diag(runtime): receiver-representation ledger — per-family headerless-receiver counters and a typed native-result classification replacing NR_PTR (371 rows, 0 unknown) - #9973
Closed
proggeramlug wants to merge 2 commits into
Conversation
Instrument headerless receiver producers and dynamic receiver funnels behind the opt-in PERRY_RECEIVER_REPR_DIAG sink. Add explicit native return storage kinds while retaining the existing pointer-boxing lowering, plus checked provider inventory and fail-capable fixtures. Claude-Session: https://claude.ai/code/session_011dhBmdn4vGgNibjo3oZqTo
Record the producer map, corrected executable-row census, sabotage proofs, gate results, and the perrymaster measurement request for campaign PR 1. Claude-Session: https://claude.ai/code/session_011dhBmdn4vGgNibjo3oZqTo
|
Warning Review limit reachedNext included review available in 59 minutes. View limit detailsLimit details: You’ve used all 8 included reviews currently available. You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository. Review configuration: ⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Team Run ID: ⛔ Files ignored due to path filters (1)
📒 Files selected for processing (54)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
proggeramlug
marked this pull request as ready for review
September 7, 2026 20:45
proggeramlug
pushed a commit
that referenced
this pull request
Sep 7, 2026
#9969 makes run_lint_gates.sh derive every command of every run: step (80 now, up from 67). Two of the newly-run gates were red on this train: - addr_class_inventory rejected #9973's two GcHeader casts in hot_diag/receiver_repr.rs. Both are allowlisted with reasons rather than converted: the block is a #[cfg(debug_assertions)] trust-the-tag audit that compares the ownership-derived header against a direct byte-offset read and counts disagreements. Routing the raw side through try_read_gc_header would validate the address first and return None for exactly the implausible cases the audit exists to catch, so the canonical predicate cannot stand in there. - RUSTFLAGS="-D warnings" cargo check --workspace --all-targets rejected #9861's doc comment on a thread_local! macro invocation, which cannot carry one. Moved inside the macro onto the static it describes. This is a warning, not an error, so it only fails under -D warnings — which the driver never replayed before #9969.
Contributor
Author
|
Landed on The three non-passing gates are accounted for: public-baseline is pre-existing on |
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 PR of the receiver-representation campaign (design:
cc-perf-campaign/DESIGN_receiver_representation.mdin the campaign repo). Measurement and metadata only — no JS-visible representation, release-path classifier, wrapper, or native-result boxing changes. Written by codex (0.154.0-alpha.3); all gates run locally.Why
On the best cc bundle the receiver family —
try_read_tracked_gc_header2.46 % self,js_native_call_method_at_site2.90,dispatch_primitive1.79, plus the Buffer/typed-array registry residue — is ≈ 9.9 % of main-thread samples. The audit (REPORT_trust_the_tag.md) showed the direct-header read cannot replace the ownership check while seven headerless POINTER_TAG families exist (integer registry ids, AsyncHook/AsyncResource boxes, leaked global symbols, external buffers, SAB backing, the null stub, bare managed addresses). Before any of them is migrated, the campaign needs (a) counts per family on a real cc reply and (b) an executable classification of every native pointer result so a wrapper migration cannot miss a producer.What changes
PERRY_RECEIVER_REPR_DIAG=<sink>(hot_diag sink, SIGKILL-safe snapshot): one line[receiver-repr-diag] constructed common= fetch= zlib= proxy= timer= text= tui= async_hook= async_resource= symbol_global= external_buffer= sab= null_stub=; observed_old …; observed_wrapped …; bare_managed= invalid_pointer_zero= direct_mismatch=. Producers bump their family at the constructors (common/fetch/zlib registries, proxy, timers, text, TUI, async hooks/resources, global symbols, external buffer, SAB, null stub — file:line in the campaign report); the five dynamic-receiver funnels (gc_pointer_and_type_from_value,dispatch_primitive,object_static_prototype, the field tail,ic_miss) classify only while armed. Every guard is one relaxed atomic load when off;direct_mismatch's header probe isdebug_assertions-only.NR_PTRresult kind is replaced byNR_GCPTR(131 rows),NR_NULLABLE_GCPTR(2),NR_HANDLE_ID(221),NR_FOREIGN_PTR(4: AsyncHook/AsyncResource backing),NR_JS_VALUE(13) — 371 executable declarations (the audit's "372" counted two prose comments and missed one helper row), all materialised through the samenative_handlebranch as before (byte-identical lowering).scripts/native_result_ledger.tsvrecords 322 runtime symbols with class, provider source and Rust return type (the design's 81 unresolved rows: 81/81 resolved);scripts/native_result_ledger.pyfails on any remainingNR_PTR, any table/provider class disagreement, or a stale provider entry, with a--self-test; a path-filtered CI workflow runs it.Tests (named; sabotage performed)
receiver_repr_family_fixtures_move_constructed_and_observed_old(each family's constructed > 0, observed_old > 0, observed_wrapped == 0),receiver_repr_every_family_producer_keeps_its_constructed_bump(pins the bump-site count per producer file; sabotage: one family token changed →left: 1, right: 2),receiver_repr_unarmed_funnels_never_enter_classification(test-only entry counter stays zero with the sink off; sabotage: drop a guard → nonzero),native_result_ledger.py --self-test(a plantedNR_PTRrow and a planted class disagreement are both rejected).Gates (local, macOS arm64)
Runtime release lib suite 3,262 passed / 0 failed; codegen tests 1,450 passed;
cargo build --release -p perry-runtime --features wasm-hostandcargo build --release -p perryok; root-holder and thread-local inventories ok; rustfmt, diff-check, file-size ok.Next
perrymaster runs one 3300-char cc reply with the diag on the best bundle (stage RL): the ledger line prices PR 2 (wrapper creations per reply) and names which headerless families actually reach a dynamic receiver on cc. Per the design, the campaign's early kill (a forced-invariant synthetic A/B) runs before any wrapper migration.
Measured (perrymaster RL, 2026-09-08; main6 bundle relinked against this runtime and stdlib, one 3300-char cc reply,
PERRY_RECEIVER_REPR_DIAG=stderr)constructed common=11 fetch=20 zlib=0 proxy=0 timer=137 text=15 tui=0 async_hook=0 async_resource=0 symbol_global=55 external_buffer=0 sab=0 null_stub=0; observed_old common=4974 fetch=63 timer=4966 text=4949 symbol_global=148 (others 0); observed_wrapped all 0; bare_managed=0; invalid_pointer_zero=0; direct_mismatch=0. Exactly five families reach a dynamic receiver on cc (timers, global symbols, fetch handles, text handles, the common pool); eight never do. 238 constructions per reply price the next PR's wrappers as negligible. The ~4.9 kobserved_oldcounts shared by common/timer/text are the same small-integer payloads claimed by each overlapping range, as the report predicts.https://claude.ai/code/session_011dhBmdn4vGgNibjo3oZqTo