multiple perl versions

[Dan Ports]

On Wed, Feb 15, 2012 at 09:21:12PM +0000, Eric Hall wrote:
> 	There are valid reasons to have multiple versions of perl
> installed at the same time - the least of which is testing new
> versions to make sure they work while still being able to build
> and run things against older versions.

I don't doubt that there are some benefits to supporting multiple
simultaneous perl installations. But doing so imposes a substantial
burden on maintainers of ports that use perl, in terms of
maintainability and reliability: they need to be careful about which
version of perl they're depending on and invoking, and non-default
configurations probably aren't tested often.

In fact, what we have now may be the worst possible world: we claim to
support multiple perl versions, and ports have to go to some effort to
specify which version they depend on, but it doesn't even work -- the
p5.x- perl module ports conflict with each other!

To give a few examples of the sort of problems this causes:

  - gnucash depends on p5.12-finance-quote (and its many dependencies).
    But autoconf detects that /opt/local/bin/perl is available, and so
    some of its scripts (e.g. gnc-fq-check) use that. This won't work
    right if perl5 +perl5_14 is installed.

  - even the scripts installed by the perl5.x ports themselves have
    this problem. For example, on my system, `cpan-5.14` invokes perl
    5.12. That seems unexpected.

  - intltool depends on various p5.12 modules. Unfortunately, most
    packages using intltool try to invoke it using /opt/local/bin/perl,
    causing it to fail if that's not perl5.12. See, for example, #32425.
    Unfortunately, this needs to be fixed by setting INTLTOOL_PERL in
    every port that *uses* intltool. Most of the ~200 such ports don't
    do this.

It isn't that any of these problems, individually, are particularly
hard to solve -- but they seem sufficiently widespread that it makes me
question whether supporting multiple perl versions is worth it.

I can't think of any other distribution that tries to do this.

Dan

-- 
Dan R. K. Ports              MIT CSAIL                http://drkp.net/