Re: Possible bug -- how to trace the dead-lock?

[Ole Tange]

On Mon, Dec 20, 2010 at 8:34 AM, Maciej Pilichowski
<pilichowski.maciej@gmail.com> wrote:

>> To have people on the email list help you, you need to provide them
>> with an example that reproduces this behaviour.

> However, I spotted dead lock again -- one of the files was not transferred
> to the remote computer, and parallel was stuck at such point (transferring
> it).

I have tried searching for this and I may be onto it. I have the
feeling ssh may sometimes get confused if it gets a tty (but not all
the time - which will explain the random behaviour you describe).

Can you try replacing the line (found in sub init_run_jobs around line
625 in parallel):

    $Global::tty_taken = 0;

with:

    $Global::tty_taken = 1;

This will cause interactive programs to fail:

       cat filelist | parallel -uXj1 emacs (** this will not work with
the above change **)

but should not cause other issues.

I suggest making a test that fails with the original version (e.g. if
parallel only deadlocks 1 out of 1000 then run it 10000 times) and see
if the modified version can survive if run 10 times that amount.


/Ole