Date: Tue, 19 Apr 2016 09:28:01 -0700 From: Nathan Whitehorn <nwhitehorn@freebsd.org> To: Alfred Perlstein <alfred@freebsd.org>, Lyndon Nerenberg <lyndon@orthanc.ca>, freebsd-pkgbase@freebsd.org, freebsd-current@freebsd.org Subject: Re: [CFT] packaging the base system with pkg(8) Message-ID: <57165C91.7070005@freebsd.org> In-Reply-To: <5715A338.5060009@freebsd.org> References: <20160302235429.GD75641@FreeBSD.org> <57152CE5.5050500@FreeBSD.org> <9D4B9C8B-41D7-42BC-B436-D23EFFF60261@ixsystems.com> <20160418191425.GW1554@FreeBSD.org> <571533B8.6090109@freebsd.org> <20160418194010.GX1554@FreeBSD.org> <57153E80.4080800@FreeBSD.org> <571551AB.4070203@freebsd.org> <5715772A.3070306@freebsd.org> <571588BB.2070803@orthanc.ca> <201604190201.u3J216NQ054020@orthanc.ca> <5715968B.303@orthanc.ca> <5715A338.5060009@freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
Well, this discussion has gone pretty far off of the rails. I am of course happy to make a patch that cuts this down to 10 packages, but that's not something that should be committed without agreement -- which we obviously don't have. It would have been good to have had meaningful discussion of this before. There are basically three workable options: 1. Have fewer packages. This is easy to implement and preserves the integrity of the base system (as well as unified versioning so that a system at some particular patch level will have the same global state). I have not seen any meaningful downside suggested to this so far except for marginally higher load on update servers. 2. Have 755 packages. This makes it harder to version the system and makes the user interface significantly worse (my opinion, but shared by others). This is the easiest to implement since it is already implemented. 3. Have ~10 meta packages that just depend on sets of the 755 packages and hide the internal details. This gives the user experience of (1) with the implementation of (2), and is marginally more complex than either. Other things (the overlapping packages idea, for instance) are way too complex and will just lead to breakage. Can anyone provide an argument against (1) or, alternatively, for (2)? (2) seems to add a lot of complexity for no clear gain and I remain pretty confused about why it was chosen. -Nathan On 04/18/16 20:17, Alfred Perlstein wrote: > Maybe what the "too many packages" folks need to do is write some code > to hide that it's so many packages. > > :) > > I think the rule of two feet should be applied here. > > What we have is people that have worked quite hard to bring us > something that we can easily work with, and on the other hand some > folks that want something they consider even better. Personally I > can't see how having the system less granular is better, since having > it MORE granular is actually harder work. > > Can someone on the "too many packages" campaign here explain to me how > having too fine a granularity stops you from making macro packages > containing packages? > > Because honestly I can't see how having granularity hurts at all when > if someone wanted to make it less granular all they would have to do > is make some meta-packages. > > -Alfred > > On 4/18/16 7:23 PM, Lyndon Nerenberg wrote: >> On 2016-04-18 7:01 PM, Roger Marquis wrote: >>> Can you explain what would be accomplished by testing all or even a >>> fraction of the possible permutations of base package combinations? We >>> don't do that for ports. >> >> The ports tree isn't a mandatory part of the system. And by >> definition it could not be tested that way, since it offers so many >> alternative implementations of specific functionality. >> >>> Other operating systems don't do that for >>> their base packages. >> >> I'm pretty sure Solaris had some fairly hard-core regression tests to >> ensure basic system functionality wouldn't be compromised by >> 'oddball' selections of packages offered up at install time. >> >> > Honestly, some of us are wondering what exactly is >> > behind some of these concerns regarding base packages. >> >> The concern is from all of us UNIX dinosaurs who predate the >> fine-grained packaging environment, which just worked, and who now >> rip our (little remaining) hair out due to unsolvable package >> dependency loops in the Linux machines we are forced to administer in >> order to pay rent. For me, as a sysadmin, I derive a negative >> benefit from this optimization. >> >> I guess what I'm really asking is: where is the peer reviewed >> research that shows this actually improves things for the not-1% of >> FreeBSD users? >> >> --lyndon >> >> P.S. Don't turn this into a pissing match. I really want to know >> how this is of net benefit to everyone. But I don't want hyperbole. >> I have looked at a lot of, e.g., USENIX and ACM, bibliographies and >> papers for justification for this, and I can't find it. It would >> really help (me, at least) if someone could take a moment to point me >> at demonstrable evidence of the benefits of this model. >> _______________________________________________ >> freebsd-current@freebsd.org mailing list >> https://lists.freebsd.org/mailman/listinfo/freebsd-current >> To unsubscribe, send any mail to >> "freebsd-current-unsubscribe@freebsd.org" >> > > _______________________________________________ > freebsd-current@freebsd.org mailing list > https://lists.freebsd.org/mailman/listinfo/freebsd-current > To unsubscribe, send any mail to > "freebsd-current-unsubscribe@freebsd.org" >
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?57165C91.7070005>