Re: Issues holding up 1.6.0 release.

[Gary Houston]

> From: Gary Houston <address@hidden>
> Date: 19 Sep 2001 23:09:05 -0000
> 
> > From: Rob Browning <address@hidden>
> > Date: Tue, 18 Sep 2001 08:36:00 -0500
> > 
> > Right now, I'm waiting on someone to tell me what I should do about
> > scm_call_with_dynamic_root.  I don't feel like that's a decision I
> > should make on my own, unless people indicate they want me to.  In
> > particular:
> > 
> >   - should we create the new function that doesn't (re/un)wind the
> >     dynamic-wind stack?
> > 
> >   - should this function go into 1.5.3, or should apps like guppi,
> >     scwm, and possibly gnucash just continue to copy-paste from root.c
> >     for the next release?
> > 
> >   - I think we already know what the function should look like, but
> >     what should the new function be named, and should it show up on
> >     the scm side, or just the C side?
> 
> Here's a test implementation of the "new function" idea.  I'm not sure
> if this would be enough for Guppi, the stack/backtrace issues need
> checking.
> 
> static void
> before (void *unused)
> {
>   SCM_SEQ (scm_rootcont) = ++n_dynamic_roots;
> }
> 
> static void
> after (void *v_oldseq)
> {
>   SCM_SEQ (scm_rootcont) = *(unsigned long *) v_oldseq;
> }
> 
> SCM 
> scm_c_with_continuation_barrier (scm_t_catch_body body, void *body_data)
> {
>   unsigned long old_seq = SCM_SEQ (scm_rootcont);
> 
>   return scm_internal_dynamic_wind (before, body, after, body_data, &old_seq);
> }
> 
> 
> Optional extras for a Scheme interface:
> 
> static SCM
> wcb_body (void *v_thunk)
> {
>   return scm_apply (*(SCM *) v_thunk, SCM_EOL, SCM_EOL);
> }
> 
> SCM_DEFINE (scm_with_continuation_barrier, "with-continuation-barrier",
>           1, 0, 0,
>           (SCM thunk),
>           "Evalutate @var{thunk} ..."
> #define FUNC_NAME s_scm_with_continuation_barrier
> {
>   return scm_c_with_continuation_barrier (wcb_body, &thunk);
> }
> #undef FUNC_NAME

I think we should experiment with Guppi and leave this stuff out of
1.6.

The continuation barrier above has the strange property that
exceptions can pass through, but escape continuations can not.  Since
it seems that only reentry to the body needs to be prevented, maybe
this would be sufficient:

static void
before (void *v_count)
{
  if (*(int *) v_count > 0)
    scm_misc_error ("scm_c_with_reentry_barrier",
                    "attempt to break reentry barrier", SCM_EOL);
  (*(int *) v_count)++;
}

static void
after (void *unused)
{
}

SCM
scm_c_with_reentry_barrier (scm_t_catch_body body, void *body_data)
{
  int count = 0;

  return scm_internal_dynamic_wind (before, body, after, body_data, &count);
}

static SCM
do_body (void *v_thunk)
{
  return scm_apply (*(SCM *) v_thunk, SCM_EOL, SCM_EOL);
}

SCM_DEFINE (scm_with_reentry_barrier, "with-reentry-barrier",
            1, 0, 0,
            (SCM thunk),
            "Evaluate @var{thunk}....")
#define FUNC_NAME s_scm_with_reentry_barrier
{
  return scm_c_with_reentry_barrier (do_body, &thunk);
}
#undef FUNC_NAME