Re: g++ unref symbols

Re: g++ unref symbols

[Alexandre Oliva]

On Aug  2, 2000, Petr Ovchenkov <ptr@ParaGraph.Ru> wrote:

> Compiler specs (indeed problem exist on all gcc compilers since
> 2.95.2, don't now about earlier versions): 

You'll see somewhere in the docs that GCC emits out-of-line copies of
every inline member-function of a class in the translation unit in
which the first non-inline non-abstract virtual member-function is
defined, if there is one.  This is a valid optimization because every
non-abstract virtual member-function must be defined.

-- 
Alexandre Oliva   Enjoy Guarana', see http://www.ic.unicamp.br/~oliva/
Red Hat GCC Developer                  aoliva@{cygnus.com, redhat.com}
CS PhD student at IC-Unicamp        oliva@{lsd.ic.unicamp.br, gnu.org}
Free Software Evangelist    *Please* write to mailing lists, not to me

Re: g++ unref symbols

[Petr Ovchenkov]

Alexandre Oliva writes:
 > On Aug  3, 2000, Petr Ovchenkov <ptr@ParaGraph.Ru> wrote:
 > 
 > > That's was 2.95.2. For later snapshots (mmm, at least since 20000530)
 > > for A::A scope instead GLOBAL will be LOCAL (on HP)
 > 
 > I see.  AFAIK, WEAK symbols are now supported and used on HP.  Maybe
 > they just show up as `t' symbols?  Or maybe it is indeed some new bug?

I never see WEAK symbols on HP.

c360> uname -a
HP-UX c360 B.11.00 A 9000/785 2002760784 two-user license

And due to native ld fails link when libraries compiled any gcc after
2.95.2, I think that's 't' isn't designate WEAK (I work with GAS).

 > 
 > So, summarizing to the list the discussion we've had in private,
 > given:
 > 
 > h.h:
 > struct A {
 >   A() {}
 >   virtual void f();
 > };
 > 
 > x.cc:
 > #include "h.h"
 > void A::f() {}
 > 
 > y.cc:
 > #include "h.h"
 > A a;
 > 
 > if you compile x.cc and y.cc without optimization, linking fails

One another issue should be payed attention:
virtual table (and type_info) of A class has local scope, so it still
be unresolved on linkage stage.

c360> c++ -fPIC -shared -o libA.sl -c lib.cc
c360> nm -C libA.sl
00000000 t $CODE$
40000018 d L$C0000
00000000 t A::A(void)
         U __rtti_user
00000000 t A type_info function
         ~~~~~~~~~~~~~~~~~~~~~~
00000008 C A type_info node
40000000 d A virtual table
         ~~~~~~~~~~~~~~~~~
00000000 t gcc2_compiled.
00000000 T A::qq(void)

 > because the constructor is defined as local in x.o and not defined at
 > all in y.  Is that correct?

Yes. That's what I want to say. Thanks.

       - Petr Ovchenkov

Re: g++ unref symbols

[Jeffrey A Law]

  In message < or4s52n2u2.fsf@guarana.lsd.ic.unicamp.br >you write:
  > On Aug  3, 2000, Petr Ovchenkov <ptr@ParaGraph.Ru> wrote:
  > 
  > > That's was 2.95.2. For later snapshots (mmm, at least since 20000530)
  > > for A::A scope instead GLOBAL will be LOCAL (on HP)
  > 
  > I see.  AFAIK, WEAK symbols are now supported and used on HP.  Maybe
  > they just show up as `t' symbols?  Or maybe it is indeed some new bug?
You must have a recent binutils snapshot to get the right behavior from
nm/objdump to detect weak symbols.  They use 'W' to denote a weak symbol.

jeff