On Tue, Nov 13, 2001 at 12:43:03PM +0100, Goswin Brederlow wrote:
> It looks eigther like a gcc bug or the atexit symbol is defined in a
> strange way that prevents it from resolving in the stripped version of
> the libc.
>
> Can someone with this problem try to run the library reduction on
>
> #include <stdlib.h>
> void bla(void) {}
> int main() {
> atexit(bla);
> return 0;
> }
>
> and the libc with nothing else cluttering the output.
>
> Its the same behaviour the broken libc had on i386.
i just tried both your test and a full powerpc b-f build and this time
i had no problems with mklibs.py. libc6 is 2.2.4-5
we still cannot use mklibs.py until someone fixes the gratuitous
i386ism regarding what ld is called i have to use the following patch
to make it work correctly (well ignore DEBUG_SPAM):
Index: scripts/rootdisk/mklibs.py
===================================================================
RCS file: /cvs/debian-boot/boot-floppies/scripts/rootdisk/mklibs.py,v
retrieving revision 1.4
diff -u -r1.4 mklibs.py
--- scripts/rootdisk/mklibs.py 2001/08/11 06:55:18 1.4
+++ scripts/rootdisk/mklibs.py 2001/11/13 15:14:55
@@ -53,7 +53,7 @@
DEBUG_VERBOSE = 2
DEBUG_SPAM = 3
-debuglevel = DEBUG_NORMAL
+debuglevel = DEBUG_SPAM
def debug(level, *msg):
if debuglevel >= level:
@@ -352,7 +352,7 @@
base_name = so_pattern.match(library).group(1)
# libc needs its soinit.o and sofini.o as well as the pic
if base_name == "libc":
- extra_flags = find_lib("ld-linux.so.2")
+ extra_flags = find_lib("ld.so.1")
extra_pre_obj = "/usr/lib/libc_pic/soinit.o"
extra_post_obj = "/usr/lib/libc_pic/sofini.o"
else:
@@ -396,6 +396,6 @@
os.remove(dest_path + "/" + lib)
# Make the dynamic linker executable
-ld_file = find_lib("ld-linux.so.2")
+ld_file = find_lib("ld.so.1")
ld_file_name = os.path.basename(ld_file)
os.chmod(dest_path + "/" + ld_file_name, 0755)
--
Ethan Benson
http://www.alaska.net/~erbenson/
msg12191/pgp00000.pgp
Description: PGP signature
