DarkRoom/core/dr-catalog/src/lib.rs

//! TRACES: FR-CAT-2 | FR-CAT-4 | FR-CAT-6 | NFR-P1
//! The catalog: a rebuildable index over the library.
//!
//! Not a source of truth. Sidecars next to the images hold the authoritative
//! edit state (ARCH §6.12), and this file is deletable at any time — rebuilt
//! by rescanning sources and reading sidecars. That inversion is deliberate:
//! darktable maintains both a database and sidecars while achieving the
//! reliability of neither.
//!
//! # What lives here
//!
//! - [`schema`] — tables and forward-only migrations
//! - [`scan`] — incremental discovery that prunes unchanged directories
//! - [`query`] — selectors compiled to indexed SQL, windowed for the grid
//! - [`jobs`] — the durable background work queue
//! - [`merge`] / [`sync`] — cross-device collection merging
//!
//! # The one thing everything is designed around
//!
//! **Work is proportional to what changed, or to what the user is looking at —
//! never to library size.** A 50k-image library that has not changed costs one
//! metadata probe per folder to verify (§scan), no thumbnails to regenerate
//! (§jobs coalescing), and no rule evaluation per grid cell (materialised
//! `tier_desired`).

use std::path::Path;

use dr_types::{Availability, ImageId};
use rusqlite::Connection;

pub mod error;
pub mod jobs;
pub mod merge;
pub mod query;
pub mod scan;
pub mod schema;
pub mod sync;

pub use error::CatalogError;
pub use jobs::{Job, JobKind, Priority};
pub use merge::MergeReport;
pub use query::{Query, Sort};
pub use scan::{DirAction, DirState, EntryAction, ScanOutcome};

/// One row of the library grid.
///
/// Exactly what a cell draws and nothing more — no join per cell, and
/// availability reads a materialised column rather than evaluating cache rules
/// (ARCH §9.5).
#[derive(Debug, Clone, PartialEq, Eq)]
pub struct GridRow {
    pub id: ImageId,
    pub name: String,
    pub availability: Availability,
    /// UTC seconds. `None` until EXIF has been read.
    pub captured_at: Option<i64>,
    /// Minutes east of UTC, for rendering the photographer's local time.
    pub captured_offset: Option<i32>,
    /// 0 = nothing, 1 = stat-only, 2 = full EXIF.
    pub metadata_state: u8,
}

/// A count of images in one time bucket, for the timeline scrubber.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub struct TimeBucket {
    /// UTC seconds at the bucket's start.
    pub start: i64,
    pub count: u32,
}

/// Time bucket size.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum Granularity {
    Year,
    Month,
    Day,
    Hour,
}

impl Granularity {
    /// SQLite `strftime` format that collapses a timestamp to this bucket.
    ///
    /// Applied to **local** time, not UTC: "everything from 3 August" means
    /// the photographer's 3 August, which is why `captured_offset` is stored
    /// alongside the UTC timestamp.
    fn strftime(self) -> &'static str {
        match self {
            Granularity::Year => "%Y",
            Granularity::Month => "%Y-%m",
            Granularity::Day => "%Y-%m-%d",
            Granularity::Hour => "%Y-%m-%dT%H",
        }
    }

    /// A sensible bucket size for a span of seconds, so the UI need not guess.
    pub fn for_span(seconds: i64) -> Self {
        const DAY: i64 = 86_400;
        match seconds {
            s if s > 5 * 365 * DAY => Granularity::Year,
            s if s > 90 * DAY => Granularity::Month,
            s if s > 2 * DAY => Granularity::Day,
            _ => Granularity::Hour,
        }
    }
}

/// A connection to the catalog.
pub struct Catalog {
    conn: Connection,
}

impl Catalog {
    /// Open or create a catalog, migrating it forward if needed.
    pub fn open(path: &Path) -> Result<Self, CatalogError> {
        let conn = Connection::open(path)?;
        schema::configure(&conn)?;
        schema::migrate(&conn)?;
        Ok(Catalog { conn })
    }

    /// An in-memory catalog, for tests and for a throwaway import preview.
    pub fn in_memory() -> Result<Self, CatalogError> {
        let conn = Connection::open_in_memory()?;
        schema::configure(&conn)?;
        schema::migrate(&conn)?;
        Ok(Catalog { conn })
    }

