch of times for some giant batch jobs across hundreds or thousands of cores... And I've found it easier to build a job file of the exact commands and redirect that into it.

I use paralel often and find the syntax reasonable:

    cat args | parallel this-program {}
    parallel this-program {} ::: arg1 arg2 arg3 ...

The only really new thing is ::: which separates the program from the arguments if you want to give them directly.

Can you show what the syntax should be for it to be intuitive to you?