fix(grpc): preserve oversized integer tool-call arguments digit-exact

[pallasathena92]

Description

Problem

process_tool_call_arguments parses each assistant tool call's arguments string into a JSON value before chat-template rendering. serde_json represents integers outside the i64/u64 range as f64, so an argument like {"id": 123456789012345678901234567890} silently re-renders as 1.2345678901234568e29. The model sees the corrupted spelling in its rendered history and echoes a wrong id back to the tool on the next call.

Solution

Pre-scan the arguments string with a minimal string-state lexer and wrap only the integer literals neither i64 nor u64 can hold in quotes, so they survive the parse digit-exact as JSON strings. The guard is deliberately narrow:

• in-range integers (up to u64::MAX / down to i64::MIN) stay native numbers — zero behavior change;
• float-shaped tokens (fraction or exponent) pass through untouched — f64 semantics are JSON-standard for those;
• digits inside string values are never touched (the lexer tracks in-string/escape state);
• non-canonical tokens (leading zeros, stray operators) are left alone so malformed input still fails in the parser, citing the original text.

Rendering such an argument quoted rather than bare is the accepted trade-off: values at that magnitude are invariably identifiers that must round-trip verbatim, and exact-but-quoted beats bare-but-corrupted.

Changes

• model_gateway/src/routers/grpc/utils/chat_utils.rs: add quote_unrepresentable_integers (Cow-returning pre-scan; borrows when no rewrite is needed) and is_canonical_json_integer; apply the guard in process_tool_call_arguments.

Test Plan

Six new unit tests in chat_utils:

• >u64::MAX positive and <i64::MIN negative integers survive digit-exact (top level and nested containers);
• u64::MAX, i64::MIN, ordinary ints, and floats stay native numbers;
• digit runs inside string values (including next to escaped quotes) are untouched;
• malformed arguments ({bad, [1-2], leading-zero integers) still error, citing the original text;
• untouched inputs take the Cow::Borrowed path.

cargo clippy -p smg --all-targets clean; cargo test -p smg --lib chat_utils 30/30.

Checklist

• Format your code: make fmt
• Run lint checks: targeted cargo clippy -p smg --all-targets -- -D warnings
• Add unit tests for new functionality
• Update documentation if needed (n/a — internal fix)

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 08e73287-a554-4baa-b49e-aa1f28b0eaca

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

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.}

[claude]

Clean, well-tested fix. The lexer correctly handles string escaping, float-shaped tokens, canonical-form validation, and the zero-copy borrowed path. Test coverage is thorough across boundaries (u64::MAX, i64::MIN), nested containers, escaped-quote edge cases, and malformed input fallthrough. No issues found.