[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

ld-2.9.1: REQ: better support for relative link script paths



Hi!

For the binutils people: debian-hurd is a list of people who are
creating an easy-to-install version of the complete GNU system.  A lot
of us are using cross-compilers right now, and the following is taken
from a discussion that we were having.

On the Hurd, libc.so (from glibc-pre2.1) is a linker script that looks
like this:

----------------------------------------------------------------------
/* GNU ld script
   Use the shared library, but some functions are only in
   the static library, so try that secondarily.  */
GROUP ( /lib/libc.so.0.2 /lib/libcrt_nonshared.a )
----------------------------------------------------------------------

This has caused problems for people who cross-compile, because even
though the runtime system uses /lib, we have it mounted as /gnu/lib
during compilation.  So, /lib/libc.so.0.2 is 

>>>>> Marcus Brinkmann writes:

 MB> I am just curious: Why was the path hardcoded at all? The old
 MB> version of the file seems to work as well. I include it here for
 MB> your reference.
[...]

The following would *seem* like the right thing:

GROUP ( libc.so.0.2 libcrt_nonshared.a )

but it has a subtle problem.  Most of the time it works, but GNU ld
searches the current working directory *before* the library path, so
if you touch a file called `libc.so.0.2' in the current directory,
everything breaks.  Try it for yourself.

This is why Ulrich changed glibc to use an absolute path.

However, I think that an absolute path is just as bad: you have to
write a wrapper or modify a script every time you put the libraries
into a new directory.

The ideal solution, IMO, would be to change GNU ld so that a certain
notation can specify a file that is only found in the same directory
as the linker script.  The current behaviour (search the CWD, then the
library search path) might be useful sometimes, but I doubt it is what
most people want when they use unqualified filenames in their scripts.

I would greatly appreciate anybody's help with this issue, including
suggestions for notation, or better yet, patches to GNU ld.

Thanks,

-- 
 Gordon Matzigkeit <gord@fig.org> //\ I'm a FIG (http://www.fig.org/)
    Lovers of freedom, unite!     \// I use GNU (http://www.gnu.org/)
[Unfortunately, www.fig.org is broken.  Please stay tuned for details.]


Reply to: