bench(packages): package matrix re-run after the fixing phase

benchmarks/packages/profile/rerun-2026-09-30/callgraph-mongodb-insert_find.md

@@ -0,0 +1,138 @@
+# Package profile: where Perry's instructions go
+
+Perry commit `9e29f59d43`; call chains: `perf record -e instructions:u --call-graph dwarf`; ~12000 samples at n2; every figure is instructions **per iteration** by the two-N method (the n1 profile is subtracted from the n2 profile key by key). Generated by `scripts/package_bench.py profile --callgraph`.
+
+## Ranked root-cause buckets
+
+Share of Perry's **excess over Node** (perry − node instructions/iter). *Equal-weight*: mean over packages (each package counts once, its workloads averaged) — the headline. *Abs-weight*: summed absolute excess (dominated by the most expensive workloads, shown for transparency). Workloads with Perry/Node < 2× are excluded.
+
+Control floor (control perry − node, clamped at 0): **0 instr/iter** — nothing to subtract, so the floor-adjusted ranking is identical.
+
+| # | bucket | equal-weight % of excess | abs-weight % | packages where ≥5% | equal-weight, floor-adjusted |
+|---|---|---:|---:|---|---:|
+| 1 | property lookup slow path / IC miss (own/by-name) (`prop_lookup`) | 34.2 | 34.2 | 1: mongodb | 34.2 |
+| 2 | GC: minor/copying collection (`gc_minor`) | 12.1 | 12.1 | 1: mongodb | 12.1 |
+| 3 | GC: major (mark-sweep) collection (`gc_major`) | 9.1 | 9.1 | 1: mongodb | 9.1 |
+| 4 | dynamic index get/set, Array element access (`dyn_index`) | 6.3 | 6.3 | 1: mongodb | 6.3 |
+| 5 | string ops / UTF-8<->UTF-16 transcoding (`string`) | 5.3 | 5.3 | 1: mongodb | 5.3 |
+| 6 | runtime method dispatch (obj.m() via dispatcher) (`method_dispatch`) | 4.9 | 4.9 | 0:  | 4.9 |
+| 7 | Buffer / typed-array access + buffer-registry probes (`buffer_typedarray`) | 4.2 | 4.2 | 0:  | 4.2 |
+| 8 | reads resolved on the prototype chain (absent keys, inherited data) (`proto_chain_read`) | 3.9 | 3.9 | 0:  | 3.9 |
+| 9 | object/array allocation entry points (`alloc`) | 3.3 | 3.3 | 0:  | 3.3 |
+| 10 | generated code (inline JS) (`generated_code`) | 2.6 | 2.6 | 0:  | 2.6 |
+| 11 | private class members (`private_fields`) | 2.5 | 2.5 | 0:  | 2.5 |
+| 12 | GC: other (safepoint bookkeeping, layout notes) (`gc_other`) | 2.2 | 2.2 | 0:  | 2.2 |
+| 13 | GC write barrier (`write_barrier`) | 1.8 | 1.8 | 0:  | 1.8 |
+| 14 | property write (add/overwrite slow path) (`prop_write`) | 1.4 | 1.4 | 0:  | 1.4 |
+| 15 | accessor / property-descriptor handling (`accessor_descriptor`) | 1.1 | 1.1 | 0:  | 1.1 |
+| 16 | other runtime (`other_runtime`) | 0.9 | 0.9 | 0:  | 0.9 |
+| 17 | raw allocation (arena/malloc, no stronger owner on the chain) (`raw_alloc`) | 0.9 | 0.9 | 0:  | 0.9 |
+| 18 | closures / boxed captures / arguments (`closure_box_args`) | 0.9 | 0.9 | 0:  | 0.9 |
+| 19 | call overhead (safepoints, shadow frames, param guards, dynamic calls) (`call_overhead`) | 0.6 | 0.6 | 0:  | 0.6 |
+| 20 | Promise / async / timers (`promise_async`) | 0.5 | 0.5 | 0:  | 0.5 |
+| 21 | Symbol-keyed properties (`symbol`) | 0.4 | 0.4 | 0:  | 0.4 |
+| 22 | BigInt (`bigint`) | 0.3 | 0.3 | 0:  | 0.3 |
+| 23 | Map / Set (`map_set`) | 0.3 | 0.3 | 0:  | 0.3 |
+| 24 | value tests (typeof/equality/truthiness) (`value_typeof`) | 0.2 | 0.2 | 0:  | 0.2 |
+| 25 | exceptions / try-catch setup (`exceptions`) | 0.1 | 0.1 | 0:  | 0.1 |
+| 26 | regex (Perex) (`regex`) | 0.0 | 0.0 | 0:  | 0.0 |
+| 27 | numeric conversion (ToInt32, fmod, trunc, Math) (`numeric_conv`) | 0.0 | 0.0 | 0:  | 0.0 |
+| 28 | UTF-8 validation / transcoding (no stronger owner on the chain) (`utf8_transcode`) | 0.0 | 0.0 | 0:  | 0.0 |
+| 29 | memcmp/memcpy (no stronger owner on the chain) (`memops`) | 0.0 | 0.0 | 0:  | 0.0 |
