Re: dlvsym support in FFI

[Kaz K. <kky...@gm...>]

On Wed, Mar 12, 2008 at 7:32 PM, Sam Steingold <sd...@gn...> wrote:
> Kaz Kylheku wrote:
> >   http://common-lisp.net/project/meta-cvs/downloads/dlvsym-ffi-support.patch
>
> coerce_ss_name is broken because coerce_ss is GC-unsafe, so after its
> first invocation name, orig_name, and orig_ver are invalid.

Is that actually true, regarding orig_name and orig_ver? These are the
car and cdr fields of name. The callers of coerce_ss_name already
ensure that this object is hooked into the stack. Surely the car and
cdr values can't be garbage collected if the parent cons is
referenced.

I just see the problem with ss_name and ss_ver

local maygc object coerce_ss_name(object name)   // <--- name is
hooked into stack
{
  if (consp(name)) {
    var object orig_name = Car(name);            // <--- transitively
reachable through name
    var object orig_ver = Cdr(name) ;            // <-/
    var object ss_name = coerce_ss(orig_name);   // <--- this is not
referenced, oops!
    var object ss_ver = coerce_ss(orig_ver);     // <--- this may GC!

    if (eq(orig_name, ss_name) && eq(orig_ver, ss_ver))
      return name;

    var object cons = allocate_cons();  // <--- this may GC
    Car(cons) = ss_name;
    Cdr(cons) = ss_ver;
    return cons;
  } else {
    return coerce_ss(name);
  }
}

> you need to push stuff on STACK to save it while you call coerce_ss.
> a good check if running your code in clisp configured like this:
> ./configure CC=g++ --with-debug --builg build-gxx

That's interesting; I will give that a try.