1 |
Tonko Mulder <tonko.mulder@×××××.com> posted |
2 |
1227247224.6358.3.camel@Gaius, excerpted below, on Fri, 21 Nov 2008 |
3 |
07:00:24 +0100: |
4 |
|
5 |
> I was wondering since flash has been released for x86_64 and simply |
6 |
> can't find a program on my laptop which uses emul-linux-x86-* If I could |
7 |
> just move to a no-multilib profile. |
8 |
> |
9 |
> As I've read, life get's easier with the no-multilib profile. :) |
10 |
|
11 |
The big question is whether you need anything 32-bit only, which is |
12 |
pretty much proprietary stuff. I'm on the no-multilib profile, but the |
13 |
question was easy for me since I don't do proprietaryware. |
14 |
|
15 |
In addition to flash, which should now cease to be a 32/64-bit issue, |
16 |
there's also various codecs, etc. While there's 64-bit options for many |
17 |
codecs (I've personally had better luck with xine-lib and its many USE |
18 |
flags than mplayer, YMMV) including normal wmv, DRMed formats (including |
19 |
DRMed wmv) and real can be problems. Specifically for xine-lib, the |
20 |
real, vidix and win32codecs USE flags are profile-masked because they are |
21 |
32-bit only. IIRC there's a win64codecs package but I'm not sure of its |
22 |
status |
23 |
|
24 |
The other big problem would be proprietary games. If you prioritize |
25 |
being able to play them very highly, you'll almost certainly want to stay |
26 |
multilib. |
27 |
|
28 |
HOWEVER, it's worth noting that there *IS* an alternative 32-bit solution |
29 |
to multilib. It's more work, but in some ways it's truer to Gentoo's |
30 |
roots. This is the 32-bit chroot solution sometimes mentioned. With |
31 |
multilib, glibc and gcc (the big ones, also sandbox and binutils) are |
32 |
compiled for both 64-bit and 32-bit support, but pretty much everything |
33 |
else is installed from precompiled binaries. This includes all the emul- |
34 |
linux-x86-* libraries and the various *-bin executables (mozilla-firefox- |
35 |
bin, mplayer-bin, etc) packages. With a 32-bit chroot solution (for |
36 |
which there's Gentoo documentation if interested), you start from a |
37 |
standard x86 (32-bit) stage tarball and do much of a 32-bit installation |
38 |
as if from scratch, so after you're done you have almost an entire 32-bit |
39 |
installation paralleling your 64-bit installation (but missing system |
40 |
services like syslog, etc, since the 64-bit main system side provides |
41 |
those services, all compiled from sources, and you maintain them both as |
42 |
you normally would. As I said, rather more work, but it's certainly a |
43 |
valid option, and really, truer to the Gentoo norm of the user |
44 |
controlling as much as possible (while still automated) of the installed |
45 |
software, including the CFLAGS used to compile it and the dependencies |
46 |
(USE flags) it builds against. Multilib, while much less work, is both |
47 |
more complex to implement and much less under installed system sysadmin |
48 |
control. |
49 |
|
50 |
The main point in bringing up the 32-bit chroot option here, however, is |
51 |
to note that rejecting multilib doesn't /necessarily/ mean rejecting 32- |
52 |
bit on the system. Rather, multilib is the middle ground between no |
53 |
32-bit support (no-multilib profile without the 32-bit chroot) and the |
54 |
full 32-bit chroot option with all the trimmings. |
55 |
|
56 |
Meanwhile, no-multilib has some serious benefits, regardless of whether |
57 |
you do the 32-bit chroot or not. Multilib being as it is, a mainly 64- |
58 |
bit system with 32-bit add-ons arranged to allow 32-bit to work as well |
59 |
as possible under the circumstances, it involves a number of compromises |
60 |
and is quite complex, complicating the system considerably. Regardless |
61 |
of whether one is keeping the 32-bit and 64-bit sides more cleanly |
62 |
separated by choosing a full 32-bit chroot on the one hand, or has |
63 |
decided they can do without 32-bit in general on the other, the result of |
64 |
choosing the no-multilib profile is a much cleaner layout, because the 64- |
65 |
bit side doesn't have to worry about 32-bit at all (with the single |
66 |
exception of the kernel's 32-bit emulation option). No double-compiling |
67 |
both the 32-bit and 64-bit parts of gcc or glibc in the same package |
68 |
means broken 32-bit can't block the successful completion and |
69 |
installation of the 64-bit package. From my experience, that alone is |
70 |
the biggest problem people have with those packages on multilib profiles, |
71 |
and a gcc or glibc that can't be upgraded is a HUGE problem that can be a |
72 |
lot of hassle to solve, so eliminating the single biggest cause of that |
73 |
is a BIG deal! No ld having both 32-bit and 64-bit library search paths |
74 |
to worry about and maintain, etc. Less complexity and removing the |
75 |
single biggest cause of broken gccs translates directly to less hassle |
76 |
maintaining your installation. =:^) |
77 |
|
78 |
As always, it's your system, so the decision is ultimately yours. |
79 |
There's serious tradeoffs, benefits and drawbacks to either multilib or |
80 |
nomultilib, 32-bit chroot or not, and no single solution is going to be |
81 |
best for everybody, but given a reasonable understanding of the benefits |
82 |
and drawbacks of each, most Gentoo/amd64 users should have little problem |
83 |
deciding what's best for their particular situation, based on just how |
84 |
much 32-bit they use and whether they prefer the less complex but more |
85 |
powerful 32-bit chroot or the less hassle multilib, should they decide |
86 |
they need to keep 32-bit support. Most but not all 32-bit-only programs |
87 |
are available only as binary-only proprietaryware, so how much of that |
88 |
you may use plays the biggest part in whether you need 32-bit or not, but |
89 |
there's still the occasional open source program that hasn't been ported |
90 |
yet, and of course for some usage, in-house-only apps that may not have |
91 |
been ported, and these may play some albeit increasingly small part in |
92 |
the consideration as well. |
93 |
|
94 |
-- |
95 |
Duncan - List replies preferred. No HTML msgs. |
96 |
"Every nonfree program has a lord, a master -- |
97 |
and if you use the program, he is your master." Richard Stallman |