+| 30 | number <-> string (`number_string`) | 0.0 | 0.0 | 0:  | 0.0 |
+| 31 | std HashMap/SipHash side tables (no stronger owner on the chain) (`side_table_hash`) | 0.0 | 0.0 | 0:  | 0.0 |
+| 32 | JSON (`json`) | 0.0 | 0.0 | 0:  | 0.0 |
+| 33 | unresolved frames (`unknown`) | 0.0 | 0.0 | 0:  | 0.0 |
+
+## Package × bucket matrix (% of the package's excess)
+
+| package | `prop_lookup` | `gc_minor` | `gc_major` | `dyn_index` | `string` | `method_dispatch` | `buffer_typedarray` | `proto_chain_read` | `alloc` | `generated_code` | `private_fields` | `gc_other` |
+|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
+| mongodb | 34 | 12 | 9 | 6 | 5 | 5 | 4 | 4 | 3 | 3 | 3 | 2 |
+
+## Construct view: runtime entry points called from generated code
+
+Everything beneath each entry up to the next JS frame (all buckets combined), equal-weight % of excess. `<inline JS>` is generated code's own instructions; `<no JS frame>` is work with no JS caller (GC cycles, event loop).
+
+| % of excess | runtime entry |
+|---:|---|
+| 22.7 | `<no JS frame>` |
+| 20.8 | `js_native_call_method_by_id` |
+| 15.6 | `js_object_get_field_by_property_id_f64` |
+| 5.5 | `js_event_emitter_on` |
+| 4.1 | `js_object_set_field_by_property_id` |
+| 3.2 | `js_object_alloc_class_inline_keys_stamped` |
+| 2.8 | `js_object_get_own_field_or_undef` |
+| 2.7 | `<inline JS>` |
+| 2.2 | `node_stream::push_chunk` |
+| 0.9 | `js_class_field_add` |
+| 0.8 | `js_async_step_done` |
+| 0.8 | `promise::microtasks::run_microtasks` |
+| 0.7 | `js_object_assign_one` |
+| 0.6 | `object::field_get_set::ic_miss::get_field_ic_miss_impl` |
+
+## Top (bucket, runtime entry) pairs
+
+Which runtime entry point — the function generated code called — carries each bucket, equal-weight % of excess over packages.
+
+| % of excess | bucket | runtime entry called from generated code |
+|---:|---|---|
+| 13.1 | prop_lookup | `js_object_get_field_by_property_id_f64` |
+| 12.1 | gc_minor | `<no JS frame>` |
+| 9.1 | prop_lookup | `js_native_call_method_by_id` |
+| 9.0 | gc_major | `<no JS frame>` |
+| 4.7 | method_dispatch | `js_native_call_method_by_id` |
+| 4.3 | dyn_index | `js_event_emitter_on` |
+| 3.5 | buffer_typedarray | `js_native_call_method_by_id` |
+| 3.2 | alloc | `js_object_alloc_class_inline_keys_stamped` |
+| 2.8 | prop_lookup | `js_object_get_own_field_or_undef` |
+| 2.7 | generated_code | `<inline JS>` |
+| 2.5 | private_fields | `js_native_call_method_by_id` |
+| 2.5 | string | `js_object_get_field_by_property_id_f64` |
+| 2.5 | prop_lookup | `js_object_set_field_by_property_id` |
+| 1.6 | gc_other | `<no JS frame>` |
+| 1.2 | write_barrier | `js_event_emitter_on` |
+| 1.1 | string | `js_native_call_method_by_id` |
+| 1.0 | prop_write | `js_object_set_field_by_property_id` |
+| 0.9 | prop_lookup | `js_class_field_add` |
+| 0.9 | raw_alloc | `node_stream::push_chunk` |
+| 0.8 | proto_chain_read | `js_async_step_done` |
+| 0.8 | dyn_index | `promise::microtasks::run_microtasks` |
+| 0.7 | other_runtime | `node_stream::push_chunk` |
+| 0.7 | prop_lookup | `js_object_assign_one` |
+| 0.7 | dyn_index | `node_stream::push_chunk` |
+| 0.6 | string | `js_object_set_field_by_property_id` |
+| 0.6 | prop_lookup | `object::field_get_set::ic_miss::get_field_ic_miss_impl` |
+
+Regex bucket split (equal-weight % of excess): execute 0.0, compile -0.0
+
+## Workloads
+
+| workload | perry/iter | node/iter | ratio | unwind→main | package code on stack (≥) | top buckets (% of Perry) |
+|---|---:|---:|---:|---:|---:|---|
+| mongodb/insert_find | 175.70M | 787.6k | 223.1× | 80% | 52% | prop_lookup 34%, gc_minor 12%, gc_major 9%, dyn_index 6% |
+
