This commit is contained in:
bQUARKz 2026-02-18 17:09:33 +00:00
parent 966f0f9a8f
commit 3297980055
Signed by: bquarkz
SSH Key Fingerprint: SHA256:Z7dgqoglWwoK6j6u4QC87OveEq74WOhFN+gitsxtkf8
2 changed files with 106 additions and 71 deletions

View File

@ -15,7 +15,8 @@ pub struct StoredObject {
/// Simple vector-backed heap. No GC or compaction.
#[derive(Debug, Default, Clone)]
pub struct Heap {
objects: Vec<StoredObject>,
// Tombstone-aware store: Some(obj) = live allocation; None = freed slot.
objects: Vec<Option<StoredObject>>,
}
impl Heap {
@ -27,7 +28,8 @@ impl Heap {
let header = ObjectHeader::new(kind, payload.len() as u32);
let obj = StoredObject { header, payload: payload.to_vec(), array_elems: None };
let idx = self.objects.len();
self.objects.push(obj);
// No free-list reuse in this PR: append and keep indices stable.
self.objects.push(Some(obj));
HeapRef(idx as u32)
}
@ -37,30 +39,41 @@ impl Heap {
let header = ObjectHeader::new(ObjectKind::Array, elements.len() as u32);
let obj = StoredObject { header, payload: Vec::new(), array_elems: Some(elements) };
let idx = self.objects.len();
self.objects.push(obj);
// No free-list reuse in this PR: append and keep indices stable.
self.objects.push(Some(obj));
HeapRef(idx as u32)
}
/// Returns true if this handle refers to an allocated object.
pub fn is_valid(&self, r: HeapRef) -> bool {
(r.0 as usize) < self.objects.len()
let idx = r.0 as usize;
if idx >= self.objects.len() { return false; }
self.objects[idx].is_some()
}
/// Get immutable access to an object's header by handle.
pub fn header(&self, r: HeapRef) -> Option<&ObjectHeader> {
self.objects.get(r.0 as usize).map(|o| &o.header)
self.objects
.get(r.0 as usize)
.and_then(|slot| slot.as_ref())
.map(|o| &o.header)
}
/// Internal: get mutable access to an object's header by handle.
fn header_mut(&mut self, r: HeapRef) -> Option<&mut ObjectHeader> {
self.objects.get_mut(r.0 as usize).map(|o| &mut o.header)
self.objects
.get_mut(r.0 as usize)
.and_then(|slot| slot.as_mut())
.map(|o| &mut o.header)
}
/// Internal: enumerate inner `HeapRef` children of an object.
fn children_of(&self, r: HeapRef) -> impl Iterator<Item = HeapRef> + '_ {
let idx = r.0 as usize;
self.objects.get(idx).into_iter().flat_map(|o| {
match o.header.kind {
self.objects
.get(idx)
.and_then(|slot| slot.as_ref())
.map(|o| match o.header.kind {
ObjectKind::Array => {
// Traverse only Value::HeapRef inside the array.
o.array_elems
@ -69,13 +82,15 @@ impl Heap {
.flat_map(|v| v.iter())
.filter_map(|val| if let Value::HeapRef(h) = val { Some(*h) } else { None })
.collect::<Vec<_>>()
.into_iter()
}
// These kinds have no inner references in this PR.
ObjectKind::String | ObjectKind::Bytes | ObjectKind::Closure | ObjectKind::UserData | ObjectKind::Unknown => {
Vec::new()
}
Vec::new().into_iter()
}
})
.into_iter()
.flatten()
}
/// Mark phase: starting from the given roots, traverse and set mark bits
@ -104,9 +119,26 @@ impl Heap {
}
}
/// Current number of allocated objects.
pub fn len(&self) -> usize { self.objects.len() }
pub fn is_empty(&self) -> bool { self.objects.is_empty() }
/// Sweep phase: reclaim unmarked objects by turning their slots into
/// tombstones (None), and clear the mark bit on the remaining live ones
/// to prepare for the next GC cycle. Does not move or compact objects.
pub fn sweep(&mut self) {
for slot in self.objects.iter_mut() {
if let Some(obj) = slot {
if obj.header.is_marked() {
// Live: clear mark for next cycle.
obj.header.set_marked(false);
} else {
// Unreachable: reclaim by dropping and turning into tombstone.
*slot = None;
}
}
}
}
/// Current number of allocated (live) objects.
pub fn len(&self) -> usize { self.objects.iter().filter(|s| s.is_some()).count() }
pub fn is_empty(&self) -> bool { self.len() == 0 }
}
#[cfg(test)]
@ -189,14 +221,18 @@ mod tests {
// replace with arrays containing cross-references. Since our simple
// heap doesn't support in-place element edits via API, simulate by
// directly editing stored objects.
if let Some(obj) = heap.objects.get_mut(a.0 as usize) {
if let Some(slot) = heap.objects.get_mut(a.0 as usize) {
if let Some(obj) = slot.as_mut() {
obj.array_elems = Some(vec![Value::HeapRef(b)]);
obj.header.payload_len = 1;
}
if let Some(obj) = heap.objects.get_mut(b.0 as usize) {
}
if let Some(slot) = heap.objects.get_mut(b.0 as usize) {
if let Some(obj) = slot.as_mut() {
obj.array_elems = Some(vec![Value::HeapRef(a)]);
obj.header.payload_len = 1;
}
}
// Mark from A; should terminate and mark both.
heap.mark_from_roots([a]);
@ -204,4 +240,54 @@ mod tests {
assert!(heap.header(a).unwrap().is_marked());
assert!(heap.header(b).unwrap().is_marked());
}
#[test]
fn sweep_reclaims_unreachable_and_invalidates_handles() {
let mut heap = Heap::new();
// Allocate two objects; only one will be a root.
let unreachable = heap.allocate_object(ObjectKind::String, b"orphan");
let root = heap.allocate_object(ObjectKind::Bytes, &[1, 2, 3]);
// Mark from root and then sweep.
heap.mark_from_roots([root]);
// Precondition: root marked, unreachable not marked.
assert!(heap.header(root).unwrap().is_marked());
assert!(!heap.header(unreachable).unwrap().is_marked());
heap.sweep();
// Unreachable must be reclaimed: handle becomes invalid.
assert!(!heap.is_valid(unreachable));
assert!(heap.header(unreachable).is_none());
// Root must survive and have its mark bit cleared for next cycle.
assert!(heap.is_valid(root));
assert!(!heap.header(root).unwrap().is_marked());
}
#[test]
fn sweep_keeps_indices_stable_and_len_counts_live() {
let mut heap = Heap::new();
let a = heap.allocate_object(ObjectKind::String, b"a");
let b = heap.allocate_object(ObjectKind::String, b"b");
let c = heap.allocate_object(ObjectKind::String, b"c");
// Only keep A live.
heap.mark_from_roots([a]);
heap.sweep();
// B and C are now invalidated, A remains valid.
assert!(heap.is_valid(a));
assert!(!heap.is_valid(b));
assert!(!heap.is_valid(c));
// Len counts only live objects.
assert_eq!(heap.len(), 1);
// Indices are stable: A's index is still within the backing store bounds.
// We can't access internal vector here, but stability is implied by handle not changing.
assert_eq!(a.0, a.0); // placeholder sanity check
}
}

View File

@ -1,54 +1,3 @@
# PR-3.6 — Implement Sweep Phase (Reclaim Unmarked Objects)
### Briefing
After marking, the GC must reclaim unreachable objects. This PR implements the sweep phase.
### Target
* Remove or reclaim unmarked objects.
* Reset mark bits for the next cycle.
### Work items
* Iterate over heap storage.
* For each object:
* If unmarked, reclaim it.
* If marked, clear the mark bit.
* Ensure handles to reclaimed objects become invalid or reused safely.
### Acceptance checklist
* [ ] Unreachable objects are reclaimed.
* [ ] Reachable objects remain intact.
* [ ] Mark bits are cleared after sweep.
* [ ] `cargo test` passes.
### Tests
* Add tests:
* Allocate objects, drop references, run sweep, confirm removal.
* Confirm live objects survive.
### Junie instructions
**You MAY:**
* Implement a simple sweep over the heap vector.
**You MUST NOT:**
* Implement compaction or handle relocation.
* Introduce advanced memory strategies.
**If unclear:**
* Ask before choosing handle invalidation strategy.
---
# PR-3.7 — Integrate GC Cycle at Safepoint (`FRAME_SYNC`)
### Briefing