On Thu, Apr 6, 2017 at 8:20 AM, Igor Gnatenko <ignatenkobrain@xxxxxxxxxxxxxxxxx> wrote: > -----BEGIN PGP SIGNED MESSAGE----- > Hash: SHA256 > > On Thu, 2017-04-06 at 14:07 +0200, Florian Weimer wrote: >> On 04/06/2017 02:00 PM, Igor Gnatenko wrote: >> > On Thu, 2017-04-06 at 12:40 +0100, Richard W.M. Jones wrote: >> > > What's not explained is why, except that it "causes issues with >> > > the >> > > package updates process". It seems as if the or-rule above would >> > > be >> > > simple enough, so what's the exact problem? >> > None (or almost none) of rel-eng (and probably some other critical >> > tools) are not supporting them. For example, "mash" uses yum which >> > just >> >> Isn't mash dead? > Some people tell me it is, some other says no. And after some time, > someone starts porting it to DNF. So it is hard to say. > > Anyhow, it *is* used right now. It is used in Bodhi now, so it's at least a zombie. >> >> > crashes when finds out rich dependency. Other tools are doing their >> > own >> > depsolving which leads to the wrong results or non-solvable >> > dependencies. >> >> But the tooling does not support weak dependencies, either. Weak >> dependencies *are* allowed, and you can't really install current >> Fedora >> using yum. > Yes, this is also true. Yum just ignores weak deps, but I thought > anaconda/pungi/whatsoever disables weak deps anyway.. So distribution > *MUST* be installable even weak deps are disabled. Anaconda does *not* disable weak deps. As far as I know, neither does Pungi in the new DNF solver mode. All tools currently assume weak deps don't exist, but solve with weak deps enabled. It is completely reasonable to have weak deps enabled, because they are "install if available, don't if not" type dependencies. So I don't want that to change at all. However, tool inputs don't assume weak deps are possible, which is a problem. That's why, for example, excluding weak dep packages on kickstarts doesn't work with livemedia-creator or Anaconda. (It does work in my livecd-tools and I did port the solution over to Anaconda and proposed it as a PR, but like most PRs, it just sits there). >> >> So maybe the tools issues have been addressed? > No, they have not been addressed. =( > > Speaking on behalf of Rust SIG, it is important to support rich > dependencies for "version range" support, but I can't do much without > support from rel-eng. I hope for more people asking for rich deps, because then maybe it can be prioritized... -- 真実はいつも一つ！/ Always, there's only one truth! _______________________________________________ devel mailing list -- devel@xxxxxxxxxxxxxxxxxxxxxxx To unsubscribe send an email to devel-leave@xxxxxxxxxxxxxxxxxxxxxxx
