differential-engine 0.13.0

//! A renderer-agnostic projection of one plan document.
//!
//! The arithmetic every reviewer surface needs — group totals, file totals,
//! resolved dependency edges, reviewed-mark keys — computed once, in the
//! domain. It lived in the TUI's constructor, which is why the stack had to
//! re-derive its own half and why the two drifted.

use std::collections::HashMap;

use crate::EngineError;
use crate::plan::ids::{HunkId, PlanIndex};
use crate::plan::{LineCounts, effort_name};
use crate::schema;

/// One resolved `depends_on` edge.
#[derive(Debug, Clone, PartialEq, Eq)]
pub struct Dependency {
    pub id: String,
    pub label: String,
    /// The dependency appears **later** in the plan.
    ///
    /// Which means the two groups depend on each other and the topological
    /// sort had to break the cycle. The plan says so rather than quietly
    /// presenting an order it could not honour.
    pub unsatisfied: bool,
    /// The symbols that produced the edge — why the dependency exists.
    pub via: Vec<String>,
    /// Why the sort could not honour it, when it could not. `unsatisfied` says
    /// that it happened; this says whether the cycle is in the change or only
    /// in the grouping.
    pub cycle: Option<schema::Cycle>,
}

#[derive(Debug, Clone, PartialEq, Eq)]
pub struct GroupView {
    pub id: String,
    pub label: String,
    /// The model's prose, carried so a renderer needs no second handle on the
    /// raw group. Between them the two renderers print these and nothing else
    /// of the document's text: the stack's commit body is both, the TUI's
    /// group header is the description alone.
    pub description: String,
    pub reason: String,
    pub effort: schema::Effort,
    pub role: Option<schema::Role>,
    pub class_ids: Vec<String>,
    /// Members in class order.
    pub hunks: Vec<HunkId>,
    /// Distinct paths touched. A rename counts twice, because the canonical
    /// view is `--no-renames`; zero-hunk changes contribute nothing.
    pub n_files: usize,
    pub counts: LineCounts,
    pub depends_on: Vec<Dependency>,
    /// The audit's back-fill: classes the model omitted, recovered by the
    /// coverage audit and read last (ADR 0001, invariant 5).
    pub unclassified: bool,
}

#[derive(Debug, Clone, PartialEq, Eq)]
pub struct FileView {
    pub path: String,
    /// Canonical hunks, file order.
    pub hunks: Vec<HunkId>,
    pub counts: LineCounts,
}

/// The projection. Owned, not borrowing the document, so a session can hold
/// both without becoming self-referential.
#[derive(Debug, Clone, PartialEq, Eq)]
pub struct ReviewView {
    pub groups: Vec<GroupView>,
    /// Every file in the document, document order — including the zero-hunk
    /// binary, submodule and mode-only changes the group view cannot surface.
    pub files: Vec<FileView>,
    group_of_hunk: HashMap<HunkId, usize>,
    /// Every hunk's own counts, document order — see `counts`.
    counts_of_hunk: Vec<LineCounts>,
    hunk_by_digest: HashMap<String, HunkId>,
    digest_of_hunk: HashMap<HunkId, String>,
    classes: HashMap<String, ClassMembers>,
}

/// One shape class, resolved: which hunk stands for it and which it holds.
///
/// Carried by the projection so a renderer never needs `PlanIndex`. The TUI
/// rebuilt one on every keypress to answer exactly these two questions, and a
/// `PlanIndex` borrows the document — which is why it could not be kept.
#[derive(Debug, Clone, PartialEq, Eq)]
pub struct ClassMembers {
    /// The hunk a skim reader is shown for this class.
    pub exemplar: HunkId,
    /// Every member, in document order.
    pub hunks: Vec<HunkId>,
}

