FreeTDS Development Group

[Jonathan Monroe <monroej_freetds AT actualtechnologies.com>]

On Dec 30, 2003, at 5:58 PM, Frediano ZIGLIO wrote:

> I searched where to apply patch in distribution and I found this code
> (in ltcf-c.sh)
>
>   darwin* | rhapsody*)
>     case "$host_os" in
>     rhapsody* | darwin1.[012])
>       allow_undefined_flag='-undefined suppress'
>       ;;
>     *) # Darwin 1.3 on
>       allow_undefined_flag='-flat_namespace -undefined suppress'
>       ;;
>     esac
>     archive_cmds='$CC $(if test .$module = .yes; then echo -bundle;  
> else
> echo -dynamiclib; fi) $allow_undefined_flag -o $lib $libobjs $deplibs
> $linkopts -install_name $rpath/$soname $verstring'
>     # We need to add '_' to the symbols in $export_symbols first
>     #archive_expsym_cmds="$archive_cmds"' && strip -s $export_symbols
> $lib'
>     hardcode_direct=yes
>     hardcode_shlibpath_var=no
>     whole_archive_flag_spec='-all_load $convenience'
>     ;;
>
> so the line is processed only on Apple machines (rhapsody is another
> Apple platform)... I don't have this file, the code it's in my  
> configure
> script, however is much more complex:
>
>     darwin* | rhapsody*)
>     if $CC -v 2>&1 | grep 'Apple' >/dev/null ; then
>       archive_cmds_need_lc=no
>       case "$host_os" in
>       rhapsody* | darwin1.[012])
>         allow_undefined_flag='-undefined suppress'
>         ;;
>       *) # Darwin 1.3 on
>         test -z ${LD_TWOLEVEL_NAMESPACE} &&
> allow_undefined_flag='-flat_namespace -undefined suppress'
>         ;;
>       esac
>       # FIXME: Relying on posixy $() will cause problems for
>       #        cross-compilation, but unfortunately the echo tests do
> not
>       #        yet detect zsh echo's removal of \ escapes.  Also zsh
> mangles
>       #        `"' quotes if we put them in here... so don't!
>         lt_int_apple_cc_single_mod=no
>         output_verbose_link_cmd='echo'
>         if $CC -dumpspecs 2>&1 | grep 'single_module' >/dev/null ; then
>           lt_int_apple_cc_single_mod=yes
>         fi
>         if test "X$lt_int_apple_cc_single_mod" = Xyes ; then
>           archive_cmds='$CC -dynamiclib -single_module
> $allow_undefined_flag -o $lib $libobjs $deplibs $compiler_flags
> -install_name $rpath/$soname $verstring'
>         else
>         archive_cmds='$CC -r ${wl}-bind_at_load -keep_private_externs
> -nostdlib -o ${lib}-master.o $libobjs~$CC -dynamiclib
> $allow_undefined_flag -o $lib ${lib}-master.o $deplibs $compiler_flags
> -install_name $rpath/$soname $verstring'
>       fi
>       module_cmds='$CC -bundle ${wl}-bind_at_load $allow_undefined_flag
> -o $lib $libobjs $deplibs$compiler_flags'
>       # Don't fix this by using the ld -exported_symbols_list flag, it
> doesn't exist in older darwin ld's
>         if test "X$lt_int_apple_cc_single_mod" = Xyes ; then
>           archive_expsym_cmds='sed -e "s,#.*,," -e "s,^[    ]*,," -e
> "s,^\(..*\),_&," < $export_symbols >
> $output_objdir/${libname}-symbols.expsym~$CC -dynamiclib -single_module
> $allow_undefined_flag -o $lib $libobjs $deplibs $compiler_flags
> -install_name $rpath/$soname $verstring~nmedit -s
> $output_objdir/${libname}-symbols.expsym ${lib}'
>         else
>           archive_expsym_cmds='sed -e "s,#.*,," -e "s,^[    ]*,," -e
> "s,^\(..*\),_&," < $export_symbols >
> $output_objdir/${libname}-symbols.expsym~$CC -r ${wl}-bind_at_load
> -keep_private_externs -nostdlib -o ${lib}-master.o $libobjs~$CC
> -dynamiclib $allow_undefined_flag -o $lib ${lib}-master.o $deplibs
> $compiler_flags -install_name $rpath/$soname $verstring~nmedit -s
> $output_objdir/${libname}-symbols.expsym ${lib}'
>         fi
>           module_expsym_cmds='sed -e "s,#.*,," -e "s,^[    ]*,," -e
> "s,^\(..*\),_&," < $export_symbols >
> $output_objdir/${libname}-symbols.expsym~$CC -bundle
> $allow_undefined_flag  -o $lib $libobjs $deplibs$compiler_flags~nmedit
> -s $output_objdir/${libname}-symbols.expsym ${lib}'
>       hardcode_direct=no
>       hardcode_automatic=yes
>       hardcode_shlibpath_var=unsupported
>       whole_archive_flag_spec='-all_load $convenience'
>       link_all_deplibs=yes
>     fi
>       ;;
>
> I don't understand why darwin it's so different !!! I hope one day will
> have more compatible tools >:-<

This patch assumes that the linker should be building darwin dylibs  
with a flat namespace.  This document from Apple discusses why they  
should actually be built using two level namespaces starting in OS X  
10.1 (darwin 5.0):   
http://developer.apple.com/documentation/ReleaseNotes/DeveloperTools/ 
TwoLevelNamespaces.html

The main reason for using two level namespaces is to avoid namespace  
collisions when the same symbol is used in multiple modules.  In the  
case of an app that links to the darwin libiodbc.dylib, the ODBC  
functions would exist in both libiodbc.dylib and libtdsodbc.dylib.   
Since libtdsodbc.dylib is built using a flat namespace, the library  
loader gets confused and can't resolve the symbol.

I'd like to propose a patch, but I don't understand anything about  
these config scripts or even where they are maintained (this particular  
script doesn't appear to be in CVS), so I'll just describe what I think  
it should do and hopefully a toolsmith can pick this up and incorporate  
it into the build:

For versions of darwin earlier than 5.0, keep the script the way it is.
For 5.0 or later, allow_undefined_flag='-twolevel_namespace'

I think its OK to remove the "-undefined" flag.

Jonathan Monroe