fix(gc)+perf(closure): root the uint8 Buffer callback dispatcher (#8179) and hoist per-element closure dispatch (#8180)

changelog.d/8188-uint8-callback-rooting-and-hoisted-dispatch.md

@@ -0,0 +1,96 @@
+### Fixed — GC rooting in the uint8 Buffer callback dispatcher (#8179)
+
+`dispatch_uint8_buffer_method` — the shared uint8 `%TypedArray%.prototype`
+dispatcher that every Buffer-backed `Uint8Array` callback method funnels
+through, on all three of its entries — kept the callback closure, the receiver,
+`map`'s freshly allocated result buffer, `sort`/`toSorted`'s permuted output and
+`reduce`/`reduceRight`'s accumulator in bare Rust locals across
+`js_closure_call{2,3,4}`.
+
+The closure is the live half. It is an ordinary nursery allocation
+(`GC_TYPE_CLOSURE`, with a `GcMoveHookKind::ClosureDynamicProps` move hook — it
+both moves and dies), and a callback handed in by a frameless caller is
+reachable only through that raw parameter plus the native stack, which an
+evacuating minor does not scan. `array::buffer_receiver_dispatch` rooted it at
+the boundary; the `%TypedArray%.prototype` thunk and `dispatch_buffer_method`'s
+catch-all did not, so both `Uint8Array.prototype.map.call(u, fn)` and a
+statically typed `u.map(fn)` were exposed.
+
+`test-files/test_gap_gc_uint8_buffer_callback_rooting.ts` fails on the **shipped
+default** before the fix (`TypeError: value is not a function`, exit 1), and
+under `PERRY_GC_SCHEDULE_RATE=1 PERRY_GC_SCHEDULE_SEED=<n>
+PERRY_GC_PROTECT_FROMSPACE=1 PERRY_GC_VERIFY_EVACUATION=1` it dies on the FIRST
+scheduled collection, for every seed tried (1, 7, 42):
+
+```
+[gc-fromspace-protect] FAULT: signal 10 at 0x2454161058c
+  last-known object: user_ptr=0x24541610580 obj_type=4 size=24
+[gc-schedule] FAILURE (signal 10) under seed=7
+[gc-schedule]   safepoints=1 scheduled_collections=1
+```
+
+`obj_type=4` is `GC_TYPE_CLOSURE` and the faulting address is `user_ptr + 12` —
+`CLOSURE_TYPE_TAG_OFFSET`, i.e. `get_valid_func_ptr`'s `CLOSURE_MAGIC` probe
+reading a retired from-space closure header. After the fix the same seeds run to
+completion with the instrument's own liveness verdict showing the subject was
+live: `safepoints=5306 scheduled_collections=5306 copying_minors=5306
+moved_objects=25760`, zero faults, exit 0. The witness is registered in
+`test-parity/gc_repsel_corpus.txt`.
+
+The receiver is the other half, and is treated differently on purpose. A Buffer
+is `arena_alloc_gc_old` + `GC_FLAG_TENURED` (`buffer/header.rs`) — the same
+old-arena space `typed_array_alloc` calls "non-movable space: raw data pointers
+are handed out" — and every `%TypedArray%` sibling already holds its receiver in
+a plain local across callbacks on that invariant. It is now rooted for
+*liveness* (the raw parameter is otherwise its only reference on two of three
+entries) with its address read from the root once per arm rather than once per