    /// Escape hatch for modules that need raw access. Not part of the UI-facing
    /// surface.
    pub fn connection(&self) -> &Connection {
        &self.conn
    }

    /// How many images match.
    ///
    /// Returned alongside the first window so the grid can size its scrollbar
    /// and paint in one round trip.
    pub fn count(&self, q: &Query, now: i64) -> Result<usize, CatalogError> {
        let c = query::compile(&q.filter, now);
        let sql = query::count_sql(&c);
        let n: i64 =
            self.conn
                .query_row(&sql, rusqlite::params_from_iter(c.params.iter()), |r| {
                    r.get(0)
                })?;
        Ok(n as usize)
    }

    /// Fetch one window of results.
    ///
    /// Never returns the whole catalog: FR-CAT-4 requires memory bounded
    /// independently of library size.
    pub fn window(
        &self,
        q: &Query,
        range: std::ops::Range<usize>,
        now: i64,
    ) -> Result<Vec<GridRow>, CatalogError> {
        let c = query::compile(&q.filter, now);
        let sql = query::window_sql(q, &c);

        let mut params = c.params.clone();
        params.push(rusqlite::types::Value::Integer(range.len() as i64));
        params.push(rusqlite::types::Value::Integer(range.start as i64));

        let mut stmt = self.conn.prepare(&sql)?;
        let rows = stmt
            .query_map(rusqlite::params_from_iter(params.iter()), |r| {
                let source_ref: String = r.get(1)?;
                let avail: i64 = r.get(2)?;
                Ok(GridRow {
                    id: ImageId(r.get::<_, i64>(0)? as u64),
                    name: source_ref
                        .rsplit(['/', ':'])
                        .next()
                        .unwrap_or(&source_ref)
                        .to_string(),
                    availability: decode_availability(avail),
                    captured_at: r.get(3)?,
                    captured_offset: r.get::<_, Option<i64>>(4)?.map(|v| v as i32),
                    metadata_state: r.get::<_, i64>(5)? as u8,
                })
            })?
            .collect::<Result<Vec<_>, _>>()?;
        Ok(rows)
    }

    /// Counts per time bucket, for the timeline scrubber.
    ///
    /// One grouped aggregate over the `images_captured` index — not 50k rows
    /// handed to the UI to bucket itself.
    pub fn timeline(
        &self,
        q: &Query,
        g: Granularity,
        now: i64,
    ) -> Result<Vec<TimeBucket>, CatalogError> {
        let c = query::compile(&q.filter, now);
        // Bucketed in local time: captured_offset is minutes east of UTC, and
        // NULL falls back to UTC rather than dropping the row.
        let sql = format!(
            "SELECT min(captured_at) AS start,
                    count(*)          AS n
               FROM images
              WHERE {} AND captured_at IS NOT NULL
              GROUP BY strftime('{}', captured_at + coalesce(captured_offset, 0) * 60,
                                'unixepoch')
              ORDER BY start ASC",
            c.where_sql,
            g.strftime()
        );

        let mut stmt = self.conn.prepare(&sql)?;
        let rows = stmt
            .query_map(rusqlite::params_from_iter(c.params.iter()), |r| {
                Ok(TimeBucket {
                    start: r.get(0)?,
                    count: r.get::<_, i64>(1)? as u32,
                })
            })?
            .collect::<Result<Vec<_>, _>>()?;
        Ok(rows)
    }

    /// Merge a downloaded remote catalog's collections into this one.
    ///
    /// See [`sync`] for why only collections cross over.
    pub fn merge_remote_catalog(&self, remote: &Path) -> Result<MergeReport, CatalogError> {
        sync::merge_remote(&self.conn, remote)
    }

    /// Write a consistent snapshot ready to upload.
    pub fn snapshot_for_upload(&self, dest: &Path) -> Result<(), CatalogError> {
        sync::snapshot_for_upload(&self.conn, dest)
    }
}

fn decode_availability(v: i64) -> Availability {
    match v {
        1 => Availability::Preview,
        2 => Availability::Original,
        3 => Availability::Offline,
        _ => Availability::MetadataOnly,
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use dr_types::Selector;

    fn seeded() -> Catalog {
        let cat = Catalog::in_memory().unwrap();
        let c = cat.connection();
        c.execute(
            "INSERT INTO roots(id, kind, label) VALUES (1, 'local', 'lib')",
            [],
        )
        .unwrap();
        // Three images across two days, one with no EXIF read yet.
        for (id, name, captured, state) in [
            (1i64, "a.CR3", Some(1_000_000i64), 2i64),
            (2, "b.CR3", Some(1_100_000), 2),
            (3, "c.CR3", None, 1),
        ] {
            c.execute(
                "INSERT INTO images(id, root_id, source_ref, captured_at, metadata_state, added_at)
                 VALUES (?1, 1, ?2, ?3, ?4, 0)",
                rusqlite::params![id, name, captured, state],
            )
            .unwrap();
        }
        cat
    }

