Oddity when linking

[Casey Marshall]

Hi.

I've been trying to compile Kaffe CVS and reach a problem when the
native parts of the class library are linked. The end of the compile
looks like this:

---
/bin/sh ../../../libtool --mode=link gcc  -g -O2 -Wall -Wstrict-prototypes   -o 
libnative.la -rpath /opt/kaffe-cvs/jre/lib/i386  -module -release 1.1.x-cvs 
-export-symbols-regex "^([Jj]ava|kaffe)_" ByteToCharDefault.lo 
ByteToCharIconv.lo CharToByteDefault.lo CharToByteIconv.lo ObjectInputStream.lo 
Class.lo ClassLoader.lo Compiler.lo Double.lo Float.lo Math.lo MemoryAdvice.lo 
NativeLibrary.lo Object.lo PrimordialClassLoader.lo Runtime.lo String.lo 
System.lo Thread.lo ThreadStack.lo Throwable.lo UNIXProcess.lo ZipFile.lo 
Array.lo Constructor.lo Field.lo Method.lo DateFormat.lo TestNative.lo 
Arrays.lo TimeZone.lo -lm
rm -fr .libs/libnative.la .libs/libnative.* .libs/libnative-1.1.x-cvs.*
generating symbol list for `libnative.la'
/usr/bin/nm -B  ByteToCharDefault.lo ByteToCharIconv.lo CharToByteDefault.lo 
CharToByteIconv.lo ObjectInputStream.lo Class.lo ClassLoader.lo Compiler.lo 
Double.lo Float.lo Math.lo MemoryAdvice.lo NativeLibrary.lo Object.lo 
PrimordialClassLoader.lo Runtime.lo String.lo System.lo Thread.lo 
ThreadStack.lo Throwable.lo UNIXProcess.lo ZipFile.lo Array.lo Constructor.lo 
Field.lo Method.lo DateFormat.lo TestNative.lo Arrays.lo TimeZone.lo  | sed -n 
-e 's/^.*[  ]\([ABCDGISTW][ABCDGISTW]*\)[   ][      
]*\(\)\([_A-Za-z][_A-Za-z0-9]*\)$/\1 \2\3 \3/p' | sed 's/.* //' | sort | uniq > 
.libs/libnative.exp
egrep -e "^([Jj]ava|kaffe)_" ".libs/libnative.exp" > ".libs/libnative.expT"
mv -f ".libs/libnative.expT" ".libs/libnative.exp"
gcc "{ global:" > .libs/libnative.ver
gcc: { global:: No such file or directory
gcc: no input files
make[2]: *** [libnative.la] Error 1
make[2]: Leaving directory `/home/rsdio/src/kaffe/libraries/clib/native'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/home/rsdio/src/kaffe/libraries/clib'
make: *** [all-recursive] Error 1
---

I've seen a handful of bug reports that end like this that come up
when searching for `gcc { global: ...' on Google.

The offending part of libtool (Kaffe is using 1.8.2 right now) appears
to be this, on line 514 (if that is meaningful):

---
# Mandrake: (gc) It's bad to link C++ code with GCC, so we need to use the 
compiler name if provided
if test "$mode" = link && test -n "$archive_cmds" && test -x "/usr/bin/perl"; 
then
    case $nonopt in
    *cc | *++ | gcc* | *-gcc* | egcs*)
            archive_cmds=`echo $archive_cmds | perl -pe 's/^\S+\s+//'`
            archive_cmds="$nonopt $archive_cmds"
            archive_expsym_cmds=`echo $archive_expsym_cmds | perl -pe 
's/^\S+\s+//'`
            archive_expsym_cmds="$nonopt $archive_expsym_cmds"
    esac
fi
---

This looks almost certainly wrong. `archive_expsym_cmds' is set to
this:

---
archive_expsym_cmds="\$echo \\\"{ global:\\\" > \$output_objdir/\$libname.ver~
cat \$export_symbols | sed -e \\\"s/\\\\(.*\\\\)/\\\\1;/\\\" >> 
\$output_objdir/\$libname.ver~
\$echo \\\"local: *; };\\\" >> \$output_objdir/\$libname.ver~
        \$CC -shared \$libobjs \$deplibs \$compiler_flags \${wl}-soname 
\$wl\$soname \${wl}-version-script \${wl}\$output_objdir/\$libname.ver -o \$lib"
---

So to me it looks like the above is assuming that the first word of
`archive_expsym_cmds' is something that links files, so stuffing gcc
($nonopt) in its place makes sense. Has `archive_expsym_cmds' changed
recently to make this not true?

Cheers.

-- 
Casey Marshall || address@hidden