[m-rev.] make nb_reference table-able
Fergus Henderson
fjh at cs.mu.OZ.AU
Wed Aug 20 13:43:53 AEST 2003
On 20-Aug-2003, David Overton <dmo at cs.mu.OZ.AU> wrote:
> On Tue, Aug 12, 2003 at 04:52:39PM +1000, Fergus Henderson wrote:
> > Branches: main
> > Estimated hours taken: 1
> >
> > extras/references/nb_reference.m:
> > Use private_builtin.ref(T) rather than c_pointer for the
> > nb_reference type, so that nb_references can be copied using
> > copy/1 or stored in `pragma memo' tables.
>
> Is there a reason that you didn't make the same change in reference.m?
Yes. The reference.reference(T) type is represented differently
than private_builtin.ref(T). The latter is just a pointer,
whereas the former is an ME_Reference struct (see c_reference.h),
which contains not just a pointer, but also an MR_ChoicepointId.
If we pretended that it was a private_builtin.ref(T), then copy/1
or GC would only copy the pointer, not the choice point id.
I did consider the following diff instead, which would make
the reference.reference(T) type table-able, at the expense of
not being able to recover memory allocated for reference.reference(T)
by heap reclamation on failure (in `--gc none' grades).
But I decided that this was not necessarily a good trade-off.
The real solution is alluded to in the XXX comment below:
define a new MR_TYPECTOR_REP for ME_Reference, and use that.
Index: reference.m
===================================================================
RCS file: /home/mercury1/repository/mercury/extras/references/reference.m,v
retrieving revision 1.5
diff -u -d -u -r1.5 reference.m
--- reference.m 10 Jan 2003 05:49:22 -0000 1.5
+++ reference.m 12 Aug 2003 06:40:50 -0000
@@ -60,8 +60,24 @@
:- impure pred new_reference_2(T::in, c_pointer::out) is det.
:- pragma inline(new_reference_2/2).
:- pragma c_code(new_reference_2(X::in, Ref::out), will_not_call_mercury, "
+#ifdef MR_MIGHT_RECLAIM_HP_ON_FAILURE
+ /*
+ ** Ideally we would like to allocate references on the Mercury heap.
+ ** However, that leads to problems if we try to put them in memo
+ ** tables, since MR_deep_copy() (which is called from
+ ** MR_make_permanent(), which is called to ensure that the memo
+ ** table survives backtracking) will abort if it finds c_pointers that
+ ** point into the Mercury heap. Therefore, allocate them on
+ ** the C heap instead.
+ **
+ ** XXX Perhaps we should define a MR_TYPECTOR_REP for ME_Reference
+ ** and handle that case explicitly in MR_deep_copy().
+ */
+ Ref = (MR_Word) MR_GC_NEW(ME_Reference);
+#else
MR_incr_hp(Ref, (sizeof(ME_Reference) + sizeof(MR_Word) - 1) /
sizeof(MR_Word));
+#endif
((ME_Reference *) Ref)->value = (void *) X;
((ME_Reference *) Ref)->id = MR_current_choicepoint_id();
").
--
Fergus Henderson <fjh at cs.mu.oz.au> | "I have always known that the pursuit
The University of Melbourne | of excellence is a lethal habit"
WWW: <http://www.cs.mu.oz.au/~fjh> | -- the last words of T. S. Garp.
--------------------------------------------------------------------------
mercury-reviews mailing list
post: mercury-reviews at cs.mu.oz.au
administrative address: owner-mercury-reviews at cs.mu.oz.au
unsubscribe: Address: mercury-reviews-request at cs.mu.oz.au Message: unsubscribe
subscribe: Address: mercury-reviews-request at cs.mu.oz.au Message: subscribe
--------------------------------------------------------------------------
More information about the reviews
mailing list