+element. The one collector arm that relocates an old-arena page is old-page
+defrag, which is opt-in and default-off (`PERRY_GC_OLD_DEFRAG=1`); making that
+safe is a tree-wide property of every holder of an old-arena raw address, and
+re-reading per element cost +28 % on the `Uint8Array` benchmark for a knob that
+is off.
+
+Two sibling families get the same treatment: `js_typed_array_reduce` /
+`js_typed_array_reduce_right` now root their accumulator (as `js_array_reduce`
+has since the 2026-07-02 audit), and the two non-BigInt arms of
+`js_typed_array_sort_with_comparator` /
+`js_typed_array_to_sorted_with_comparator` now root the comparator closure —
+`sorted_bigint_lanes` beside them already did.
+
+### Performance — hoisted per-element closure dispatch (#8180)
+
+`js_closure_callN` re-derived, on every element of every fused array-callback
+loop, three answers that cannot change while one closure is being called:
+`get_valid_func_ptr` (two address-band checks, a volatile `CLOSURE_MAGIC` probe
+through `*(closure + 12)`, a volatile `func_ptr` load), `resolve_strategy` (a
+`perry_thread_local!` single-slot cache — on Darwin a `tlv_get_addr` CALL plus a
+load and a compare even on a hit), and the `DispatchStrategy` match before the
+indirect jump.
+
+New `closure/dispatch/direct.rs` generalises `array/sort.rs`'s `ComparatorCall`
+trick — introduced to "skip ~50M HashMap lookups over a 1.25M-element sort", and
+until now its only consumer — into `DirectCall{1,2,3,4}`: resolve once, call
+directly, fall back to `js_closure_callN` for a bound method/function, a rest
+parameter, a declared arity above the call arity, or an invalid closure pointer.
+`resolve_call2_direct` is deleted rather than left standing beside it.
+
+Hoisted at 31 call sites: `array/iter_methods.rs` (14), `array/reduce_right.rs`
+(1), `typedarray/iterate.rs` (9), `typedarray/transform.rs` (5 — including the
+BigInt lane comparator, which resolved once per *comparison*) and the uint8
+`%TypedArray%.prototype` dispatcher.
+
+Instructions retired, quiet M1 mini, best of 5, arms interleaved:
+
+| benchmark | main | +#8179 | +#8179+#8180 | vs main |
+|---|---|---|---|---|
+| 21M plain-`Array` callback invocations | 5,027,207,909 | 5,027,015,176 | 3,970,592,563 | **−21.0 %** |
+| 7.9M Buffer-`Uint8Array` callback invocations | 1,979,043,224 | 2,535,814,615 | 1,978,697,176 | **−0.02 %** |
+
+Peak RSS is flat: 46,628,864 B on both arms of the first benchmark;
+14,794,752 → 14,876,672 B (+0.55 %, 20 pages) on the second.
+
+`array/generic.rs`'s `js_arraylike_*` engine is deliberately not converted: its
+per-element cost is dominated by generic array-like property access (`al_has` +
+`al_get`), not dispatch, and the spec order it implements reads
+`LengthOfArrayLike` before `IsCallable`, so a hoisted resolve would have to be
+sequenced after the existing `callable()` call rather than inserted at the top.

