1 |
On 11-01-2008 21:52:08 -0500, Mike Frysinger wrote: |
2 |
> On Tuesday 08 January 2008, Fabian Groffen wrote: |
3 |
> > - sort out the 64-bits targets with their multilib-hell forced upon us |
4 |
> |
5 |
> dont know exactly what you're referring to, but multilib is completely |
6 |
> optional. |
7 |
|
8 |
http://article.gmane.org/gmane.linux.gentoo.alt/3329 |
9 |
|
10 |
In short: gcc inserts 64-bits library paths which causes the linker |
11 |
first to look inside the host dirs, then in my prefix lib dirs, which |
12 |
creates interesting problems, since the runtime linker gets our runpath |
13 |
directions to look in the prefix lib dirs first. Anyway, it makes |
14 |
linking/runtime fail in cases where the host provided libs are |
15 |
incompatible with the prefix provided ones. |
16 |
|
17 |
Added to that that when I implemented the ldwrapper on amd64 (fedora) |
18 |
linux I didn't fully understand the full multilib picture, some |
19 |
decisions I made there now just feel plain wrong, especially given that |
20 |
each distro seems to implement the multilib thing different (Gentoo: |
21 |
/lib = native bits size, Fedora: /lib = 32-bits, Debian ...). |
22 |
I didn't get it fully right in my post above though, because every |
23 |
distro/os has a kernel configured in such a way that for a 64-bits |
24 |
object, the search path points to the 64-bits host-specific lib paths. |
25 |
So it seems that only binutils doesn't want to know about 64-bits |
26 |
host-specific lib paths, and gcc takes actions to compensate that. |
27 |
|
28 |
Thanks. |
29 |
|
30 |
-- |
31 |
Fabian Groffen |
32 |
Gentoo on a different level |
33 |
-- |
34 |
gentoo-dev@l.g.o mailing list |