PTS 2023: The Lyon Amendment (minimum perl) (2/5)
Ten years ago, the “toolchain gang” who manage the libraries most central to deploying and testing CPAN libraries came to an agreement on the minimum supported version of perl. Nobody needed to keep maintaining v5.6, and everyone who hadn’t already, moved to v5.8. That was ten years ago, but the toolchain was still pinned to v5.8. I am not enthusiastic about targeting v5.8, and have written about this before. I bumped some modules to v5.12, earlier this year. I got some feedback, both positive and negative, but I felt good about it.
This year, a number of people wanted to talk about bumping the maximum target version for toolchain. To me, the interesting question wasn’t about the toolchain per se, but about what version of perl people feel they can rely upon in CPAN without being a pain to others. I didn’t have strong feelings, coming into this conversation, but I thought it would be nice if the version number moved.
There were a few positions voiced in these conversations:
- Supporting v5.8 should continue, in part because any change in required
version may inconvenience people, and in part because it makes bisecting
changes in how code behaves more difficult (you can’t, for example, test a
library on v5.8 if it
configure_requires
v5.22). - We should provide an alternative CPAN index or repository that will only index code known to work with older versions of Perl. If you use that version of CPAN, you won’t ever get the newer Test-Deep, but you can get the last one that worked on v5.8.
- We should agree to a newer minimum, both for the convenience of the toolchain authors in coding, but also in debugging: it takes real time to work around issues in perl that were fixed 12 years ago.
- The new version shouldn’t be a one-time bump, but a policy about what version we support over time.
The short version here is that the last two positions won out, and many people were interested in the “alternative index” idea. For my part, I saw many complications in making something reliable and long-lived, so my goal was to not couple the idea of a new version to the existence of this system. As for the first position, basically it was outvoted. I do like that Perl has a long tradition of backward compatibility, but for me, it can only go so far. All the way back to 2003 is too far.
In the end, we wrote an amendment to the Lancaster consensus, stating that the toolchain would track the perl of 10 years ago. That would mean that this year we’d target v5.18 — but we also said we’d let Red Hat v7 reach end of maintenance support before moving past its version of perl, v5.16.
That means v5.16 this year, v5.20 next, then v5.22. To my mind, this is huge progress. I’m not in a rush to go change everything, but I will definitely freely update my code when I touch it. CPAN::Meta::Requirements is likely to get an update soon, requiring at least v5.10, maybe v5.12. Test::Deep may follow on toward newer versions over time.
If you want to see older versions of perl continue to have cpan Some::Library
work (by finding an older version of that library), keep an eye out for news
about projects to make that possible. I don’t really plan to follow that work.