The AIX saga continues

[Daniel Sands]

I've been hacking libtool to make it work fully on AIX, and ran across 
another issue while running the checks:
#24 of the new tests fails.  This is because it hardcodes rpaths into 
the libraries but does not build old libraries.  Apparently check 24 
depends on libtool adding the appropriate -R's into the .la file.  This 
only happens if 'test "$hardcode_into_libs" != yes || test 
"$build_old_libs" = yes' evaluates true.  On AIX, hardcode_into_libs=yes 
and build_old_libs=no.

This is fine for AIX since it uses the runpath of loaded libs to load 
their dependencies.  It only becomes a problem if a library of the same 
name is already loaded in the same "domain" and it is not the one you 
wanted.  A simple example of this would be to create libfoo.a in one 
directory, create libfoo.a in another directory with dependence on the 
first libfoo.a, and then link the second libfoo into main.  So main -> 
d2/libfoo.a -> d1/libfoo.a.  This will fail to run because it won't load 
d1/libfoo.a with d2/libfoo.a in memory and so it won't resolve d1's export.

Since I wouldn't advise the above scenario (and extra runpath entries 
wouldn't help this anyway), I'd suggest that test 24 should be modified 
to run only under the conditions where '-R's should be expected in the 
.la files.


Now to find out why test 25 seems to hang forever....