This commit is contained in:
bQUARKz 2026-02-18 16:56:09 +00:00
parent 13f025e448
commit 966f0f9a8f
Signed by: bquarkz
SSH Key Fingerprint: SHA256:Z7dgqoglWwoK6j6u4QC87OveEq74WOhFN+gitsxtkf8
2 changed files with 136 additions and 54 deletions

View File

@ -1,11 +1,15 @@
use crate::{ObjectHeader, ObjectKind};
use prometeu_bytecode::HeapRef;
use prometeu_bytecode::{HeapRef, Value};
/// Internal stored object: header plus opaque payload bytes.
#[derive(Debug, Clone)]
pub struct StoredObject {
pub header: ObjectHeader,
/// Raw payload bytes for byte-oriented kinds (e.g., String, Bytes).
pub payload: Vec<u8>,
/// Optional typed elements for `ObjectKind::Array`.
/// When present, `header.payload_len` must equal `array_elems.len() as u32`.
pub array_elems: Option<Vec<Value>>,
}
/// Simple vector-backed heap. No GC or compaction.
@ -21,7 +25,17 @@ impl Heap {
/// Returns an opaque `HeapRef` handle.
pub fn allocate_object(&mut self, kind: ObjectKind, payload: &[u8]) -> HeapRef {
let header = ObjectHeader::new(kind, payload.len() as u32);
let obj = StoredObject { header, payload: payload.to_vec() };
let obj = StoredObject { header, payload: payload.to_vec(), array_elems: None };
let idx = self.objects.len();
self.objects.push(obj);
HeapRef(idx as u32)
}
/// Allocate a new `Array` object with the given `Value` elements.
/// `payload_len` stores the element count; raw `payload` bytes are empty.
pub fn allocate_array(&mut self, elements: Vec<Value>) -> HeapRef {
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);
HeapRef(idx as u32)
@ -37,6 +51,59 @@ impl Heap {
self.objects.get(r.0 as usize).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)
}
/// 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 {
ObjectKind::Array => {
// Traverse only Value::HeapRef inside the array.
o.array_elems
.as_ref()
.into_iter()
.flat_map(|v| v.iter())
.filter_map(|val| if let Value::HeapRef(h) = val { Some(*h) } else { None })
.collect::<Vec<_>>()
}
// These kinds have no inner references in this PR.
ObjectKind::String | ObjectKind::Bytes | ObjectKind::Closure | ObjectKind::UserData | ObjectKind::Unknown => {
Vec::new()
}
}
})
}
/// Mark phase: starting from the given roots, traverse and set mark bits
/// on all reachable objects. Uses an explicit stack to avoid recursion.
pub fn mark_from_roots<I: IntoIterator<Item = HeapRef>>(&mut self, roots: I) {
let mut stack: Vec<HeapRef> = roots.into_iter().collect();
while let Some(r) = stack.pop() {
if !self.is_valid(r) { continue; }
// If already marked, skip.
let already_marked = self.header(r).map(|h| h.is_marked()).unwrap_or(false);
if already_marked { continue; }
// Set mark bit.
if let Some(h) = self.header_mut(r) { h.set_marked(true); }
// Push children.
for child in self.children_of(r) {
if self.is_valid(child) {
// Check child's mark state cheaply to reduce stack churn.
let marked = self.header(child).map(|h| h.is_marked()).unwrap_or(false);
if !marked { stack.push(child); }
}
}
}
}
/// Current number of allocated objects.
pub fn len(&self) -> usize { self.objects.len() }
pub fn is_empty(&self) -> bool { self.objects.is_empty() }
@ -52,7 +119,7 @@ mod tests {
let r1 = heap.allocate_object(ObjectKind::String, b"hello");
let r2 = heap.allocate_object(ObjectKind::Bytes, &[1, 2, 3, 4]);
let r3 = heap.allocate_object(ObjectKind::Array, &[]);
let r3 = heap.allocate_array(vec![]);
assert!(heap.is_valid(r1));
assert!(heap.is_valid(r2));
@ -71,4 +138,70 @@ mod tests {
assert_eq!(h3.kind, ObjectKind::Array);
assert_eq!(h3.payload_len, 0);
}
#[test]
fn mark_reachable_through_array() {
let mut heap = Heap::new();
// Target object B (unreferenced yet)
let b = heap.allocate_object(ObjectKind::Bytes, &[9, 9, 9]);
// Array A that contains a reference to B among other primitives
let a = heap.allocate_array(vec![
Value::Int32(1),
Value::HeapRef(b),
Value::Boolean(false),
]);
// Mark starting from root A
heap.mark_from_roots([a]);
// Both A and B must be marked; random other objects are not allocated
assert!(heap.header(a).unwrap().is_marked());
assert!(heap.header(b).unwrap().is_marked());
}
#[test]
fn mark_does_not_mark_unreachable() {
let mut heap = Heap::new();
let unreachable = heap.allocate_object(ObjectKind::String, b"orphan");
let root = heap.allocate_object(ObjectKind::Bytes, &[1, 2, 3]);
heap.mark_from_roots([root]);
assert!(heap.header(root).unwrap().is_marked());
assert!(!heap.header(unreachable).unwrap().is_marked());
}
#[test]
fn mark_handles_cycles() {
let mut heap = Heap::new();
// Create two arrays that reference each other: A -> B, B -> A
let a_placeholder = HeapRef(0); // temporary
let b_placeholder = HeapRef(0);
// Allocate empty arrays first to get handles
let a = heap.allocate_array(vec![]);
let b = heap.allocate_array(vec![]);
// Now mutate their internal vectors via re-allocation pattern:
// 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) {
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) {
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]);
assert!(heap.header(a).unwrap().is_marked());
assert!(heap.header(b).unwrap().is_marked());
}
}

View File

@ -1,54 +1,3 @@
# PR-3.5 — Implement Mark Phase (Reachability Traversal)
### Briefing
This PR introduces the mark phase of a simple mark-sweep collector. It must traverse from roots and mark reachable objects.
### Target
* Implement reachability marking from root set.
* Set mark bits on visited objects.
### Work items
* Implement a `mark()` function in the heap.
* Traverse roots using the root iterator.
* For each `HeapRef`, mark the object.
* Recursively traverse references inside objects.
* Ensure no infinite loops on cycles.
### Acceptance checklist
* [ ] Mark phase visits all reachable objects.
* [ ] Cycles are handled safely.
* [ ] Unreachable objects remain unmarked.
* [ ] `cargo test` passes.
### Tests
* Add tests:
* Reachable object stays marked.
* Unreachable object remains unmarked.
* Cyclic references do not crash.
### Junie instructions
**You MAY:**
* Implement a simple DFS or stack-based marking.
**You MUST NOT:**
* Add generational, incremental, or parallel GC.
* Change object layout.
**If unclear:**
* Ask before deciding traversal structure.
---
# PR-3.6 — Implement Sweep Phase (Reclaim Unmarked Objects)
### Briefing