/// Whether a set of hunks — a group's, a file's, a directory's — reads as done.
///
/// Every hunk marked, and at least one hunk to mark: a binary file or a
/// gitlink carries no hunks, and "all of nothing" must not light it up as
/// reviewed. Four rows in the TUI each re-derived that guard by hand.
pub fn all_reviewed<'a>(
    hunks: impl IntoIterator<Item = &'a HunkId>,
    reviewed: &std::collections::HashSet<usize>,
) -> bool {
    let mut any = false;
    for h in hunks {
        if !reviewed.contains(&h.index()) {
            return false;
        }
        any = true;
    }
    any
}

impl ReviewView {
    /// Project a document, validating it on the way through.
    ///
    /// Needs no store and therefore no port: reviewed-mark keys are a pure
    /// function of the hunk digests the document already carries, which is
    /// why `ReviewSession` can stop computing its own copy of them.
    pub fn project(doc: &schema::PlanDocument) -> Result<Self, EngineError> {
        let index = PlanIndex::build(doc)?;

        let groups = index.groups();
        let label_of: HashMap<&str, &str> = groups
            .iter()
            .map(|g| (g.id.as_str(), g.label.as_str()))
            .collect();
        let rank_of: HashMap<&str, usize> = groups
            .iter()
            .enumerate()
            .map(|(i, g)| (g.id.as_str(), i))
            .collect();

        // The back-fill group is assembled last and the ordering stage keeps
        // it trailing, so its position identifies it. Positional — but
        // positional in ONE place, instead of once per renderer.
        let backfilled = doc.audit.classes_missing.unwrap_or(0) > 0;

        let projected: Vec<GroupView> = groups
            .iter()
            .enumerate()
            .map(|(rank, g)| {
                let hunks = index.group_hunks(g);
                let files: std::collections::HashSet<&str> =
                    hunks.iter().map(|&h| index.hunk(h).file.as_str()).collect();
                GroupView {
                    id: g.id.clone(),
                    label: g.label.clone(),
                    description: g.description.clone(),
                    reason: g.reason.clone(),
                    effort: g.effort,
                    role: g.role,
                    class_ids: g.class_ids.clone(),
                    n_files: files.len(),
                    counts: hunks
                        .iter()
                        .map(|&h| LineCounts::of_hunk(index.hunk(h)))
                        .sum(),
                    depends_on: g
                        .depends_on
                        .iter()
                        .map(|e| Dependency {
                            label: label_of
                                .get(e.on.as_str())
                                .map(|l| (*l).to_string())
                                .unwrap_or_else(|| e.on.clone()),
                            unsatisfied: rank_of.get(e.on.as_str()).copied().unwrap_or(0) > rank,
                            id: e.on.clone(),
                            via: e.via.clone(),
                            cycle: e.cycle,
                        })
                        .collect(),
                    unclassified: backfilled && rank + 1 == groups.len(),
                    hunks,
                }
            })
            .collect();

        let mut group_of_hunk = HashMap::new();
        for (i, g) in projected.iter().enumerate() {
            for &h in &g.hunks {
                group_of_hunk.insert(h, i);
            }
        }

        let files: Vec<FileView> = doc
            .files
            .iter()
            .map(|f| {
                let hunks = index.file_hunks(f);
                FileView {
                    path: f.path.clone(),
                    counts: hunks
                        .iter()
                        .map(|&h| LineCounts::of_hunk(index.hunk(h)))
                        .sum(),
                    hunks,
                }
            })
            .collect();

        // One `LineCounts` per hunk, so any SUBSET of the document can be
        // sized without the projection holding the document. Group totals and
        // file totals are two such subsets; the group's part of a file is the
        // third, and it is the one no field can carry.
        let counts_of_hunk: Vec<LineCounts> = doc.hunks.iter().map(LineCounts::of_hunk).collect();

        let hunk_by_digest = doc
            .hunks
            .iter()
            .enumerate()
            .map(|(i, h)| (h.digest.clone(), HunkId::from_index(i)))
            .collect();
        let digest_of_hunk = doc
            .hunks
            .iter()
            .enumerate()
            .map(|(i, h)| (HunkId::from_index(i), h.digest.clone()))
            .collect();

        let classes = doc
            .classes
            .iter()
            .map(|c| {
                (
                    c.id.clone(),
                    ClassMembers {
                        exemplar: index.exemplar(&c.id),
                        hunks: index.class_hunks(&c.id),
                    },
                )
            })
            .collect();

        Ok(ReviewView {
            groups: projected,
            files,
            group_of_hunk,
            counts_of_hunk,
            hunk_by_digest,
            digest_of_hunk,
            classes,
        })
    }

    pub fn group_position(&self, id: &str) -> Option<usize> {
        self.groups.iter().position(|g| g.id == id)
    }

