Gentoo Archives: gentoo-dev

From: Duncan <1i5t5.duncan@×××.net>
To: gentoo-dev@l.g.o
Subject: [gentoo-dev] Re: Package file name requirement for binary ebuilds
Date: Tue, 18 Oct 2016 05:37:39
Message-Id: pan$147df$2bf6beb9$818fa3da$e771aead@cox.net
In Reply to: Re: [gentoo-dev] Re: Package file name requirement for binary ebuilds by "William L. Thomson Jr."
1 William L. Thomson Jr. posted on Mon, 17 Oct 2016 01:36:33 -0400 as
2 excerpted:
3
4 > On Monday, October 17, 2016 4:37:50 AM EDT Duncan wrote:
5 >> William L. Thomson Jr. posted on Sun, 16 Oct 2016 18:30:44 -0400 as
6 >>
7 >> excerpted:
8 >> > Then how would you test that against non official? You cannot install
9 >> > the same package twice at the same time with different USE flags. You
10 >> > can't even make binaries easily of the same package with different
11 >> > USE flags. The previous binary will get overwritten.
12 >>
13 >> AFAIK, with current portage now you can have multiple binaries of the
14 >> same package, with different USE flags or built with different CFLAGS
15 >> or whatever, tho the feature's off by default.
16 >
17 > Very nice, any chance that also includes generated binaries?
18 >
19 > I make binaries for other systems to merge allot (binpkg). It sucks to
20 > have them clobbered with different USE flags. I always wanted to be able
21 > to keep the various binaries with different USE flags. Like some systems
22 > are hardened others not, so recompile gcc vs having 2.
23 >
24 > Not sure if that is a step in that direction, but it is a cool feature
25 > just the same. Allows for further testing.
26
27 I was talking about binpkg (not actually installing, that is qmerging,
28 multiple different versions of the package). However, now that you asked
29 the question, I can see that what I wrote was somewhat ambiguous.
30
31 Again, I've not followed this extremely closely as real life has been
32 crowding out much of my gentooing recently (I bought a house and moved,
33 was in a hotel temporarily for a few months in the mean time and am still
34 rather bare-bones in the new place), but the general idea is pretty much
35 exactly as you suggest.
36
37 With the appropriate option set, portage will bump a sequence number on
38 the binpkg files when the same version is remerged instead of replacing
39 the old binpkg. There's infrastructure for tracking multiple binpkgs of
40 the same version as well, and portage should be smart enough to match USE
41 flags at least, selecting an existing binpkg that matches where possible,
42 and building a new one to add to the collection when there's not an
43 existing match.
44
45 I'm not likely to use it for that, but I've been going to look into the
46 feature and see if it could help managing -9999 live-vcs packages at some
47 point... when I get the time. The biggest problem with live-vcs packages
48 ATM is the possibility of something breaking and not having a good record
49 of your commit-build history, making bisecting more difficult than it
50 should be. This feature could conceivably help not only with that, but
51 in allowing binpkg remerge of the last good build when something breaks,
52 too, thus shortcutting the actual rebuild when there's not time to bisect
53 and you just want to get a working package back again.
54
55 But that multi-binpkg should now be possible with current ~arch portage
56 is about all I know, ATM, due to lack of time to do more than skim the
57 proposed and ultimately approved commits as they're posted to the portage-
58 dev list (which I follow via gmane, as I do all my lists including this
59 one). I don't even know what to toggle to turn it on as I've not been
60 following that closely.
61
62 But I imagine it could be awhile still until it's in stale for those that
63 don't run ~arch...
64
65 --
66 Duncan - List replies preferred. No HTML msgs.
67 "Every nonfree program has a lord, a master --
68 and if you use the program, he is your master." Richard Stallman