Struct CodeQualityAnalyzer 

pub struct CodeQualityAnalyzer { /* private fields */ }

The code-quality analyzer.

Built once per process. The chain and the cache are both passed in rather than constructed here: they are process-wide resources. The chain in particular carries the concurrency limiter for each provider and the record of which providers have been demoted, so a second chain would both double the in-flight requests against one endpoint and re-discover a dead provider the first one already knew about.

The model and temperature are not duplicated onto this struct. They belong to the provider that ends up answering, which is not known until the chain has run, and a second copy is exactly what lets a request go to one model while the cache key names another.

Implementations

impl CodeQualityAnalyzer

pub fn new(chain: ProviderChain, cache: Cache) -> Self

Build from a provider chain and a shared cache.

Infallible: ProviderChain::new has already rejected an empty or misconfigured chain, so there is nothing left for this to validate. cache is a parameter rather than constructed here so the key stays independent of the Cache root (see Cache::key) and a test can point it at a TempDir.

pub fn chain(&self) -> &ProviderChain

The provider chain, for a caller reporting which providers served.

pub async fn analyze_file(&self, hunks: &[Hunk]) -> AnalysisResult

Analyze one file’s hunks.

Returns an AnalysisResult populated with whatever the file produced: findings, a failure marker, or both. The result is never a bare Vec<Finding>; the failure axis is part of the return type so the caller cannot forget it.

pub async fn analyze_files(&self, by_file: &[Vec<Hunk>]) -> AnalysisResult

Analyze many files concurrently, bounded by the limiter.

Each entry of by_file is one file’s hunks; the limiter bounds the in-flight requests, so we spawn them all and let it queue. The per-file results are merged with AnalysisResult::merge.