[shaurya703]

Summary

parse_context_limit_from_error panics when a multi-byte character comes right before the number in the "<N> maximum context length" (Mistral) shape.

Bug. Reproduction on current main:

parse_context_limit_from_error("too large for model with\u{a0}32768 maximum context length");
// panicked at crates/tinyinference-llm/src/failure/context_limit.rs:138:25
// (byte index is not a char boundary)

The same happens when a CJK character precedes the number ("…模型32768 maximum context length"). This runs on provider overflow error bodies through model::discover::record_overflow_error, so a provider error with that wording crashes the caller instead of recording the window.

Root cause. trailing_number finds the last non-digit character with rfind and starts the digit run at index + 1. That is one byte past the start of the character, so a no-break space (2 bytes) or CJK character (3 bytes) leaves the slice start inside it.

Fix. Walk char_indices() from the end and start at index + character.len_utf8().

Related issue

None. Found while reading the overflow-error parsers.

API or behavior changes

None. Messages that used to panic now return the stated limit.

Validation

• cargo fmt --all -- --check: clean
• cargo clippy -p tinyinference-llm --all-targets --all-features -- -D warnings: clean
• cargo test -p tinyinference-llm --all-features: all pass (873 lib tests). The new test panics without the fix.

Tests

failure::context_limit::tests::parses_trailing_limit_after_a_multibyte_character covers a no-break space and a CJK character before the number, and expects Some(32_768) for both.

Documentation

Not needed.

Checklist

• The change is focused on one logical change
• No new #[allow(...)], #[ignore], or relaxed lints
• No secrets, tokens, or .env contents in the diff or the description

[tinysweeper]

Tiny Sweeper review

Reviewed head: 9bdca33c07d5b0c4b1d33795054dde5ac16a52c7

Verdict: Ready for maintainer review · Priority: none
Coverage: Reviewed 2 of 2 files.
Pull request purpose (title): fix(failure): read a…

Diagnostics and supporting details

Full summary

Pull request purpose (title): fix(failure): read a trailing context limit after a multi-byte character. Tiny Sweeper completed its review; deterministic results follow.

Review snapshot

Change surface	Files	Review signal	Count
Production	1	Active findings	0
Tests	1	Noted findings	0
Documentation	0	Resolved findings	0
Configuration	0	Pending checks/questions	0

Completeness: Complete
Test assessment: No supported feature-to-test mapping was available; this does not mean tests are absent or passed.

What changed

`trailing_number` in the context-limit parser replaces its `rfind`-based byte offset with a reverse `char_indices()` scan that advances by the matched character's `len_utf8()`, preserving the `map_or(0, ...)` fallback for all-digit/comma input and leaving the downstream `trim_start_matches(',')`/parse handling unchanged. Messages that previously panicked on a non-char-boundary slice now return the stated limit. A new unit test in the sibling tests file exercises a no-break space and a CJK character before the number, both expected to parse as `Some(32_768)`.

Features

• Modified — Panic-free trailing context-limit parsing after multi-byte characters: Provider overflow error bodies with a no-break space or CJK character before the limit number no longer panic the caller (via `model::discover::record_overflow_error`) and instead yield the parsed window; behavior is unchanged for other inputs, including all-digit/all-comma strings. (crates/tinyinference-llm/src/failure/context_limit.rs fn first_number(text: &str, window: usize) -> Option<u64> {, crates/tinyinference-llm/src/failure/context_limit.rs)

Tests

No supported feature-to-test mapping was produced. Test execution is not inferred.

Findings

No active actionable findings.

Before merge

None.