    #[test]
    fn count_and_window_agree() {
        let cat = seeded();
        let q = Query::default();
        assert_eq!(cat.count(&q, 0).unwrap(), 3);
        assert_eq!(cat.window(&q, 0..10, 0).unwrap().len(), 3);
    }

    #[test]
    fn window_is_bounded_by_the_requested_range() {
        // FR-CAT-4: memory independent of catalog size.
        let cat = seeded();
        let rows = cat.window(&Query::default(), 0..2, 0).unwrap();
        assert_eq!(rows.len(), 2);
    }

    #[test]
    fn paging_covers_every_row_exactly_once() {
        let cat = seeded();
        let q = Query::default();
        let mut seen = Vec::new();
        for start in (0..3).step_by(2) {
            seen.extend(cat.window(&q, start..start + 2, 0).unwrap());
        }
        let mut ids: Vec<u64> = seen.iter().map(|r| r.id.0).collect();
        ids.sort_unstable();
        assert_eq!(ids, vec![1, 2, 3]);
    }

    #[test]
    fn an_image_without_capture_time_sorts_last_not_first() {
        // Otherwise a freshly scanned library leads with whatever has not been
        // read yet, which looks like corruption to the user.
        let cat = seeded();
        let rows = cat.window(&Query::default(), 0..10, 0).unwrap();
        assert_eq!(rows.last().unwrap().id, ImageId(3));
    }

    #[test]
    fn metadata_state_reaches_the_grid() {
        // The grid needs it to distinguish "no photos on this date" from
        // "EXIF not read yet" (FR-NC-6c's honesty principle).
        let cat = seeded();
        let rows = cat.window(&Query::default(), 0..10, 0).unwrap();
        let pending = rows.iter().find(|r| r.id == ImageId(3)).unwrap();
        assert_eq!(pending.metadata_state, 1);
    }

    #[test]
    fn a_filter_narrows_the_count() {
        let cat = seeded();
        let q = Query {
            filter: Selector::Text("a.CR3".into()),
            ..Default::default()
        };
        assert_eq!(cat.count(&q, 0).unwrap(), 1);
    }

    #[test]
    fn timeline_buckets_and_skips_unread_images() {
        let cat = seeded();
        let buckets = cat
            .timeline(&Query::default(), Granularity::Day, 0)
            .unwrap();
        // Two images with timestamps, one day apart in UTC; the third has no
        // capture time and cannot be placed on a timeline at all.
        let total: u32 = buckets.iter().map(|b| b.count).sum();
        assert_eq!(total, 2);
    }

    #[test]
    fn timeline_granularity_follows_the_span() {
        const DAY: i64 = 86_400;
        assert_eq!(Granularity::for_span(10 * 365 * DAY), Granularity::Year);
        assert_eq!(Granularity::for_span(120 * DAY), Granularity::Month);
        assert_eq!(Granularity::for_span(10 * DAY), Granularity::Day);
        assert_eq!(Granularity::for_span(3600), Granularity::Hour);
    }

    #[test]
    fn names_are_derived_for_both_paths_and_saf_ids() {
        let cat = Catalog::in_memory().unwrap();
        let c = cat.connection();
        c.execute(
            "INSERT INTO roots(id, kind, label) VALUES (1, 'saf', 'tree')",
            [],
        )
        .unwrap();
        c.execute(
            "INSERT INTO images(id, root_id, source_ref, added_at)
             VALUES (1, 1, 'primary:DCIM/Camera/IMG_1.CR3', 0)",
            [],
        )
        .unwrap();
        let rows = cat.window(&Query::default(), 0..10, 0).unwrap();
        assert_eq!(rows[0].name, "IMG_1.CR3");
    }
}