Re: package system: questions, suggestions, alien

Ville> On Thu, 2002-05-23 at 00:24, Nix wrote: [Alec Schroeder, I think] Emacs is sufficiently full-featured that they probably don't know what they're missing. XEmacs users, on the other hand, regularly post "the package system is really helpful! if only it would ...". I think most Unix people download the sumos for initial install; use pui for upgrades later. This probably is not true for netinstaller customers, since they can select packages from the same UI as they select XEmacs binaries. But I think most of them use PUI not netinstaller for upgrades. Ville> ...and if I've understood correctly, the Sumos shouldn't Ville> really exist (no references available as of now, but I know Ville> I've seen some comments against them somewhere). True, but possibly not in the sense you mean. The main complaint against the SUMO is that we don't distribute new releases of XEmacs as xemacs-21.4.8+sumo-20011219+mule-sumo-20011219.tar.gz. Ville> I think that a big Sumo tarball gives Ville> users/admins/distro-makers/those-who-don't-use-pui a strong Ville> hint like "These packages have been tested to interoperate Ville> well, install this and don't worry about the dependencies." Ville> Well, at least stronger than N different packages in the Ville> same dir, even though the dir is for stable packages. Ville> IMHO, this kind of limits the power of the package system. This is possible, but I've seen no evidence (in the form of questions on xemacs-beta or c.e.x) to indicate that it is true. On the contrary, Adrian posts (often 3 or 4 times a day) "Update to net-utils 1.43" and people Just Do It, nobody ever asks how. The basic use is _there_, it does what typical users want it to do: give them Emacs capabilities they need (or we ask them to install!) without muss or fuss. The limits to the power of the package system from my point of view are: o no help for build from source people -- we should support them better o the pui internals were architected by Motie watchmakers, no rhyme or reason but some how they work o Red Hat (SuSE, et al) hoses us every so often with gratuitous changes to their FTP clients, and we're stuck with an annoying EFS bootstrap problem; Mike Sperber's conservatism about updating EFS doesn't help with this Ville> I also think that a more comprehensive and more enforced Ville> dependency system could have a positive effect on this; Ville> still assuming that Sumos are not considered a good thing. Ville> Versioned dependencies wouldn't hurt here, I guess... Actually, I think the only noticable effects of versioned dependencies would be slightly fewer bug reports to package maintainers, and massively increased complaints about the package system not doing its job correctly. I think it's worth worrying about dependency issues for the future, we may get to the point where dependencies recurse as much as ten layers deep as in Debian. But right now we rarely see recursion deeper than 2, and most packages are chock-full of compatibility cruft---it will take massive effort to increase that depth by moving compatibility and sub-package functionality to separate packages. Eg consider Gnus, which has mml, message, the various backends, etc all split out into nominal subpackages. Does anybody use them? No. Could we really split them into separate packages? I doubt it. On the other hand, consider JDE, which is the only substantial user of eieio and semantic, and thus we distribute whatever version JDE requires. -- Institute of Policy and Planning Sciences http://turnbull.sk.tsukuba.ac.jp University of Tsukuba Tennodai 1-1-1 Tsukuba 305-8573 JAPAN My nostalgia for Icon makes me forget about any of the bad things. I don't have much nostalgia for Perl, so its faults I remember. Scott Gilbert c.l.py