(I'm still not clear on why we don't have epoch numbers in pkgsrc.)
I've always been of the opinion that they're not needed - see previous discussions of this on pkgsrc-related lists over the years.
So to 2023's reiteration:
1. By nature, a change in the numbering scheme means that they're uncommon - someone new takes over a software project, wishes to stamp authority/establish credibility/show it's loved/establish a cultural revolution/stroke their own ego, all without alienating existing users - all that means that it's a 1-off event, unlikely to happen again (a fork with a new name would likely ensue if that were the case). For me, that kind of event is best denoted/mirrored/acknowledged in pkgsrc by using a different package name (directory and ${PKGBASE}), rather than convoluting existing package-numbering and versioning with an epoch. This has the nice property of even allowing old and new to co-exist (they did WHAT to the library version numbers?????), whilst migration takes place. No user is forced to year-0 their installation, especially as part of the larger whole. There is also confusion amongst operating systems, as Joerg pointed out a while ago, as to where epoch sorts.
2. You'll need to back-port it into pkg_install, the related perl modules, pkgin, and anything that uses pkgsrc versioning - but also to downlevel checkers, vulnerability checkers, external websites, and some other things outside pkgsrc control. This will take a while to switch over, and any bodges could be deceptive and alarming. Not fatal, but something to consider, we've not done well in this area recently
3. While there are numerous things I'd go back and change in the versioning stuff were we to do it again, I don't find epochs top of the list - regexp parsing, local change indicators, remove some of the 2000-era NetBSD abbreviations and hardcoded names, maybe re-think the alpha to numeric automatic conversion (seems no longer needed, except maybe current openssl), restore the 64-bit integers for each component that were lost really early on.
This conversation happens about once every 5 years, maybe longer, so it's not exactly "a huge fail", even for the SMOP side of this discussion.
Best,
Alistair
PS. All cultural references and snide asserts have been toned down for publication