Struct GraphContext 

pub struct GraphContext {
    pub items: Vec<ContextItem>,
    pub dropped_items: usize,
}

The graph context handed to one file’s review.

GraphContext::none is the diff-only arm, and build_prompt renders no context section for it, so the baseline prompt carries no vestigial heading promising something that is not there.

What is in it, and why those and not the rest

The menu roteiro review already computes is governing ADRs, callers, callees, blast radius and authored drift. The arm takes two of those and records the omission of the others as a decision:

• Governing ADR and blueprint sections (authored). The one thing a per-file reviewer structurally cannot obtain: a decision written in another file that the code under review contradicts. This is contract-drift’s defining shape and the authored layer’s whole purpose.
• The file’s own doc surface outside the diff (derived). build_prompt instructs the model to stay silent unless both halves of a conflict are visible, and a -U3 diff shows at most three lines either side. A doc comment at the top of a file and the code that betrays it 700 lines down are never both in the hunks. The graph holds each symbol’s doc comment, so this makes the promise-half visible without pasting the file.
• Callers, callees and blast radius are deliberately excluded. Measured on this repository a single changed symbol carries dozens of caller keys, and their bodies would dominate the prompt — the failure mode above, bought knowingly. Recorded here so the absence reads as a decision rather than an oversight, and so a later arm that adds them knows it is changing a pre-registered variable.

Fields

items: Vec<ContextItem>

The items, in the order they are rendered.

dropped_items: usize

Items dropped by GraphContext::fit to stay inside the cap.

Counted rather than silently absorbed, for the same reason Prompt::dropped_tokens is: a run that quietly shed the context it was measuring would report the graph arm as having been tested when it was partly the diff-only arm wearing its name.

Implementations

impl GraphContext

pub fn none() -> Self

No context — the diff-only arm.

pub fn is_empty(&self) -> bool

Whether any context is present.

pub fn tokens(&self) -> usize

Estimated tokens this context will cost, by estimate_tokens over the text build_prompt actually renders — label, provenance tag and body.

pub fn fit(items: Vec<ContextItem>, diff_tokens: usize) -> Self

Drop whole items, lowest-priority last-first, until the context fits the cap for a diff of diff_tokens.

Whole items, never a truncated one. A half-quoted ADR is worse than no ADR: it reads as a complete statement of a decision and is not one, so a model can be handed a promise whose exception was cut off and report the code as contradicting it. build_prompt makes the same choice for the same reason and truncates only the diff.

Callers pass items in priority order — most valuable first — because this drops from the tail.

Trait Implementations

impl Clone for GraphContext

fn clone(&self) -> GraphContext

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for GraphContext

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

Formats the value using the given formatter.

impl Default for GraphContext

fn default() -> GraphContext

Returns the “default value” for a type.

impl Eq for GraphContext

impl PartialEq for GraphContext

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

Equality operator ==.

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

Inequality operator !=.

impl StructuralPartialEq for GraphContext