Re: darwin/Fortran: -single_module

[Peter O'Gorman]

Ralf Wildenhues wrote:
> No testing done, and little experience with darwin, so if you deem ok
> and tests pass, then please go ahead, after looking at these comments:

Hi Ralf,
Thanks for the review! Had not noticed these bugs.

> > Index: ChangeLog
> > 2005-11-28  Peter O'Gorman  <address@hidden>
> >
> >         * libltdl/config/ltmain.m4sh, libltdl/m4/libtool.m4  [darwin]:
> >         Use -single_module by default when linking shared libraries. Can
> >         be overridden by specifying -multi-module or by setting the env
> >         var MULTI_MODULE before configure.
>
>
> Please mention the fact that MACOSX_DEPLOYMENT_TARGET is not used as a
> criterion any more in the ChangeLog.

Sure.

> Is MULTI_MODULE used by other tools on darwin?  If yes, should we not,
> for the sake of non-interacting, use something like LT_MULTI_MODULE?

It is not used anywhere else, but LT_ strikes me as better.

> Can it ever be useful to want to specify neither -single_module nor
> -multi_module?

When neither is specified then the default is used. Now, accourding to the 
man pages the default is multi_module, but that is only true for the 32 bit 
linker, the 64 bit linker defaults to -single_module. Where 
-single_module/-multi_module flags are not supported (older versions of the 
linker/compiler/OS) the default is multi_module.

> May users defeat your change by passing -Wl,-multi_module?  They
> wouldn't do that, right?

They'd get linker errors. The two conflicting flags may not be specified on 
the link line.

> Note indenting is inconsistenly wrt. TABs/spaces.

Hmm, I had thought I'd fixed them all to be spaces. Sorry.

> More comments inline.

> > +        output_verbose_link_cmd=echo
> > +        if test -z "$MULTI_MODULE"; then
> > +          # By default we will add the -single_module flag. You can override
> > +          # by either setting the environment variable MULTI_MODULE non-empty
> > +          # at configure time, or by adding -multi-module to the link flags.
> > +          # All ld's that accept -v accept -single_module
>
>
> Plural is `lds'.  This criterion seems quite dubious; if it's true..

Let me think about a better test, thanks.

> > +          if test -n "`ld -v 2>/dev/null`" ||
> > +            $CC -dumpspecs 2>&1 | $GREP 'single_module' >/dev/null; then
> > +              lt_int_apple_cc_single_mod=yes
> > +          fi
> > +       fi
> > +        if test "X$lt_int_apple_cc_single_mod" = Xyes ; then
> > +          _LT_TAGVAR(archive_cmds, $1)='$CC -dynamiclib $single_module 
> > $allow_undefined_flag -o $lib $libobjs $deplibs $compiler_flags -install_name 
> > $rpath/$soname $verstring'
> > +         _LT_TAGVAR(archive_expsym_cmds, $1)='sed -e "s,#.*,," -e "s,^[    ]*,," -e 
> > "s,^\(..*\),_&," < $export_symbols > $output_objdir/${libname}-symbols.expsym~$CC $single_module 
> > -dynamiclib $allow_undefined_flag -o $lib $libobjs $deplibs $compiler_flags -install_name $rpath/$soname 
> > $verstring~nmedit -s $output_objdir/${libname}-symbols.expsym ${lib}'
>
>
> Am I correct in that the
>    "s,^[    ]*,,"
> which will, after m4 quote removal, be
>    "s,^    *,,"
> is supposed to remove 3 or more leading spaces, and not mess with any
> TABs at all?  (Just asking because this is a likely editor-messes-it-up
> thing; several occasions.)

Meant to remove all leading whitespace. There should be a TAB in there. IIRC 
when this first went in I was quite  unsure of the required format of the 
libtool symbols file, and wanted to be extra careful. I'll look at this 
again later.

> I wonder whether all this export_symbols munging should not rather
> happen someplace else.  In global_symbol_pipe maybe?  The macro there
> even tests for leading underscores and stuff.  Where do the comment
> lines and the spaces come from?
>
> Surely, since you are not introducing this you may leave fixing it to
> another patch.

Thanks,
Peter