You are correct about the $^X, I have just uploaded a 1.2.2 to fix.
The eval probably is not needed for test code like this, but added it for good form after banging my head for a while in a similar construct. Actually for the 050_pod.t test which
I expected to be a no-brainer --- just wanted to verify that if you run in Taint mode passing
any tainted data to the code aborts. Slurm::Sshare/Sacctmgr do a similar fork and exec in
the child construct. exec can throw exceptions in Taint mode (e.g. tainted PATH), and if you wrap a higher level function in eval{} and the exec throws an exception, it gets caught in the higher level eval in the child process, and code actually runs twice. A lot of head banging trying to work that one out:) (I think I've actually learned more about Perl in handling edge cases of the testing code than in writing the main code:)
On Tue Mar 08 01:30:50 2016, SREZIC wrote:
Show quoted text> I still see the problem with 1.2.1. Maybe the $^X has to go into the
> first argument in the exec call, that is instead of
>
> eval { exec { $SBALANCE } $^X, $SBALANCE, @args; };
>
> it should be
>
> eval { exec { $^X } $^X, $SBALANCE, @args; };
>
> ?
>
> Also, I don't think the eval {} is doing anything here ... exec cannot
> throw an exception (except if using "use autodie").
>
> On 2016-03-07 18:25:30, PAYERLE wrote:
> > Thanks again for catching my bugs:)
> >
> > Should be fixed in 1.2.1 which just uploaded. Also fixes a hash
> > order
> > randomization bug.
> > Will update ticket tomorrow as cannot see 1.2.1 yet.
> > On Sat Mar 05 12:49:10 2016, SREZIC wrote:
> > > See
http://www.cpantesters.org/cpan/report/e5770622-e2a6-11e5-8efc-
> > > 839bfcd2507e for a sample fail report. The @INC array in the
> > > failing
> > > test lists paths like /usr/local/lib/perl5/5.20/mach which
> > > indicates
> > > that FreeBSD's system perl is used here, but not the perl used to
> > > build and test the distribution.
> > >
> > > Probably $^X has to be used here (or a similar solution).
