On Tue, 2008-08-12 at 12:00 -0600, Steve Dibb wrote: > Okay, this is something that I've wondered about for a while, but need > to ask -- what is the best way (do we even have a policy) for using > package.use.mask in profiles? > > A couple of specific questions: > > If I need to mask a use flag because of use flag dependencies that won't > work on a particular arch, do I need to contact the arch teams to modify > their package.use.mask profile? If the answer is yes, I can see that as > a huge blocker since I'd have to wait on the arches to do something > before I can even put an ebuild in the tree. I realize this is a > per-arch question depending on how each one might respond, but a common > consensus would be good. >
What happens now is that the ebuild gets added, keywords get dropped as needed, and whoever added the new ebuild opens a rekeywording bug. > Are there ever any cases where we could just simply put the use flag as > restricted in the global package.use.mask and then unrestrict them in > the profiles ones if, for example, it only worked on one or a few > arches? Or is the best policy always to mask it on each profile? > Personally, I prefer the first. But then, if a package is not going to work someplace, sparc is often one of those places. Down side comes if perhaps we are actually testing the package out of /usr/local/portage or some such, and suddenly the use flag for it comes up masked. > As for a specific example, mplayer's dxr2/dxr3 use flag now pulls in a > dependency (media-video/em8300-libraries) which is only keyworded for > x86, ppc, and amd64. That means I'd have to mask the use flag in alpha, > hppa, ia64, ppc64 and sparc (according to repoman). I could skirt the > issue completely and just run an if statement checking if they are using > any of those three arches, but I'd prefer to do it the right way. And > not piss off any arch teams in the process. > > So I guess my question is, can individual ebuild devs freely edit > package.use.mask files in profiles? > Freely? Of course not. We (the arch developers) need to know about it. :) > Steve > I see what you are after. I don't see a good answer for your specific request that does not usually involve a bug of some sort, asking the arch teams to look at what you have done or what you want to do. There are edge cases, of course. Like, "I've package.use.masked fast-x86 for bigmath-3.3.3 on sparc because it pulls in the fast-x86 package which is a fast math x86 package written in x86 assembler." But we still want to know what you've done and why, although in a case like that, a ChangeLog entry would likely be enough. Speaking for myself and not for all of sparc: If you do what seems best at the time (drop keywords and ask for rekeyword, package.use.mask, use.mask with selective unmasking) and document it, along with just asking on IRC when there is doubt, you won't go far wrong. We might scream at you, but we do that to package developers all the time anyway. Default now seems to be to drop keywords and open bugs requesting rekeywording, and that seems to work fine. But unnecessary in edge cases like the one I made up above (and yes, there are some like that). And if you know ahead of time you have something like this coming up, as I mentioned before, ask on IRC if you think of it before playing with profiles. I didn't answer your question. Mostly, I guess, do what seems right and let us know what you did. The best way to do that is through a bug usually. You might not find us on IRC when you need us, and we probably won't read your mail. :) Sorry for not helping, Regards, Ferris -- Ferris McCormick (P44646, MI) <[EMAIL PROTECTED]> Developer, Gentoo Linux (Devrel, Sparc, Userrel, Trustees)
signature.asc
Description: This is a digitally signed message part
