bench: cover four inbound/group hot paths for CodSpeed baselines

[jlucaso1]

Why

Establishes CodSpeed baselines for four receive-path hot paths that had no benchmark coverage, so a flamegraph can then point at where to optimize. No production code changes — benchmarks only. (#856 already covered the two biggest inbound gaps, plaintext decode and the appstate index-MAC dedup; this fills the rest.)

What it adds

• bench_parse_message_info (message_utils_benchmark) — the stanza→MessageInfo metadata parse that runs once per inbound message before any Signal work, across four shapes (DM, LID group, status broadcast, self-sent). The input is a marshal round-trip decoded back into an OwnedNodeRef, so JID attributes arrive wire-typed (ValueRef::Jid) exactly as the decoder hands them to the receive loop — not string-parsed via Jid::from_str, which production never does. Each shape carries an <enc> child so the get_optional_child scans run.
• bench_decode_record (appstate_benchmark) — the per-mutation app-state decode (AES-256-CBC + content HMAC-SHA512 + index HMAC-SHA256 + prost + JSON index parse), up to ~1000× per resume patch. process_patch covered the whole loop but never isolated this inner per-record cost. Two shapes (5-part STAR index, contact-name) built via encode_record so MAC validation runs the success path.
• bench_group_out_of_order_decrypt_worst_case (libsignal_benchmark) — a ~2000-deep skipped-key backlog, the full out-of-order group decrypt. Honest framing in the doc: the backlog-sized SenderKeyRecord clone in load_sender_key and the signature check are the bulk of it, with the O(n) remove_sender_message_key scan a smaller slice — a baseline, not a scan-isolating microbench. (The clone showing up this large is itself a lead worth a flamegraph look.)
• bench_process_sender_key_distribution_message (libsignal_benchmark) — SKDM ingest into a fresh store, the first-group-message-per-sender cost, previously executed only in other benches' setup.

Precision

Every bench black_boxes its inputs and fully observes its output (the thin-LTO dead-code-elimination lesson from the reporting-token bench), and keeps all expensive setup in with_inputs (untimed). Drafted in parallel and then adversarially verified via a workflow — that pass caught and fixed the two issues above (the string-typed vs wire-typed JID realism gap in parse_message_info, which had overstated it ~25-43% and skewed the profile toward a from_str the receive loop never runs; and the misleading scan-only framing of the group OOO bench). No Cargo.toml changes (all bench targets already registered); cargo clippy --workspace --all-targets clean.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Repository UI

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: 0296d7ab-170c-499d-aef0-72c0814643a4

📥 Commits

Reviewing files that changed from the base of the PR and between b956722 and 2928bd0.

📒 Files selected for processing (3)

• wacore/appstate/benches/appstate_benchmark.rs
• wacore/benches/message_utils_benchmark.rs
• wacore/libsignal/benches/libsignal_benchmark.rs

---

📝 Walkthrough

Summary by CodeRabbit

• Tests
  • Enhanced performance benchmarking for core operations across inbound record processing, message metadata parsing, and cryptographic functions.

Walkthrough

This PR adds three independent benchmark suites to measure critical inbound data path performance: app state record decoding ("star" and "contact" shapes), message metadata parsing (DM, group, status broadcast, self-sent), and libsignal group decryption (worst-case out-of-order backlog plus SKDM ingest). All use Divan parameterization.

Changes

Inbound Data Path Benchmarking

Layer / File(s)	Summary
App state record decode benchmark wacore/appstate/benches/appstate_benchmark.rs	setup_record constructs encoded SyncdRecord test fixtures for "star" and "contact" shapes, and bench_decode_record measures per-shape decode cost. Import list is reformatted for readability.
Message info metadata parsing benchmark wacore/benches/message_utils_benchmark.rs	msg_info_node builds parameterized <message> stanzas with addressing-mode attributes (PN, LID, status, self-sent). bench_parse_message_info measures parse latency for each addressing mode using marshalled binary input.
Libsignal group and SKDM benchmarks wacore/libsignal/benches/libsignal_benchmark.rs	setup_group_out_of_order_worst_case pre-encrypts backlog and selects tail ciphertext for worst-case linear scanning, measured by bench_group_out_of_order_decrypt_worst_case. setup_skdm_ingest and bench_process_sender_key_distribution_message measure sender-key distribution message ingestion. Import list is reformatted.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~12 minutes

Possibly related PRs

• oxidezap/whatsapp-rust#224: Established libsignal benchmark infrastructure that this PR extends with additional group decryption and SKDM scenarios.
• oxidezap/whatsapp-rust#828: Migrated benchmark harness to Divan, which provides the parameterization and sampling framework used throughout this PR.

Suggested labels

performance

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title accurately summarizes the main change: adding benchmark coverage for four inbound/group hot paths to establish CodSpeed baselines.
Description check	✅ Passed	The description is comprehensive and directly related to the changeset, clearly explaining the motivation, what each benchmark covers, and implementation precision details.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch bench/inbound-hotpaths

---

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.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[cubic-dev-ai]

No issues found across 3 files

_{Re-trigger cubic}

[codspeed-hq]

Merging this PR will improve performance by 28.6%

⚡ 2 improved benchmarks
❌ 1 (👁 1) regressed benchmark
✅ 154 untouched benchmarks
🆕 16 new benchmarks

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
⚡	Memory	bench_group_decrypt_message	2.6 KB	1.6 KB	+64.02%
⚡	Memory	bench_group_encrypt_message	2.4 KB	1.6 KB	+49.21%
🆕	Memory	bench_decode_record[contact]	N/A	385 B	N/A
🆕	Memory	bench_decode_record[star]	N/A	484 B	N/A
🆕	Memory	bench_parse_message_info[dm_pn]	N/A	31 B	N/A
🆕	Memory	bench_parse_message_info[group_lid]	N/A	33 B	N/A
🆕	Memory	bench_parse_message_info[self_sent]	N/A	21 B	N/A
🆕	Memory	bench_parse_message_info[status_broadcast]	N/A	21 B	N/A
🆕	Memory	bench_group_out_of_order_decrypt_worst_case	N/A	126.4 KB	N/A
🆕	Memory	bench_process_sender_key_distribution_message	N/A	2.3 KB	N/A
🆕	Simulation	bench_decode_record[contact]	N/A	40.5 µs	N/A
🆕	Simulation	bench_decode_record[star]	N/A	42.7 µs	N/A
🆕	Simulation	bench_parse_message_info[dm_pn]	N/A	9.1 µs	N/A
🆕	Simulation	bench_parse_message_info[group_lid]	N/A	9.6 µs	N/A
🆕	Simulation	bench_parse_message_info[self_sent]	N/A	8.9 µs	N/A
🆕	Simulation	bench_parse_message_info[status_broadcast]	N/A	9.5 µs	N/A
🆕	Simulation	bench_group_out_of_order_decrypt_worst_case	N/A	651.8 µs	N/A
🆕	Simulation	bench_process_sender_key_distribution_message	N/A	93.7 µs	N/A
👁	Simulation	bench_decrypt_with_previous_session	5 µs	5.7 µs	-13.1%

Tip

Curious why this is faster? Comment @codspeedbot explain why this is faster on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing bench/inbound-hotpaths (2928bd0) with main (b956722)}