Struct AnalysisResult 

pub struct AnalysisResult {
    pub findings: Vec<Finding>,
    pub failed_files: BTreeMap<PathBuf, FailureReason>,
    pub dropped_out_of_range: usize,
}

What one analysis pass produced.

findings and failed_files are independent axes: a file can contribute findings AND be unanalyzed (a truncated response gives a partial list). Reporting the findings while forgetting the failure is the exact bug this type exists to prevent.

Fields

findings: Vec<Finding>

Findings the analyzer could attribute to a real line of code.

failed_files: BTreeMap<PathBuf, FailureReason>

Files that could not be fully analyzed, with the reason. A BTreeMap because two passes over the same file set must UNION, never sum — summing counts one unreachable endpoint twice.

dropped_out_of_range: usize

Findings discarded because their line was not in the payload’s valid_lines. Counted rather than silently dropped, so the drop is observable.

Implementations

impl AnalysisResult

pub fn failed(path: PathBuf, reason: FailureReason) -> Self

One file, one failure, no findings.

The shape was hand-assembled at four call sites - default(), insert, return - each of which independently had to know that findings and dropped_out_of_range stay at their defaults. Forgetting the insert at any one of them reports an unanalyzed file as clean, which is the single failure this whole type exists to prevent, so it gets a constructor.

pub fn merge(&mut self, other: AnalysisResult)

Fold other into self: findings concatenate, failed_files unions, dropped_out_of_range sums.

The merge semantics let a caller combine per-file and per-layer results without losing the failure signal.

On a key collision in failed_files, the first reason wins. A file failing twice is still one failure, and the two reasons are not meaningfully combinable - the first one is at least specific to the file, while a hypothetical last-wins policy would let a later analyzer overwrite a more informative first reason with a generic one.

pub fn has_failures(&self) -> bool

True when any file went unanalyzed.

The caller maps this to process exit 2: “could not analyze” is distinct from both “clean” (exit 0) and “found issues” (exit 1), because a gate that cannot distinguish them rubber-stamps the day the LLM endpoint goes down.

Trait Implementations

impl Clone for AnalysisResult

fn clone(&self) -> AnalysisResult

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for AnalysisResult

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for AnalysisResult

fn default() -> AnalysisResult

Returns the “default value” for a type.

impl Eq for AnalysisResult

impl PartialEq for AnalysisResult

fn eq(&self, other: &AnalysisResult) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl StructuralPartialEq for AnalysisResult