moon Phase 1 (start): ARC reclamation engine#86
Merged
Conversation
… free) Implements the core of automatic reference counting — the algorithmic heart of the post-GC memory model: - moonC_retain(o): increment refcount. - moonC_release(L, o): decrement; at zero, recursively release the object's GC children (a release-instead-of-mark mirror of the marking traversal — tables, closures, protos, userdata; strings/numbers are leaves), unlink from the global 'allgc' list, and reclaim via the existing freeobj(). Reference cycles never reach zero from an external release, so they leak by design. - A debug deinit counter (moonC_deinitcount / reset) for verification. Adds test_arc.cpp (GCC-only, like test_moonallocator) driving the engine on real object graphs with explicit ownership modelling (the VM/stack is not yet ARC-instrumented). Verified: an acyclic graph (root -> child table + leaf string) is reclaimed deterministically (3/3 objects), and an x<->y cycle leaks (0 freed) without crashing. Clean under ASan + UBSan with leak detection on (moon_close reclaims the leaked cycle at shutdown). NOT yet done (next increment): wiring retain/release into every TValue slot write across the VM stack and tables so ordinary moon programs free automatically — that needs the init-vs-assign / nil-on-free stack discipline and is the larger Phase 1 follow-up. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Refactor the ARC engine to a deferred-free model and wire it into table stores. Engine (mgc): - moonC_incref/moonC_decref adjust refcounts WITHOUT needing a moon_State — decref queues zero-count objects on a pending list. moonC_drain(L) reclaims the queue at a safe point, iteratively releasing children (no deep recursion, no freeing mid-write). moonC_release(L,o) = decref + drain. This solves the problem that the low-level table write primitives have no L to free with. Tables (mtable): - Table::set / Table::setInt now do ARC accounting: capture the old value, retain the new value (and, for a new entry, a collectable key), then release the old value after the write. resize() moves entries via insertkey and bypasses this path, so rehash moves are not double-counted. (Fixed an uninitialized capture-slot that caused a spurious decref.) Safety: the interpreter never calls moonC_drain yet, so no object is freed during normal execution — stores/overwrites/removes just adjust counts (which leak, exactly as in Phase 0). Real frees arrive once the stack is ARC-instrumented and drain is called at safe points. test_arc now builds graphs through the real Table::setInt API (no manual retain) and verifies: acyclic graph fully reclaimed (3/3), cycle leaks (0), clean under ASan+UBSan. Full interpreter smoke + a table build/overwrite/remove/iterate script pass. Not yet wired: table array-part fast paths and VM fast-set bypass Table::set/ setInt (safe while drain is off); the stack; calling drain at runtime. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
First increment of Phase 1 — ARC core: the reclamation engine that replaces tracing as moon's memory model.
What's implemented
moonC_retain(o)— increment refcount.moonC_release(L, o)— decrement; at zero, recursively release the object's GC children, unlink fromallgc, and reclaim via the existingfreeobj(). The child traversal is a release-instead-of-mark mirror of the marking code (tables incl. metatable/array/hash keys+values, Lua/C closures, protos, userdata; strings/numbers are leaves). Reference cycles never reach zero from an external release → they leak by design.moonC_deinitcount) for verification.test_arc.cpp(GCC-only) drives the engine on real object graphs.Verification
moon_closereclaims the leaked cycle at shutdown).Scope / what's next (honest status)
This is the reclamation engine — the hard algorithmic core — verified in isolation. It is not yet wired into the VM, so ordinary moon programs don't auto-free yet. The next increment (the larger, more invasive part) wires
retain/releaseinto everyTValueslot write across the stack and tables, which requires the init-vs-assign / nil-on-free stack discipline to stay use-after-free-free. Tracked as tasks #8/#9.Builds green; Phase 0 smoke still passes (engine is additive, not yet on the hot path).
🤖 Generated with Claude Code