+### mongodb/insert_find — 175.70M instr/iter (223.1× Node)
+
+Top chains: JS site → runtime entry the generated code called → bucket (dominant leaf):
+
+| instr/iter | % | JS function | JS source | runtime entry | bucket | dominant leaf |
+|---:|---:|---|---|---|---|---|
+| 21.18M | 12.05 | `<no JS frame>` | — | `<no JS frame>` | gc_minor | `::mark_classified` (21%) |
+| 15.72M | 8.95 | `<no JS frame>` | — | `<no JS frame>` | gc_major | `::step_arena_walk` (16%) |
+| 7.47M | 4.25 | `mongodb_src_cmap_wire_protocol_on_data_ts::onData$this` | mongodb/src/cmap/wire_protocol/on_data.ts:41 `const unconsumedPromises = new List<PendingPromises>();` | `js_event_emitter_on` | dyn_index | `js_array_get_f64` (60%) |
+| 5.64M | 3.21 | `mongodb_src_utils_ts::List.push` | mongodb/src/utils.ts:710 `prev: this.head.prev as ListNode<T>,` | `js_object_alloc_class_inline_keys_stamped` | alloc | `object::alloc::object_alloc_class_inline_keys_impl` (61%) |
+| 5.59M | 3.18 | `mongodb_src_cmap_wire_protocol_on_data_ts::10` | mongodb/src/cmap/wire_protocol/on_data.ts:113 `else unconsumedEvents.push(value);` | `js_native_call_method_by_id` | prop_lookup | `object::class_registry::dispatch::call_vtable_method_value` (35%) |
+| 5.53M | 3.15 | `mongodb_src_utils_ts::List.push` | mongodb/src/utils.ts:709 `next: this.head as HeadNode<T>,` | `js_object_get_field_by_property_id_f64` | prop_lookup | `object::native_get::try_data_get_bytes` (34%) |
+| 5.27M | 3.0 | `mongodb_src_utils_ts::List.shift` | mongodb/src/utils.ts:753 `return this.remove(this.head.next);` | `js_native_call_method_by_id` | prop_lookup | `object::class_registry::dispatch::call_vtable_method_value` (27%) |
+| 4.84M | 2.76 | `mongodb_src_utils_ts::List.push` | mongodb/src/utils.ts:710 `prev: this.head.prev as ListNode<T>,` | `js_object_get_field_by_property_id_f64` | prop_lookup | `object::native_get::try_data_get_bytes` (41%) |
+| 4.82M | 2.74 | `mongodb_src_cmap_wire_protocol_on_data_ts::10` | mongodb/src/cmap/wire_protocol/on_data.ts:111 `const promise = unconsumedPromises.shift();` | `js_native_call_method_by_id` | prop_lookup | `object::class_registry::dispatch::call_vtable_method_value` (55%) |
+| 4.73M | 2.69 | `mongodb_src_utils_ts::List.push` | mongodb/src/utils.ts:714 `this.head.prev = newNode;` | `js_object_get_field_by_property_id_f64` | prop_lookup | `object::native_get::try_data_get_bytes` (45%) |
+| 4.34M | 2.47 | `mongodb_src_utils_ts::List.push` | mongodb/src/utils.ts:709 `next: this.head as HeadNode<T>,` | `js_object_set_field_by_property_id` | prop_lookup | `object::shapes::shape_descriptor_by_id` (65%) |
+| 3.33M | 1.9 | `mongodb_src_utils_ts::List.push` | mongodb/src/utils.ts:713 `this.head.prev.next = newNode;` | `js_object_get_field_by_property_id_f64` | prop_lookup | `object::native_get::try_data_get_bytes` (28%) |
+
+Top inclusive runtime functions: `abi_trampoline::call_all_f64_x86_64` 64.33%; `object::class_registry::dispatch::call_vtable_method_value` 64.33%; `js_native_call_value` 63.77%; `js_native_call_method` 63.73%; `js_native_call_method_by_id` 57.26%; `node_stream::event_emitter::emit_stream_event` 53.63%; `closure::dispatch::bound::dispatch_bound_method` 53.55%; `js_run_stdlib_pump` 52.69%
+
+Top self: `object::shapes::shape_descriptor_by_id` 7.76%; `object::native_get::try_data_get_bytes` 4.89%; `js_array_get_f64` 4.11%; `object::class_registry::dispatch::call_vtable_method_value` 3.47%; `buffer::header::is_registered_buffer_slow` 2.69%; `::mark_classified` 2.54%; `__memmove_avx512_unaligned_erms` 2.38%; `__memcmp_evex_movbe` 2.02%
+
+Mechanism view (leaf-first): prop_lookup 35%, gc_minor 12%, gc_major 9%, dyn_index 6%, raw_alloc 5%, method_dispatch 5%