refactor(cu-tracker): take explicit instruction name

[dev-jodee]

Summary

• Make the CU tracker generic: record_cu(name, cus) instead of decoding the program instruction enum inside the tracker, so the same cu_tracker.rs can be shared across program repos (rewards/escrow/vault).
• Move the SubscriptionsInstruction::from_bytes name decode into the build_and_send_transaction test glue (the single recording site) — zero call-site churn.
• Behavior and the cu_report.md output format are unchanged.

Why

Part of standardizing CU benchmarking across program repos. The shared reusable GitHub Action consumes the cu_report.md table; the tracker that produces it should be program-agnostic (explicit name) so non-Pinocchio repos can reuse it.

Test Plan

• cargo check -p tests-subscriptions --tests passes.
• cargo fmt -p tests-subscriptions --check clean.
• Full CU_REPORT=1 just integration-test run regenerates the same report (requires the program .so); change is a behavior-equivalent relocation of the name decode.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
solana-subscriptions-program	Ready	Preview, Comment	Jun 8, 2026 8:43pm

[greptile-apps]

Greptile Summary

This PR makes the CU tracker program-agnostic by replacing record_transaction(result, ix) (which internally decoded SubscriptionsInstruction) with record_cu(name, cus), moving the name-decode responsibility into build_and_send_transaction in test_helpers.rs. The benchmark workflow's inline PR-comment shell script is replaced by the reusable solana-developers/github-actions/cu-benchmark action.

• cu_tracker.rs: Drops all dependencies on litesvm, solana_instruction, and SubscriptionsInstruction; collapses min/max/avg statistics into a single minimum-CU value; fixes the long-standing LAMPOSTS_PER_SOL typo.
• test_helpers.rs: Moves SubscriptionsInstruction::from_bytes decode to the single recording site; behaviour is equivalent to before.
• benchmark.yml: Delegates CU-report comment posting to the pinned external action, but drops contents: read from the permissions block, which will prevent actions/checkout from completing.

Confidence Score: 4/5

The Rust-side refactor is clean and correct, but the workflow change drops contents: read from its permissions block, which will cause the actions/checkout step to fail on every benchmark run until restored.

The benchmark workflow now omits contents: read while still using actions/checkout@v6. Any permission not listed in an explicit permissions: block defaults to none, so the checkout step will receive a 403 and the CI job will never reach the benchmark or report steps. The Rust refactor itself — decoupling the tracker, moving the name decode, and fixing the typo — is straightforward and correct.

.github/workflows/benchmark.yml needs contents: read added back to the permissions block.

Important Files Changed

Filename	Overview
.github/workflows/benchmark.yml	Replaces inline PR-comment shell script with a reusable cu-benchmark action; removes contents: read permission which will break the actions/checkout step, and widens fetch-depth to full history.
tests/integration-tests/src/utils/cu_tracker.rs	Decouples the tracker from SubscriptionsInstruction; changes public API to record_cu(name, cus); collapses min/max/avg stats into a single minimum value; fixes LAMPOSTS_PER_SOL typo.
tests/integration-tests/src/utils/test_helpers.rs	Moves SubscriptionsInstruction::from_bytes name decode from the tracker into build_and_send_transaction; logic is equivalent to the previous record_transaction call.

Sequence Diagram

sequenceDiagram
    participant Test as Integration Test
    participant TH as test_helpers.rs
    participant SI as SubscriptionsInstruction
    participant CT as cu_tracker.rs
    participant FS as cu_report.md

    Test->>TH: build_and_send_transaction(litesvm, signers, payer, ix)
    TH->>TH: litesvm.send_transaction(tx)
    TH->>TH: is_tracking_enabled()?
    alt CU_REPORT set
        TH->>SI: "from_bytes(&ix.data) → name"
        TH->>CT: record_cu(name, meta.compute_units_consumed)
        CT->>CT: tracker.record(name, cus)
    end
    TH-->>Test: TransactionResult

    Note over CT,FS: On process exit (dtor)
    CT->>CT: compute_stats() → min CUs per instruction
    CT->>FS: write_to_file("cu_report.md")

Loading

_{Reviews (9): Last reviewed commit: "ci: drop unused CU prettier-ignore (repo..." | Re-trigger Greptile}

[github-actions]

Compute Unit Report

Instruction	Samples	CUs	Est Cost (Low) [SOL]	Est Cost (Med) [SOL]	Est Cost (High) [SOL]	Δ Avg vs main
cancel_subscription	22	1720	0.000005000	0.000005068	0.000005860
close_subscription_authority	10	1803	0.000005000	0.000005072	0.000005901
create_fixed_delegation	41	3520	0.000005001	0.000005140	0.000006760
create_plan	97	3436	0.000005001	0.000005137	0.000006718
create_recurring_delegation	30	3555	0.000005001	0.000005142	0.000006777
delete_plan	9	359	0.000005000	0.000005014	0.000005179
init_subscription_authority	174	6226	0.000005001	0.000005249	0.000008113
resume_subscription	3	1723	0.000005000	0.000005068	0.000005861
revoke_delegation	19	255	0.000005000	0.000005010	0.000005127
subscribe	32	6485	0.000005001	0.000005259	0.000008242
transfer_fixed	9	5479	0.000005001	0.000005219	0.000007739
transfer_recurring	20	5591	0.000005001	0.000005223	0.000007795
transfer_subscription	10	5838	0.000005001	0.000005233	0.000007919
update_plan	22	424	0.000005000	0.000005016	0.000005212

🔺 increase · 🔻 decrease · – unchanged · 🆕 new · 🗑 removed (vs main)

Generated: 2026-06-08