| 1 |
On Fri, 2008-11-28 at 09:05 +0100, Michael Haubenwallner wrote: |
| 2 |
> On Thu, 2008-11-27 at 18:43 +0000, Alan Hourihane wrote: |
| 3 |
> > On Thu, 2008-11-27 at 13:56 +0000, Alan Hourihane wrote: |
| 4 |
> > > On Thu, 2008-11-27 at 10:21 +0000, Alan Hourihane wrote: |
| 5 |
> > > > Let me try the ".irrelevant" flag and see what breaks though. |
| 6 |
> > > |
| 7 |
> > > O.k. let's go for this then, and I can bump into issues as they come, as |
| 8 |
> > > it obviously works for the readline case. |
| 9 |
> > |
| 10 |
> > Argh. Nope libperl breaks and creates libperl.irrelevant. |
| 11 |
> |
| 12 |
> IMO this is the best thing that can happen on your platform, because |
| 13 |
> sys-devel/libperl only is useful as a *shared* library (with 'soname'), |
| 14 |
> because the static libperl.a is provided by dev-lang/perl (according to |
| 15 |
> the comments in sys-devel/libperl.ebuild). |
| 16 |
> |
| 17 |
> When sys-devel/libperl provides the static libperl.a, you would likely |
| 18 |
> have a collision in dev-lang/perl lateron. |
| 19 |
|
| 20 |
It doesn't here. sys-devel/libperl provides the static, and |
| 21 |
dev-lang/perl never produces a lib. |
| 22 |
|
| 23 |
> Eventually, sys-devel/libperl should generally be package.masked for |
| 24 |
> FreeMiNT (as 'irrelevant', lol), or at least install nothing. |
| 25 |
|
| 26 |
I did try that a while ago, but it threw up lots of horrible problems |
| 27 |
with DEPEND/RDEPEND. So what's the correct way to package.masked this to |
| 28 |
install nothing ?? |
| 29 |
|
| 30 |
> NOTE: The whole tree as well as the dependency resolving within portage |
| 31 |
> is designed for shared libraries: When there is a bug in one library, |
| 32 |
> only this one package gets rebuilt with the fix, and binaries using this |
| 33 |
> shared library will use the fixed one when restarted. |
| 34 |
> There is no mechanism to rebuild packages which have the broken library |
| 35 |
> linked in statically, so they still use the broken code. |
| 36 |
|
| 37 |
I realize this, and use Gentoo on my x86, mips, alpha, ppc & m68k-linux |
| 38 |
boxes too. |
| 39 |
|
| 40 |
But, any distro will have this problem, and it's just something we've |
| 41 |
had to live with on FreeMiNT. So I don't feel Gentoo should be unable to |
| 42 |
support this configuration. emerge is powerful enough to tell us when |
| 43 |
things change, so I think it's liveable with in a static world. |
| 44 |
|
| 45 |
> > So I'm back to fixing readline, as I'm loathe to go fixing everything |
| 46 |
> > this will break. |
| 47 |
> |
| 48 |
> It's not that I just don't want to change readline to *-aix*, instead I |
| 49 |
> still feel having get_libname() on FreeMiNT returning '.a' simply is |
| 50 |
> wrong - and doing it right as early as possible is better than doing it |
| 51 |
> fast. |
| 52 |
|
| 53 |
Just tried emerging pam, and cracklib has the same "mv" problem. So I |
| 54 |
guess it's inevitable I'm going to have to spend a bit more time on this |
| 55 |
issue. |
| 56 |
|
| 57 |
> > But I still need the patch below applying. |
| 58 |
> > > in gen_usr_ldscript(). |
| 59 |
> |
| 60 |
> This is ok. |
| 61 |
|
| 62 |
Right. |
| 63 |
|
| 64 |
Alan. |
Archives