Agent review details

critique

• Conclusion: Success
• Scope reviewed: all assigned evidence
• Positive: It noted the tests feed ASCII-lowercased text, while the code still handles arbitrary multi-byte input.
• Lane summary: The change fixes a real byte-boundary panic in `trailing_number`: `rfind` returns a byte index of the matched non-digit character's start, so `index + 1` could land inside a multi-byte character and panic on slicing. Replacing it with `char_indices().rev().find(...)` and adding `character.len_utf8()` is correct, the `map_or(0, ...)` case (all digits/commas) is preserved, and the two new tests exercise the no-break-space and CJK cases. Note both tests feed `lower` (ASCII-lowercased) text, so the multibyte character preceding the digits is the no-break space / CJK run — the code itself still handles arbitrary multibyte correctly. _The code index is behind this pull request (indexed at `b3a7c5fb6ca2`), so retrieved context may be out of date._

security

• Conclusion: Success
• Scope reviewed: all assigned evidence
• Positive: The security lane found that `begin` always lands on a char boundary at or after the last non-digit, keeping the slice valid and the downstream trim/parse handling consistent, and introduced no security-relevant sink or untrusted-input issue.
• Lane summary: The change rewrites `trailing_number` to step past a whole multi-byte character instead of one byte, fixing a real panic (a no-break space or CJK character before the limit would slice inside a char boundary). The reverse `char_indices` scan, `map_or(0, ...)` fallback, and subsequent `trim_start_matches(',')`/`parse_grouped` handling are consistent: `begin` always lands on a char boundary at or after the last non-digit, so the slice is valid. The two new tests exercise exactly the fixed inputs and are unit tests in the sibling `*_tests.rs` file. No security-relevant sink, untrusted-input path, or repository-rule violation is introduced. _The code index is behind this pull request (indexed at `b3a7c5fb6ca2`), so retrieved context may be out of date._

tests

• Conclusion: Success
• Scope reviewed: all assigned evidence
• Positive: The tests lane reported nothing reportable, noting the new test exercises inputs that panicked before the change and parse correctly now.
• Lane summary: The change fixes a real panic: `rfind` returns a byte index of the non-digit character, so `index + 1` could land inside a multi-byte character and `trimmed[begin..]` would panic. The new `char_indices` walk adds `len_utf8()`, and the new test exercises both a no-break space and a CJK character — inputs that panicked before and parse correctly now. I found nothing reportable. _The code index is behind this pull request (indexed at `b3a7c5fb6ca2`), so retrieved context may be out of date._

commits

• Conclusion: Success
• Scope reviewed: all assigned evidence
• Positive: The commits lane reported no sensitive content in committed files.
• Lane summary: No sensitive content in committed files. _The code index is behind this pull request (indexed at `b3a7c5fb6ca2`), so retrieved context may be out of date._

description

• Conclusion: Success
• Scope reviewed: all assigned evidence
• Positive: The description lane found the description matches the diff, explaining the panic, the `char_indices()`/`len_utf8()` fix, and the added test, with no unmentioned functional changes.
• Lane summary: The description matches the diff: it explains the byte-index panic in `trailing_number`, the fix using `char_indices()` and `len_utf8()`, and the added test covering a no-break space and a CJK character, with no unmentioned functional changes. One wording matter: the test is named `parses_trailing_limit_after_a_multibyte_character`, which is accurate for a trailing number even though the character is leading it — not a defect. There are no reportable code defects in the changed lines. _The code index is behind this pull request (indexed at `b3a7c5fb6ca2`), so retrieved context may be out of date._

Evidence and run details

• Models: ladder/vectors-oai3, deepseek-v4.1-flash
• Spend: $0.000096
• Tokens: 53638 input · 1621 output · 8704 cached · 150 embedding
• Updated: 2026-10-26T06:25:56Z

Head	State	Pass summary
9bdca33c07d5	ready for maintainer review	0 active finding(s), 0 resolved finding(s) (at 2026-10-26T06:25:56Z)

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

Run configuration

• Source mode: Auto
• Repository: Unavailable
• Branch: Unavailable
• File: No configuration file from this repository applied
• Commit: Unavailable
• Layers: Organization UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: 4f2c1293-d5ea-4af2-a8cf-c26fc7426047

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

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

[tinysweeper]

tinysweeper found nothing blocking. Approving.

Diagnostics

Total spend: $0.0001