I have a script that builds a perl and then uses CPANPLUS to install 200-odd CPAN tarballs. Previously I had all that install into a directory under /opt/perl which is where it would finally be installed.

I decided to change and do the installation somewhere else, but soon found @INC wasn't what I needed after moving under /opt/perl. I tried

-Duserelocatableinc

which is supposed incomptible with useshrplib (no longer needed with Darwin for acceptable performance).

@INC then changes as the path to perl does.

However one module, PDF-Report-1.33.tar.gz, now failed "make test" and therefore wasn't installed.

#   Failed test 'use PDF::Report;'
#   at t/00-load.t line 6.
#     Tried to use 'PDF::Report'.
#     Error:  Insecure dependency in eval while running with -T switch
      at .../lib/5.10.1/Math/Complex.pm line 36.
[download]

as that test runs under -T from Math::Complex

    my $nvsize = $Config{nvsize} ||
                ($Config{uselongdouble} && $Config{longdblsize}) ||
                 $Config{doublesize};
    die "Math::Complex: Could not figure out nvsize\n"
        unless defined $nvsize;
    die "Math::Complex: Cannot not figure out max nv (nvsize = $nvsize
+)\n"
        unless defined $DBL_MAX{$nvsize};
    my $DBL_MAX = eval $DBL_MAX{$nvsize};
[download]

The final part of my perl build script checks modules by trying a one line script "use $mname" for each module listed in 02packages.details.txt against each tarball installed. Changing that to run under '-T' didn't result in any other module having a problem. That's slightly different than building and testing all modules under '-T'.

There's still a reasonable chance I'm doing something wrong, but is this sort of problem expected from 'relocatableinc' ?

Fortunately (?) very little of our code uses taint-mode so I could just remove the '-T' in that test or do the installation manually, or maybe create a symlink under /opt/perl.

I'm really some wisdom on the best approach I should use rather than just whether @INC is tainted.

Danny Thomas

This 5.10.1 was under RHEL5, but don't expect different results under FreeBSD and Darwin.