Struct QueueReport 

pub struct QueueReport {
    pub merged: Vec<u64>,
    pub rejected: Vec<(u64, Rejection)>,
    pub final_state: String,
    pub window_trace: Vec<usize>,
    pub ci_runs: usize,
    pub merge_invocations: usize,
    pub replayed: Vec<u64>,
    pub provider_error: Option<String>,
    pub unreported_checks: Vec<String>,
}

Outcome of draining a queue.

Fields

merged: Vec<u64>

Changes merged, in merge order.

rejected: Vec<(u64, Rejection)>

Changes rejected, with reasons.

final_state: String

Final repository state after all merges.

window_trace: Vec<usize>

Window size after each head decision (for observing TCP dynamics).

ci_runs: usize

Total CI executions, including retests behind failures.

merge_invocations: usize

Total strategy-pipeline invocations. A change resolved from memory::ResolutionMemory does not invoke the pipeline, which is what a test counts to prove a replay happened.

replayed: Vec<u64>

Changes whose conflict was resolved from memory, in train order. Every one of them still ran CI: replay produces a candidate, never a landing.

provider_error: Option<String>

Why the drain stopped early without blaming a change (D18).

Set when the executor produced no verdict, or an inconclusive one: a provider that could not be reached, a batch whose verdict count did not match its jobs, a job that errored or timed out. Everything not landed is back in the queue, in order, and nothing was rejected on account of it – which is the distinction the old bool seam could not make.

unreported_checks: Vec<String>

Check reports the sequencer refused, verbatim (D49).

Empty unless MergeQueue::set_check_reporter was called. A refused report is not a landing decision – the merge already happened or did not on the verdict itself – but silence about it would leave the log missing checks with nothing saying so, which is the shape of a bug nobody finds.

Trait Implementations

impl Debug for QueueReport

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

Formats the value using the given formatter.