rjbs forgot what he was saying

not logged in (root) | by date | tagcloud | help | login

Will your perl remain supported -- and what does that mean?

by rjbs, created 2011-01-28 11:11
last modified 2011-01-28 11:21
tagged with: @markup:md journal perl programming

Yesterday, I tweeted this:

Remember, Perl shops: if you're still on 5.8 come April, you're on an unsupported legacy version. Current versions are 5.10.1 and 5.12.3

A few people asked for more details, and in giving them, I said this:

It's more an amalgam of truths than an actual truth.

Right now, the average company using Perl (or Python, or Ruby, etc.) has no support contract for the language. It's free, open source software that comes with no warranty, guarantee, or promises. Of course, everybody knows that this doesn't mean that it's every man for himself. There are a number of volunteers who put in incredible amounts of work to fix bugs, ensure portability, and improve the language itself. The key word above is volunteer, which means, fundamentally, that nobody is under any obligation to fix anything. If you absolutely need work done, you just might have to pay for it. (In my experience, this is exceedingly rare; the lengths to which I have seen the core Perl team to go to fix bugs that don't even affect them directly are both staggering and humbling.)

Still, the Perl team wants people to have the right kind of expectations, and that comes in two parts: the team will investigate and try to fix bugs in recent perls, but it won't promise to spend its valuable time on old versions. After all, the internals of perl change over time, and it takes a significant mental effort to keep various major versions of perl's VM and its implementation fresh in one's mind.

It was recently my great pleasure to perform the routine work required to release perl-5.12.3, which contained the most up-to-date copy of perlpolicy, which contains the promises that the core team will try to keep, regarding perl support. Here are two of the key points:

In April, the release process for 5.14.0 begins, meaning we'll probably have it in April or May, barring strange circumstances. Once 5.14.0 is out, the official support period for 5.10.x will end, and the chances of bugfixes being applied to the maint-5.10 branch will become very slim. The chances of a new 5.10.x release will become tiny. There simply won't be enough interested volunteers to do the work for such an old version. Maybe if there is a big influx of workers who want to support 5.10, the policy will change -- but that seems a pretty unlikely scenario.

Not only will 5.10.x be out of its normal "official support period," but it will be out of its security update period, too. perlhist tells us that 5.10.0 was released in December, 2007 -- already more than three years ago.

So, with spring's 5.14.0 release obsoleting 5.10, why was I talking about 5.8? Well, perl 5.10.1 was released in August, 2009, and I suspect that almost anybody running 5.10 is running 5.10.1 -- it has too many critical bugfixes for me to stomach the idea that there's a lot of 5.10.0 out there. (Let's not burst my bubble, okay? Optimism is all I have.) People who upgraded their perl in 2009 can probably manage to do it again in the next year or two without serious pain. They remember how.

On the other hand, experience on IRC, mailing lists, and the rest of the world has told me that the most common subversions of 5.8 in use are 8, 5, and 4, probably in that order. My gut tells me that 5.8.1 comes next, but I'm a lot less confident. Those releases were in 2006, 2004, 2004, and 2003, respectively. Assuming that these upgrades were done within a year or so of the language release (which isn't a great assumption, but a tolerable one) then there are a lot of places who haven't upgraded their primary programming tool in over five years. Think of all the other technical tasks you last performed five years ago, and you might realize how little you remember how you did it, and what all the little details were that cost you your 80% time overrun.

That's why I try to track current versions whenever possible. It's not that older versions are always a serious liability, it's that it can cost much more to upgrade only very rarely, compared with frequently. It also drives you to have better integration tools, since you integrate frequently and want it to have a low cost. It means that when the next maintenance release comes out, you can easily perform an automated test of your systems, build a new deployment build, and upgrade, all as routine tasks.

So, my advice is this: if you're building your company's software on a doubly-obsolete version of a tool that's still under development, it's time to begin tooling up to stay up to date.