Re: How much to remove from UPDATING (was: Re: git: ff0c7816db69 - main - Remove UPDATING entries from old branches.)
Date: Mon, 28 Nov 2022 07:38:24 UTC
On Mon, Nov 28, 2022 at 12:20 AM Warner Losh <imp@bsdimp.com> wrote: > > > On Sun, Nov 27, 2022 at 11:34 PM Alexander Leidinger <netchild@freebsd.org> > wrote: > >> Quoting Warner Losh <imp@bsdimp.com> (from Sun, 27 Nov 2022 20:12:08 >> -0700): >> >> >> >> On Sun, Nov 27, 2022 at 7:17 PM Warner Losh <imp@bsdimp.com> wrote: >> >>> >>> >>> On Sun, Nov 27, 2022 at 2:35 PM Alexander Leidinger < >>> netchild@freebsd.org> wrote: >>> >>>> Quoting Warner Losh <imp@bsdimp.com> (from Fri, 25 Nov 2022 09:41:28 >>>> -0700): >>>> >>>> > Please revert this. We keep older updating entries on purpose. You >>>> purged >>>> > way too much. Let's chat about how much to remove in arch@. They are >>>> for >>>> > more than just source updates, so your reasoning is wrong. They are >>>> also >>>> > there for users updating their products which can have a larger leap >>>> in >>>> > time. We've traditionally kept closer to 5-10 years here for that >>>> reason. >>>> >>>> Reverted. >>>> >>>> UPDATING as far back as stable/10 (= 4 major updates) is a little bit >>>> excessive (more than 9 years of development work so far), isn't it? >>>> >>> >>> Yes. It's about one release too old, maybe two. More on one or two in a >>> bit. >>> >>> >>>> I don't get the "more than just src updates" part. If we don't talk >>>> about the source code, isn't src/UPATING not the wrong place to store >>>> it? >>>> >>> >>> More than just 'make buildworld updating' or ''updating a system from >>> src' >>> is what I mean. >>> >>> >>>> In terms of updating products, I understand that updating them every 2 >>>> years may be a little bit expensive/excessive for some vendors, but >>>> taking every UPDATING from every stable branch in-between doesn't look >>>> too much time consuming to me. And compared to the huge amount of >>>> changes between N-2 and N... taking UPDATING from all stable branches >>>> in-beteen is nothing. Nevertheless, 4-5 years I consider OK-ish, >>>> nearly 10 years is ... ugh ... a life-time or two in the computer >>>> world. If we look e.g. at the PlayStation (yes, just one of the >>>> products which has FreeBSD inside, but personally I consider it one of >>>> the more stable ones than some network products which have a shorter >>>> shelf-time than the PS-line from an OS-version-tracking point of >>>> view), there are around 6 years in-between models, and they surely >>>> haven't started developing a month before the release date. >>>> >>> >>> So, let's look at what it's used for to see how much we need. If you >>> look at it that way, you'll see that we're not crazy lagging. >>> >>> >>>> So where do we draw the line for UPDATING, 2 major versions (~4 >>>> years), 3 major versions (~6 years)? ~10 years (~5 major versions) >>>> looks overly excessive to me. That's not something you want to try to >>>> catch up, that's rather a new development than a catch-up >>>> >>> >>> OK. Traditionally we've lagged a major release or two from what's >>> officially supported by the project. Right now the 10.x stuff is >>> definitely >>> too old. The 11.x stuff is borderline (but likely relevant), the 12.x >>> stuff >>> is still quite relevant. >>> >>> We need to look at who is updating. Many people have only recently >>> updated from 11. Almost everybody has updated from 10 by now. Lots >>> of people are using 12 and it's still supported. >>> >>> Most of the folks that have source products with lots of changes have >>> updated to at least 12 as far as I've been able to tell. But many haven't >>> jumped to 13 or current yet. >>> >>> There are many people still updating their VMs from 11. Traditionally, >>> they >>> wait until after 11.x goes unsupported before they update. It's only been >>> unsupported for just over 1 year. In the past, this is where upgrading is >>> hitting full speed (I've received feedback in the past at conferences >>> that >>> people often put it off for up to 18 months)... 10.x has been unsupported >>> for more than 3 years, so historically everybody has moved on. So the >>> >> >> I can't do math.... More than 4 years... >> >> >>> 10.x entries are definitely stale... The 11 entries are on the edge... >>> I'd >>> normally have removed the 10.x entries when 13 was branched, but >>> I was asleep at the wheel this time.... Though looking at the logs, I've >>> been not so great about this. Better at some times, worse at others.... >>> >> >> >>> So in my opinion, 10.x entries should have already been gone. 11.x >>> entries are likely useful enough to keep, but they are waning fast. 12.x >>> entries are likely being used all the time by people upgrading from >>> still-supported >>> releases. We've traditionally weighted towards retention because the >>> cost of retention has been super low. >>> >>> This suggests we delete up to the 11 branch point now, and to the 12 >>> branch point when 14 branches in 6 months or so... >>> >> >> 13.x was branched about 6.5 years ago. When 14 is branched, it will be >> 7 years and we'll removing the to the 12 branch point which will be >> four and half years. This seems like a good range to oscillate between. >> >> >> If I understand it correctly, you want to target a policy of: >> Just before the branch of stable/N we remove old entries from UPDATING >> and keep the data of N-2 branches = deleting the data of N-3. >> >> stable/14 -> keep 13+12 and delete 11. >> >> Basically we both are aligned and think N-2 is on the edge. I don't mind >> to live with this edge... >> >> Do we want to document that somewhere? RE-tasklist? Inside UPDATING (top >> or bottom)? >> >> What about removing the entries of 10? Now or with next branch? I would >> vote to do it now, what's done is done. >> > > I think we should remove entries before the 11 branch now. I'll create a > review with that. > https://reviews.freebsd.org/D37514 has the changes for UPDATING. We likely should document this in the RE-tasklis, with the caveat that the sequence is 'create a new branch, trim UPDATING in -current only' rather than the opposite to the stable/XX branch has the older data. I think I trimmed the right amount. We likely should also have $SOMEBODY review the build / updating / etc instructions at each release so we can keep them up to date as well as technology evolves. Warner