1 |
"Mark Knecht" <markknecht@×××××.com> posted |
2 |
5bdc1c8b0803242145le54b0c9we8ba99a79c5c9f9@××××××××××.com, excerpted |
3 |
below, on Mon, 24 Mar 2008 21:45:30 -0700: |
4 |
|
5 |
> Well, my experience is a bit different than yours. I'm sure the way I |
6 |
> maintain my systems, reading between your lines, is almost certainly |
7 |
> more lax than you care for. I do an emerge -DuN system in general about |
8 |
> once a week, but emerge -DuN world isn't going to happen any more often |
9 |
> than once every 6-8 weeks here. I don't have time to deal with the |
10 |
> issues that come up with all this stuff to do it more than maybe 5-6 |
11 |
> times per year. Even that often is pushing it, but it's about what I do. |
12 |
|
13 |
Well, if you note, I didn't push the --deep (= -D). That'll save you |
14 |
some, but at the expense of rather more pain (including a longer revdep- |
15 |
rebuild list since you'll likely have every dependency on the library |
16 |
listed by then, instead of just a couple) when you /do/ eventually update |
17 |
those deep dependencies. |
18 |
|
19 |
I'd say do the system once a week, do the world every couple weeks, |
20 |
always do a revdep-rebuild after the world and keep your --depclean up to |
21 |
date so the revdep-rebuild isn't rebuilding dependencies you don't even |
22 |
need any more, but don't worry too much about --deep, especially on |
23 |
primarily stable systems (like the myth boxes you mention). |
24 |
|
25 |
Also, there's certainly less reason to worry about updates on machines |
26 |
that don't do Internet and/or don't hold private data (as may be the case |
27 |
with some of your myth boxes) anyway. Six weeks or two months is still |
28 |
pushing it and may lead to a lot of pain when you /do/ upgrade such |
29 |
internal boxes, but there's no urgency other than that. |
30 |
|
31 |
> My experience with revdep-rebuild is that it wants to build some old |
32 |
> things, but then these things have been removed from portage and it |
33 |
> cannot, so I have to start looking for solutions. That requires reading |
34 |
> and thinking so it gets put on the back burner. |
35 |
|
36 |
That's really part of the pain of not upgrading frequently enough. As |
37 |
such, it's avoidable. Upgrade before something is removed, and you won't |
38 |
have that problem. Of course, there's a trade-off here between --deep |
39 |
and this pain, as well. If you don't use --deep, you'll encounter this |
40 |
problem relatively more frequently. However, it's a trade-off that's up |
41 |
to you. |
42 |
|
43 |
> The other reason I'll almost NEVER do a real emerge -DuN world is that |
44 |
> we use MythTV here. We have 5 machines that run Myth, either as front |
45 |
> ends or back ends. Unfortunately, with Myth if you update the server you |
46 |
> have to update every machine on the network which causes problems for |
47 |
> the family so I don't do it. |
48 |
|
49 |
You have multiple machines. Are you making use of binary packages? |
50 |
Assuming you use portage, FEATURES=buildpkg on your first upgraded |
51 |
machine, with all machines set to use the same PKGDIR and using emerge |
52 |
--usepkgonly (-K) on everything but the first one (at least where USE |
53 |
flags are the same, and assuming compatible machine and instruction |
54 |
types) should be a real boon to you. |
55 |
|
56 |
IOW, if you are recompiling the package for each machine it's deployed |
57 |
on, you are wasting way more time and energy on that than you are saving |
58 |
by updating so infrequently. If it's possible to do the build once and |
59 |
use it on multiple machines, it's very likely worth doing so, even if it |
60 |
means modifying USE flags and CFLAGS/CXXFLAGS a bit so you /can/ use a |
61 |
common binary. That could just make the upgrades on those myth machines |
62 |
much less trouble for you. |
63 |
|
64 |
Even if my "big" machine was amd64 and the others were x86, as just might |
65 |
be your case, I'd still consider FEATURES=buildpkg, and either decide to |
66 |
run the big machine at 32-bit, or run it 64-bit but with a 32-bit chroot, |
67 |
so it could still could build the binaries for the others. (Of course, |
68 |
for same machine arch, you could also run distcc if desired, but that's |
69 |
getting to be less of an advantage as quad-core multi-gig memory machines |
70 |
work their way into the mainline.) |
71 |
|
72 |
> revdep-rebuild today wanted to rebuild 16 apps. To me that's lots and |
73 |
> lots. Maybe not to you. |
74 |
|
75 |
Well, I'd call that "lots", but not "lots and lots". =8^) "Lots and |
76 |
lots" would to me start at say a couple dozen, so 24, maybe 30. |
77 |
Certainly, a KDE upgrade, typically 90-120 packages on my machine, would |
78 |
qualify for three "lots", tho I'd more likely term it "a hundredish" |
79 |
packages. |
80 |
|
81 |
Interesting how such words mean slightly different things to different |
82 |
people, isn't it? =8^) |
83 |
|
84 |
-- |
85 |
Duncan - List replies preferred. No HTML msgs. |
86 |
"Every nonfree program has a lord, a master -- |
87 |
and if you use the program, he is your master." Richard Stallman |
88 |
|
89 |
-- |
90 |
gentoo-amd64@l.g.o mailing list |