C++ demangler horrors

[H. J. Lu]

On Tue, Jul 01, 2003 at 10:47:11AM -0300, Alexandre Oliva wrote:
> On Jun 28, 2003, "H. J. Lu" <hjl@lucon.org> wrote:
> 
> > On Sat, Jun 28, 2003 at 07:59:00AM -0700, H. J. Lu wrote:
> 
> >> My patch doesn't requrie it. But some people don't like the way I solved
> >> the problem with
> >> 
> >> ar -d ....
> >> ar rc ....
> 
> It's not about not liking.  Consider a parallel build in which someone
> is linking with libiberty while demangler goes and modifies it.  Bad
> idea.  And splitting this into demangler like this won't fix it:
> 
> > all:
> > 	$(MAKE) -C ../libiberty [new_demangler|old_demangler]
> 

I don't get it. Are you saying you can link with libiberty before
libiberty is done? In my post:

http://gcc.gnu.org/ml/gcc/2003-06/msg02404.html

I said

> 3. Change the top level dependency such that targets which used to depend on
> libiberty, now depend on demangler, demangler depends on libiberty and
> libstdc++ if it is needed. Please keep in mind that only demangler for target
> needs libstdc++ for target. I don't believe it has much impact on parallel
> build.

Can you tell me how a parallel build will fail with my proposal?


H.J.