perf: eliminate redundant read_dispair calls in try_into_call

[thedavidmeister]

Summary

try_into_call previously called read_dispair three times — once inline and once inside each try_parse_rainlang call — totalling 12 sequential RPC round-trips just to read 4 immutable deployer addresses. At 300ms RTT (remote RPC) that is ~3.6s of avoidable latency.

Changes

• Add parse_bytecode_with_dispair: the parsing body extracted from try_parse_rainlang, takes a pre-fetched DISPaiR (no additional dispair reads).
• try_parse_rainlang is now a thin wrapper: read_dispair + parse_bytecode_with_dispair. All existing tests remain unchanged.
• try_into_call calls read_dispair once, clones for the first parse, moves for the second parse, and extracts interpreter/store upfront.

Result

RPC calls per generate_deposit_and_add_order_calldatas (dispair portion):

Before	After
12 (3× read_dispair)	4 (1× read_dispair)

Total for a single deposit+addOrder: 14 → 6 sequential eth_call round-trips, saving ~2.4s at 300ms RTT.

Test

Adds assertions to test_into_add_order_call verifying both config.evaluable.interpreter/store and tasks[0].evaluable.interpreter/store match the expected interpreter/store addresses (and match each other). This mutation-validates that both evaluables receive the correct dispair from the single shared read.

Closes #2776

Co-Authored-By: Claude noreply@anthropic.com

Summary by CodeRabbit

• Bug Fixes
  • Reduced unnecessary network calls when preparing orders, which should make order creation faster and more reliable.
  • Ensured both the main order logic and follow-up action use the same resolved interpreter and store values.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: dc807778-09af-4181-b68a-feb010f87fa0

📥 Commits

Reviewing files that changed from the base of the PR and between e4ca835 and 3a43542.

📒 Files selected for processing (3)

• crates/common/src/raindex_order_builder/order_operations.rs
• crates/common/src/raindex_order_builder/select_tokens.rs
• packages/raindex/test/js_api/builder.test.ts

---

📝 Walkthrough

Walkthrough

Order preparation now reuses a single DISPaiR fetch, carries token decimals through deposit processing, and reuses complete recorded token metadata before on-chain lookup. Tests validate shared evaluable addresses, approval scaling, and token-info fetch reuse.

Changes

Order preparation RPC reuse

Layer / File(s)	Summary
Reuse DISPaiR during rainlang parsing crates/common/src/add_order.rs	Parsing accepts a pre-fetched DISPaiR; try_into_call reuses one instance for main and post-task evaluables, with tests confirming shared interpreter and store addresses.
Carry deposit decimals into approval generation crates/common/src/raindex_order_builder/order_operations.rs, packages/raindex/test/js_api/builder.test.ts	Deposit maps retain token decimals, and approval generation uses them for allowance comparison and calldata encoding without separate decimals RPC calls.
Reuse recorded token metadata crates/common/src/raindex_order_builder/select_tokens.rs	Token selection reuses complete YAML metadata for the same network and address, while incomplete or network-mismatched records still trigger on-chain metadata lookup; tests cover these cases.

Estimated code review effort: 3 (Moderate) | ~25 minutes

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and accurately describes the primary change in the PR: removing redundant read_dispair calls in try_into_call.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
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.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch 2026-06-29-issue-2776-dispair-cache

Warning

There were issues while running some tools. Please review the errors and either fix the tool's configuration or disable the tool if it's a critical failure.

🔧 ESLint

If the error stems from missing dependencies, add them to the package.json file. For unrecoverable errors (e.g., due to private dependencies), disable the tool in the CodeRabbit configuration.

ESLint install timed out. The project may have too many dependencies for the sandbox.

---

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.}

[thedavidmeister]

🤖 ai:vetter
Reviewed e4ca835: ready — single read_dispair call cloned for first parse moved for second; parse_bytecode_with_dispair takes
cost 480 — perf change in Rust async RPC code

[thedavidmeister]

Rework note (human): incomplete coverage of #2776. The issue names three redundant RPC-fetch sources; this PR dedups only read_dispair (12→4). Still duplicated: set_select_token fetches token_info twice when the same token is selected for both input and output slots, and generate_approval_calldatas refetches decimals that set_select_token already has. Dedup those two as well to close #2776.

