Re: HP-UX 11.00 and parsing /bin/nm -p output (failing)

[Gary V . Vaughan]

On Monday 25 June 2001 12:50 am, Gary V. Vaughan wrote:
> On Thursday 03 May 2001  7:11 pm, address@hidden wrote:
> > On Mon, Apr 23, 2001 at 03:28:52PM -0500,
>
> address@hidden wrote:
> > > On Mon, Apr 23, 2001 at 02:11:52PM -0500, address@hidden
>
> wrote:
> > > > This pertains to the HEAD branch.
> > > >
> > > > When parsing the output of /bin/nm -p, libtool.m4 contains the
> > > > following sed statement to massage the output:
> > > >
> > > > # Try without a prefix undercore, then with it.
> > > > for ac_symprfx in "" "_"; do
> > > >
> > > >   # Write the raw and C identifiers.
> > > > [lt_cv_sys_global_symbol_pipe="sed -n -e 's/^.*[       
> > > > ]\($symcode\)[ ][      ]*\($ac_symprfx\)$sympat$opt_cr$/$symxfrm/p'"]
> > > >
> > > > So, there can be only *one* $symcode between spaces in the /bin/nm -p
> > > > output. While creating demo/libtool during a run of 'make tests' for
> > > > demo-conf.test, /bin/nm -p generates the following on the sample
> > > > conftest.o file:
> > > >   $ /bin/nm -p conftest.o
> > > >   0000000008 t  D$0
> > > >   0000000060 t  D$1
> > > >   0000000008 T  main
> > > >   0000000000 T  nm_test_func
> > > >   0000000001 CS nm_test_var
> > > >
> > > > [snip ...]
> > > >
> > > > So, it seems there can be *more* than one $symcode. So, how about we
> > > > change the sed statement to look for *more* than one $symcode
> > > > separated by spaces? I have looked at the man pages of nm on Solaris
> > > > and Tru64 UNIX and don't see anywhere that there is only one $symcode
> > > > output.
> > >
> > > Patch below.
> >
> > Ok, one more problem. Check out the following from ltmain.in (line
> > ~3300):
> >
> >    sed -n -e 's/^: \([^ ]*\) $/  {\"\1\", (lt_ptr_t) 0},/p' \
> >        -e 's/^. \([^ ]*\) \([^ ]*\)$/  {"\2", (lt_ptr_t) \&\2},/p' \
> >          < "$nlist" >> "$output_objdir/$dlsyms"
> >
> > It still assumes one $symcode. The side-effect of this is that self
> > dlopen fails (well, variables cannot be found). Making the change here
> > is more difficult because we don't have access to $symcode anymore. In
> > my patch, I used $symcode to be more precise in the sed statement than
> > simply adding 's/^.. ' to the already-existing 's/^. '.  However, what
> > do we do here? Do we add the ugly hack of:
> >
> >    sed -n -e 's/^: \([^ ]*\) $/  {\"\1\", (lt_ptr_t) 0},/p' \
> >        -e 's/^. \([^ ]*\) \([^ ]*\)$/  {"\2", (lt_ptr_t) \&\2},/p' \
> >        -e 's/^.. \([^ ]*\) \([^ ]*\)$/  {"\2", (lt_ptr_t) \&\2},/p' \
> >          < "$nlist" >> "$output_objdir/$dlsyms"
> >
> > I have tested this and it works.
>
> Why not pass the expression to ltmain as a config variable?  I propose the
> attached patch.  Okay to commit?

I'm invoking the 72 hour rule to commit this patch.

Albert, if this doesn't fix the problem you were having, let me know...

Cheers,
        Gary.
-- 
  ())_.  Gary V. Vaughan     gary@(oranda.demon.co.uk|gnu.org)
  ( '/   Research Scientist  http://www.oranda.demon.co.uk        ,_())____
  / )=   GNU Hacker          http://www.gnu.org/software/libtool   \'      `&
`(_~)_   Tech' Author        http://sources.redhat.com/autobook    =`---d__/