Struct Watermark 

pub struct Watermark {
    pub checkpoint_seq: u32,
    pub last_frame: u64,
}

Position in the WAL: a (checkpoint_seq, last_frame) pair. max_frame resets to 0 every time the WAL header restarts (WalAutoActions::Restart fires), and checkpoint_seq_no increments alongside it — so the pair is the right primary key for sink objects, not raw frame_no.

R858-B19 — checkpoint_seq alone does NOT identify a WAL generation, and this doc used to claim it did (“increments monotonically across restarts”). That claim is false and it cost a silent wrong restore. It holds only for an in-process restart, where the same WAL file is reused. When the last connection to a SQLite database closes, the engine checkpoints and DELETES the -wal file; the next writer creates a fresh WAL back at checkpoint-sequence 0 with a brand-new random salt. Across that fold checkpoint_seq goes 0 -> 0 over two completely unrelated WALs (measured 2026-09-06 against system sqlite3 3.51.0 by examples/foreign_checkpoint_probe.rs, probes B and F).

Generation identity therefore lives in WalGeneration, which pairs the sequence with the WAL header’s salt. Never compare two Watermarks’ checkpoint_seq to decide “same WAL” — that is exactly the inference this ticket exists to delete. last_frame is only meaningful relative to a known generation.

Fields

checkpoint_seq: u32

last_frame: u64

Trait Implementations

impl Clone for Watermark

fn clone(&self) -> Watermark

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for Watermark

impl Debug for Watermark

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

Formats the value using the given formatter.

impl Default for Watermark

fn default() -> Watermark

Returns the “default value” for a type.

impl Eq for Watermark

impl PartialEq for Watermark

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

Equality operator ==.

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

Inequality operator !=.

impl StructuralPartialEq for Watermark