crates/perry-runtime/src/array/iter_methods.rs

@@ -1,6 +1,6 @@
 //! Higher-order array methods.
 use super::*;
-use crate::closure::{js_closure_call3, js_closure_call4, ClosureHeader};
+use crate::closure::ClosureHeader;
 use std::ptr;
 
 /// NaN-box an array header pointer as the JS `array` receiver value passed as
@@ -195,6 +195,10 @@ pub extern "C" fn js_array_forEach(arr: *const ArrayHeader, callback: *const Clo
         let length = (*arr).length;
         let scope = crate::gc::RuntimeHandleScope::new();
         let rooted = RootedIterArray::new(&scope, arr);
+        // #8180: resolve the callback's dispatch ONCE. It is invariant for a
+        // fixed closure (see closure/dispatch/direct.rs), and this loop calls
+        // exactly one.
+        let cb_site = crate::closure::DirectCall3::resolve(callback);
         // The override is a movable `ObjectHeader` held across user callbacks
         // that allocate — root it for the duration of the loop.
         let self_handle = self_override.map(|recv| scope.root_nanbox_f64(recv));
@@ -210,7 +214,7 @@ pub extern "C" fn js_array_forEach(arr: *const ArrayHeader, callback: *const Clo
                     continue;
                 }
                 let element = crate::array::array_spec_get(arr, i as u32);
-                js_closure_call3(callback, element, i as f64, self_value(&rooted));
+                cb_site.call(callback, element, i as f64, self_value(&rooted));
             }
             return;
         }
@@ -222,7 +226,7 @@ pub extern "C" fn js_array_forEach(arr: *const ArrayHeader, callback: *const Clo
             // dispatch path supports call3 safely, so bound native
             // methods like `array.forEach(console.log)` can observe the
             // source array just like Node.
-            js_closure_call3(callback, element, i as f64, self_value(&rooted));
+            cb_site.call(callback, element, i as f64, self_value(&rooted));
         }
     }
 }
@@ -261,6 +265,10 @@ pub extern "C" fn js_array_map(
         let length = (*arr).length;
         let scope = crate::gc::RuntimeHandleScope::new();
         let rooted = RootedIterArray::new(&scope, arr);
+        // #8180: resolve the callback's dispatch ONCE. It is invariant for a
+        // fixed closure (see closure/dispatch/direct.rs), and this loop calls
+        // exactly one.
+        let cb_site = crate::closure::DirectCall3::resolve(callback);
         // Root the callback closure across the iteration. A callback allocated
         // by a frameless caller (arrow/method — #6081) is reachable ONLY via
         // this raw param + the native stack, which an evacuating minor does NOT
@@ -307,7 +315,7 @@ pub extern "C" fn js_array_map(
             };
             // JS .map() callback receives (element, index, array).
             let callback = cb_handle.get_raw_const_ptr::<ClosureHeader>();
-            let mapped = js_closure_call3(callback, element, i as f64, rooted.receiver());
+            let mapped = cb_site.call(callback, element, i as f64, rooted.receiver());
             if is_plain {
                 let result = result_arr(&result_rooted);
                 let result_elements =
@@ -363,6 +371,10 @@ pub extern "C" fn js_array_map_discard(arr: *const ArrayHeader, callback: *const
         let length = (*arr).length;
         let scope = crate::gc::RuntimeHandleScope::new();
         let rooted = RootedIterArray::new(&scope, arr);
+        // #8180: resolve the callback's dispatch ONCE. It is invariant for a
+        // fixed closure (see closure/dispatch/direct.rs), and this loop calls
+        // exactly one.
+        let cb_site = crate::closure::DirectCall3::resolve(callback);
         // The callback needs the same root its sibling `js_array_map` gives it
         // (#6081), and for the same reason: a callback allocated by a frameless
         // caller — the arrow in `xs.map(x => …)` — is reachable ONLY through this
@@ -396,15 +408,15 @@ pub extern "C" fn js_array_map_discard(arr: *const ArrayHeader, callback: *const
                     continue;
                 }
                 let element = crate::array::array_spec_get(arr, i as u32);
-                let _ = js_closure_call3(current_callback(), element, i as f64, rooted.receiver());
+                let _ = cb_site.call(current_callback(), element, i as f64, rooted.receiver());
             }
             return;
         }
         for i in 0..length as usize {
             let Some(element) = rooted.present(i) else {
                 continue;
             };
-            let _ = js_closure_call3(current_callback(), element, i as f64, rooted.receiver());
+            let _ = cb_site.call(current_callback(), element, i as f64, rooted.receiver());
         }
     }
 }
@@ -443,6 +455,10 @@ pub extern "C" fn js_array_filter(
         let length = (*arr).length;
         let scope = crate::gc::RuntimeHandleScope::new();
         let rooted = RootedIterArray::new(&scope, arr);
+        // #8180: resolve the callback's dispatch ONCE. It is invariant for a
+        // fixed closure (see closure/dispatch/direct.rs), and this loop calls
+        // exactly one.
+        let cb_site = crate::closure::DirectCall3::resolve(callback);
         // Root the callback across the loop — see js_array_map / gh #6206.
         let cb_handle = scope.root_raw_const_ptr(callback);
         let _tg = DenseThisGuard::bind_undefined();
@@ -473,7 +489,7 @@ pub extern "C" fn js_array_filter(
                 }
             };
             let callback = cb_handle.get_raw_const_ptr::<ClosureHeader>();
-            let keep = js_closure_call3(callback, element, i as f64, rooted.receiver());
+            let keep = cb_site.call(callback, element, i as f64, rooted.receiver());
             // Proper truthy check: handles NaN-boxed booleans (TAG_FALSE != 0.0 but is falsy)
             if crate::value::js_is_truthy(keep) != 0 {
                 if is_plain {
@@ -527,6 +543,10 @@ pub extern "C" fn js_array_find(arr: *const ArrayHeader, callback: *const Closur
         let length = (*arr).length;
         let scope = crate::gc::RuntimeHandleScope::new();
         let rooted = RootedIterArray::new(&scope, arr);
+        // #8180: resolve the callback's dispatch ONCE. It is invariant for a
+        // fixed closure (see closure/dispatch/direct.rs), and this loop calls
+        // exactly one.
+        let cb_site = crate::closure::DirectCall3::resolve(callback);
         let _tg = DenseThisGuard::bind_undefined();
         let exotic = crate::array::array_iteration_is_exotic(arr);
 
