Re: binutils wrapper for aix

On 21-03-2007 15:39:21 -0500, Marshall McMullen wrote:
> I'll ask our loader guys, but what is an rpath
instruction? 

Runtime Path.

In short, when the kernel executes a program, it calls the
runtime
linker to resolve the shared libraries the program needs.  
This runtime
linker has the task to find the libraries.  Of course this
isn't black
magic, so it usually just searches the default path
/usr/lib, /lib.  In
prefix, however, we don't put our libs there, so we need to
instruct the
runtime linker to look in $EPREFIX/usr/lib and $EPREFIX/lib.
 Runtime
path instructions are hardcoded paths in the binary that are
given to
the runtime linker to search there first.

It may not be necessary to do it, if you syste, like Darwin
for instance
"remembers" the location it found the library
during linking time, and
finds the right lib during runtime back like that.

I remember Haubi saying something about this on AIX, but I'm
not sure.
The binutils-config wrapper only injects the
-L$EPREFIX/usr/lib things,
no rpath instructions.  So I don't understand where your
error comes
from.

-- 
Fabian Groffen
Gentoo on a different level

-- 
gentoo-altgentoo.org mailing list

Ah, I understand what you mean now.

Yeah, I don't think that is necessary on AIX.  At
compilation / link
time, the location of the libraries are added into the
binary itself.
So at runtime, it should be able to find it properly.

Haubi, you have any additional insight on this problem I'm
having ?

On 3/21/07, Fabian Groffen <grobiangentoo.org> wrote:
> On 21-03-2007 15:39:21 -0500, Marshall McMullen wrote:
> > I'll ask our loader guys, but what is an rpath
instruction? 
>
> Runtime Path.
>
> In short, when the kernel executes a program, it calls
the runtime
> linker to resolve the shared libraries the program
needs.   This runtime
> linker has the task to find the libraries.  Of course
this isn't black
> magic, so it usually just searches the default path
/usr/lib, /lib.  In
> prefix, however, we don't put our libs there, so we
need to instruct the
> runtime linker to look in $EPREFIX/usr/lib and
$EPREFIX/lib.  Runtime
> path instructions are hardcoded paths in the binary
that are given to
> the runtime linker to search there first.
>
> It may not be necessary to do it, if you syste, like
Darwin for instance
> "remembers" the location it found the library
during linking time, and
> finds the right lib during runtime back like that.
>
> I remember Haubi saying something about this on AIX,
but I'm not sure.
> The binutils-config wrapper only injects the
-L$EPREFIX/usr/lib things,
> no rpath instructions.  So I don't understand where
your error comes
> from.
>
> --
> Fabian Groffen
> Gentoo on a different level
>
> --
> gentoo-altgentoo.org mailing list
>
>


-- 
Marshall McMullen
marshall.mcmullengmail.com
mjmcmullengmail.net
mdmcmullenus.ibm.com

In France, there's a popular saying about Linux that
essentially translates as:
"Linux? You can get a less powerful system, but it will
cost you more."
-- 
gentoo-altgentoo.org mailing list

On 22-03-2007 13:04:19 -0500, Marshall McMullen wrote:
> In case you're curious, I just added the following
lines into the
> bootstrap_gnu() function right before you call econf. 
e.g.:
> 
>    [[ $CHOST == *-aix* ]] || myconf="$
--disable-nls"

This one is no problem.  Stupidly enough it cannot go
unconditional,
because else Solaris breaks.

>        if [[ $ == "wget" ]]; then
>           for patch in `find
> $/usr/portage/net-misc/wget/files/*.patch`;
do
>               patch -p0 < $ || exit 1
>           done
>        fi

Problem here is that it
1) relies on the portage tree
2) needs patch (gpatch? gnupatch? already bootstrapped?)

> Though you could do this only if it is AIX or
something.  Again, this
> is just a hack for my purposes .

If the bootstrap process/toolsbox does patching, it needs to
have/grab
the patches itself IMO.  Second, it needs to make sure the
to use patch
is sane/working/existing.

This does not mean that I want to avoid AIX, but I'd rather
make
pre-patched sources available, hope upstream incorporates
them and
releases, or somehow figure out how to do a "binary
installer" to avoid
bootstrapping in the first place.

-- 
Fabian Groffen
Gentoo on a different level

-- 
gentoo-altgentoo.org mailing list

Yeah, these are all very good points.  Obviously I'm just
hacking this
right now to try to get through the bootstrap process again.
 I think
making sources that are already patched upstream is probably
the
simplest solution....

On 3/22/07, Fabian Groffen <grobiangentoo.org> wrote:
> On 22-03-2007 13:04:19 -0500, Marshall McMullen wrote:
> > In case you're curious, I just added the following
lines into the
> > bootstrap_gnu() function right before you call
econf.  e.g.:
> >
> >    [[ $CHOST == *-aix* ]] ||
myconf="$ --disable-nls"
>
> This one is no problem.  Stupidly enough it cannot go
unconditional,
> because else Solaris breaks.
>
> >        if [[ $ == "wget" ]]; then
> >           for patch in `find
> >
$/usr/portage/net-misc/wget/files/*.patch`; do
> >               patch -p0 < $ || exit 1
> >           done
> >        fi
>
> Problem here is that it
> 1) relies on the portage tree
> 2) needs patch (gpatch? gnupatch? already
bootstrapped?)
>
> > Though you could do this only if it is AIX or
something.  Again, this
> > is just a hack for my purposes .
>
> If the bootstrap process/toolsbox does patching, it
needs to have/grab
> the patches itself IMO.  Second, it needs to make sure
the to use patch
> is sane/working/existing.
>
> This does not mean that I want to avoid AIX, but I'd
rather make
> pre-patched sources available, hope upstream
incorporates them and
> releases, or somehow figure out how to do a
"binary installer" to avoid
> bootstrapping in the first place.
>
> --
> Fabian Groffen
> Gentoo on a different level
>
> --
> gentoo-altgentoo.org mailing list
>
>


-- 
Marshall McMullen
marshall.mcmullengmail.com
mjmcmullengmail.net
mdmcmullenus.ibm.com

In France, there's a popular saying about Linux that
essentially translates as:
"Linux? You can get a less powerful system, but it will
cost you more."
-- 
gentoo-altgentoo.org mailing list