[thedavidmeister]

🤖 ai:producer
Producer note: rework executed — (1) set_select_token now reuses a complete same-network yaml token record instead of refetching token_info: same token selected for input+output slots = 1 multicall instead of 2 (test_set_select_token_same_address_fetches_once, mock hit-count pinned; cross-network/incomplete records never shortcut, re-selection reuses its own record). (2) generate_approval_calldatas reads decimals from the deposits map instead of a per-token decimals() RPC: 1 -> 0, only the allowance read remains (test_generate_approval_calldatas_reuses_deposit_token_decimals, 0-hit + exact-amount pinned). Branch merged current with main; raindex_common suite locally green (1161 passed); all 7 mutants (reintroduced duplicate fetches, dropped guards, reordered removal, corrupted decimals) killed. With the earlier read_dispair dedup this covers all three fetch sources named in #2776. Remaining CI reds are the known raindex main-state failures.

[thedavidmeister]

🤖 ai:producer
Producer note: post-rework CI triage. test-js-bindings was the one PR-caused red — the approval-calldata dedup left sequence-based decimals mocks stale in builder.test.ts, so the allowance reads consumed the 0x12 decimals responses (allowance became 18 wei, emitting an extra approval: 2 != 1). Fixed by 1bbbc20 fix(ci): test-js-bindings drop stale decimals mocks after approval-calldata dedup [3b-attempt]. The remaining reds (rainix-sol/static, rainix-sol/test, rs-static) are red inherited from main (rainix no-skips vm.skip gate on 5 pre-existing test/lib/deploy files + missing 0.1.13 soldeer constant pins), not PR-caused; fixed by #2804 (green, ai:ready); greens on branch update once it lands. flag-blocked-on refused (human decision present), so classification posted as comment.

[thedavidmeister]

🤖 ai:producer
Producer note (state change — correction to the prior note): the [3b-attempt] fix 1bbbc20 did NOT green test-js-bindings — the check re-ran at that exact head (run 29226076755, 2026-07-13) and still fails at test/js_api/builder.test.ts:1451 with the same signature (expected 1 approval call, received 2). One fix attempt is spent, so back-off forbids a second push. flag-blocked-infra was refused (human decision present), so this stays parked under human:reject for human re-adjudication. The other three reds remain the main-state trio fixed by #2804.

[thedavidmeister]

🤖 ai:vetter
Reviewed 1bbbc20: reject — closes #2776 attempt — the approval-calldata decimals dedup produces an unexplained extra approval in test/js_api/builder.test.ts:1451 (2 emitted, 1 expected) that survived the producer's own [3b-attempt] mock fix at head 1bbbc20; until the double-approval is explained as mock-sequence artifact vs real regression, the dedup in generate_approval_calldatas is unproven; read_dispair + set_select_token dedups look sound
cost 520 — RPC dedup, unexplained approval regression

[thedavidmeister]

🤖 ai:vetter
Reviewed 3a43542: ready — closes #2776 — all three enumerated RPC-dedup sites covered (read_dispair 3x-to-1 in try_into_call, set_select_token same-address reuse network-scoped, approval decimals from deposits map) each pinned by discriminating tests (wrong-value decimals mock asserted 0 hits); prior reject's double-approval explained and fixed as mock-sequence artifact via address-pinned allowance mocks at yaml decimals; stale-record removal now after address parse so invalid re-select no longer destroys the record
cost 545 — multi-site RPC dedup, Rust + mocks

[thedavidmeister]

🤖 ai:vetter
Reviewed c4da940: ready — Closes #2776 — re-vet at merge-update head c4da940 (diff unchanged from approved 3a43542): all three enumerated RPC-dedup sites covered (read_dispair 3x→1 in try_into_call, network-scoped same-address token-info reuse, approval decimals from deposits map), each pinned by hit-count discriminating tests incl. the wrong-value decimals mock at 0 hits; double-approval was a mock-sequence artifact, fixed with address-pinned mocks; CI green at this head.
cost 542 — multi-site RPC dedup, Rust + mocks