@@ -536,7 +556,7 @@ pub extern "C" fn js_array_find(arr: *const ArrayHeader, callback: *const Closur
             } else {
                 rooted.get_or_undefined(i)
             };
-            let result = js_closure_call3(callback, element, i as f64, rooted.receiver());
+            let result = cb_site.call(callback, element, i as f64, rooted.receiver());
             // Proper truthy check: handles NaN-boxed booleans
             if crate::value::js_is_truthy(result) != 0 {
                 return element;
@@ -583,6 +603,10 @@ pub extern "C" fn js_array_findIndex(
         let length = (*arr).length;
         let scope = crate::gc::RuntimeHandleScope::new();
         let rooted = RootedIterArray::new(&scope, arr);
+        // #8180: resolve the callback's dispatch ONCE. It is invariant for a
+        // fixed closure (see closure/dispatch/direct.rs), and this loop calls
+        // exactly one.
+        let cb_site = crate::closure::DirectCall3::resolve(callback);
         let _tg = DenseThisGuard::bind_undefined();
         let exotic = crate::array::array_iteration_is_exotic(arr);
 
@@ -592,7 +616,7 @@ pub extern "C" fn js_array_findIndex(
             } else {
                 rooted.get_or_undefined(i)
             };
-            let result = js_closure_call3(callback, element, i as f64, rooted.receiver());
+            let result = cb_site.call(callback, element, i as f64, rooted.receiver());
             // Proper truthy check: handles NaN-boxed booleans
             if crate::value::js_is_truthy(result) != 0 {
                 return i as i32;
@@ -624,6 +648,10 @@ pub extern "C" fn js_array_find_last(
         let length = (*arr).length as usize;
         let scope = crate::gc::RuntimeHandleScope::new();
         let rooted = RootedIterArray::new(&scope, arr);
+        // #8180: resolve the callback's dispatch ONCE. It is invariant for a
+        // fixed closure (see closure/dispatch/direct.rs), and this loop calls
+        // exactly one.
+        let cb_site = crate::closure::DirectCall3::resolve(callback);
         let _tg = DenseThisGuard::bind_undefined();
         let exotic = crate::array::array_iteration_is_exotic(arr);
         for i in (0..length).rev() {
@@ -632,7 +660,7 @@ pub extern "C" fn js_array_find_last(
             } else {
                 rooted.get_or_undefined(i)
             };
-            let result = js_closure_call3(callback, element, i as f64, rooted.receiver());
+            let result = cb_site.call(callback, element, i as f64, rooted.receiver());
             if crate::value::js_is_truthy(result) != 0 {
                 return element;
             }
@@ -662,6 +690,10 @@ pub extern "C" fn js_array_find_last_index(
         let length = (*arr).length as usize;
         let scope = crate::gc::RuntimeHandleScope::new();
         let rooted = RootedIterArray::new(&scope, arr);
+        // #8180: resolve the callback's dispatch ONCE. It is invariant for a
+        // fixed closure (see closure/dispatch/direct.rs), and this loop calls
+        // exactly one.
+        let cb_site = crate::closure::DirectCall3::resolve(callback);
         let _tg = DenseThisGuard::bind_undefined();
         let exotic = crate::array::array_iteration_is_exotic(arr);
         for i in (0..length).rev() {
@@ -670,7 +702,7 @@ pub extern "C" fn js_array_find_last_index(
             } else {
                 rooted.get_or_undefined(i)
             };
-            let result = js_closure_call3(callback, element, i as f64, rooted.receiver());
+            let result = cb_site.call(callback, element, i as f64, rooted.receiver());
             if crate::value::js_is_truthy(result) != 0 {
                 return i as i32;
             }
@@ -757,6 +789,10 @@ pub extern "C" fn js_array_some(arr: *const ArrayHeader, callback: *const Closur
         let length = (*arr).length;
         let scope = crate::gc::RuntimeHandleScope::new();
         let rooted = RootedIterArray::new(&scope, arr);
+        // #8180: resolve the callback's dispatch ONCE. It is invariant for a
+        // fixed closure (see closure/dispatch/direct.rs), and this loop calls
+        // exactly one.
+        let cb_site = crate::closure::DirectCall3::resolve(callback);
         let _tg = DenseThisGuard::bind_undefined();
         let exotic = crate::array::array_iteration_is_exotic(arr);
 
@@ -773,7 +809,7 @@ pub extern "C" fn js_array_some(arr: *const ArrayHeader, callback: *const Closur
                     None => continue,
                 }
             };
-            let result = js_closure_call3(callback, element, i as f64, rooted.receiver());
+            let result = cb_site.call(callback, element, i as f64, rooted.receiver());
             if crate::value::js_is_truthy(result) != 0 {
                 return f64::from_bits(TAG_TRUE);
             }
@@ -816,6 +852,10 @@ pub extern "C" fn js_array_every(arr: *const ArrayHeader, callback: *const Closu
         let length = (*arr).length;
         let scope = crate::gc::RuntimeHandleScope::new();
         let rooted = RootedIterArray::new(&scope, arr);
+        // #8180: resolve the callback's dispatch ONCE. It is invariant for a
+        // fixed closure (see closure/dispatch/direct.rs), and this loop calls
+        // exactly one.
+        let cb_site = crate::closure::DirectCall3::resolve(callback);
         let _tg = DenseThisGuard::bind_undefined();
         let exotic = crate::array::array_iteration_is_exotic(arr);
 
@@ -832,7 +872,7 @@ pub extern "C" fn js_array_every(arr: *const ArrayHeader, callback: *const Closu
                     None => continue,
                 }
             };
-            let result = js_closure_call3(callback, element, i as f64, rooted.receiver());
+            let result = cb_site.call(callback, element, i as f64, rooted.receiver());
             if crate::value::js_is_truthy(result) == 0 {
                 return f64::from_bits(TAG_FALSE);
             }
@@ -857,6 +897,10 @@ pub extern "C" fn js_array_flatMap(
         let length = (*arr).length;
         let scope = crate::gc::RuntimeHandleScope::new();
         let rooted = RootedIterArray::new(&scope, arr);
+        // #8180: resolve the callback's dispatch ONCE. It is invariant for a
+        // fixed closure (see closure/dispatch/direct.rs), and this loop calls
+        // exactly one.
+        let cb_site = crate::closure::DirectCall3::resolve(callback);
         // Root the result across callbacks and pushes (a push both allocates
         // — possibly triggering a moving GC — and may reallocate the array).
         let result_rooted = scope.root_nanbox_f64(f64::from_bits(
@@ -879,7 +923,7 @@ pub extern "C" fn js_array_flatMap(
             let Some(element) = rooted.present(i) else {
                 continue;
             };
-            let mapped = js_closure_call3(callback, element, i as f64, rooted.receiver());
+            let mapped = cb_site.call(callback, element, i as f64, rooted.receiver());
             // Root first: detecting a lazy array may materialize it, and a
             // push in the inner loop can move the callback result's target.
             sub_rooted.set_nanbox_f64(mapped);
@@ -958,6 +1002,10 @@ pub extern "C" fn js_array_reduce(
         let length = (*arr).length as usize;
         let scope = crate::gc::RuntimeHandleScope::new();
         let rooted = RootedIterArray::new(&scope, arr);
+        // #8180: resolve the callback's dispatch ONCE. It is invariant for a
+        // fixed closure (see closure/dispatch/direct.rs), and this loop calls
+        // exactly one.
+        let cb_site = crate::closure::DirectCall4::resolve(callback);
 
         if length == 0 {
             if has_initial != 0 {
@@ -1005,7 +1053,7 @@ pub extern "C" fn js_array_reduce(
                 continue;
             };
             // Spec callback is `(accumulator, currentValue, currentIndex, array)`.
-            let next = js_closure_call4(
+            let next = cb_site.call(
                 callback,
                 acc_rooted.get_nanbox_f64(),
                 element,