    /// The group owning a hunk, via its class.
    ///
    /// `None` only for a hunk whose class is in no group — impossible after
    /// the coverage audit, but the type says so rather than a comment.
    pub fn group_of_hunk(&self, hunk: HunkId) -> Option<&GroupView> {
        self.group_of_hunk.get(&hunk).map(|&i| &self.groups[i])
    }

    /// One group's part of one file: the hunks it has there, file order.
    ///
    /// The question a count beside a file answers while the reader is standing
    /// inside a group — the group map and the diff pane's file list both ask
    /// it, and both used to print `files[file].counts` instead, which is a
    /// number about the rest of the file as much as about what is on screen.
    ///
    /// A hunk another group owns is not this group's, however a widened window
    /// came to draw it: the answer is a fact about the plan, so it does not
    /// move under `z`.
    pub fn hunks_in(&self, group: usize, file: usize) -> Vec<HunkId> {
        let Some(f) = self.files.get(file) else {
            return Vec::new();
        };
        f.hunks
            .iter()
            .filter(|h| self.group_of_hunk.get(h) == Some(&group))
            .copied()
            .collect()
    }

    /// What a set of hunks adds and removes.
    ///
    /// `GroupView::counts` and `FileView::counts` are this over two fixed
    /// sets. Any other set — a group's part of a file, a directory, a
    /// selection — needs the arithmetic in the domain rather than a renderer
    /// reaching back into the document for `new_count`/`old_count`.
    pub fn counts(&self, hunks: &[HunkId]) -> LineCounts {
        hunks
            .iter()
            .filter_map(|h| self.counts_of_hunk.get(h.index()))
            .copied()
            .sum()
    }

    /// Findings anchor on digests, which survive regeneration where positional
    /// ids do not.
    pub fn hunk_by_digest(&self, digest: &str) -> Option<HunkId> {
        self.hunk_by_digest.get(digest).copied()
    }

    /// A hunk's exact content digest — the reviewed-mark key.
    pub fn digest(&self, hunk: HunkId) -> &str {
        &self.digest_of_hunk[&hunk]
    }

    /// Every hunk whose digest is marked.
    ///
    /// Walks the hunks rather than the marks, because a digest is content and
    /// content repeats: two byte-identical hunks carry one key, and marking
    /// either marks both. `hunk_by_digest` can only name one of them.
    pub fn hunks_marked<'k>(
        &self,
        marked: impl Fn(&str) -> bool + 'k,
    ) -> std::collections::HashSet<HunkId> {
        self.each_marked(marked).collect()
    }

    /// How many hunks of THIS document are marked.
    ///
    /// A count, not a set. The status bar prints this number every frame, and
    /// reaching it through `hunks_marked` allocated a whole `HashSet` to ask
    /// for its length — twice over, because the caller then rebuilt it as a
    /// set of indices.
    pub fn count_marked<'k>(&self, marked: impl Fn(&str) -> bool + 'k) -> usize {
        self.each_marked(marked).count()
    }

    fn each_marked<'a, 'k: 'a>(
        &'a self,
        marked: impl Fn(&str) -> bool + 'k,
    ) -> impl Iterator<Item = HunkId> + 'a {
        self.digest_of_hunk
            .iter()
            .filter(move |(_, digest)| marked(digest))
            .map(|(h, _)| *h)
    }

    /// One class's exemplar and members.
    ///
    /// Total for any id a group names: `PlanIndex::build` proved every one of
    /// them resolves before this projection was made.
    pub fn class(&self, id: &str) -> &ClassMembers {
        &self.classes[id]
    }

    /// The tier's domain name, or `unclassified` for the audit back-fill.
    ///
    /// One answer for both renderers: the stack has always labelled the
    /// back-fill distinctly and the TUI has always shown it as an ordinary
    /// focus group, which is the same document described two ways.
    pub fn tier_name(&self, group: &GroupView) -> &'static str {
        if group.unclassified {
            "unclassified"
        } else {
            effort_name(group.effort)
        }
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::plan::test_support::{doc_with, group, hunk_ids};

    #[test]
    fn all_reviewed_needs_every_hunk_and_at_least_one() {
        let hunks = [HunkId::from_index(0), HunkId::from_index(1)];
        let marked = |ids: &[usize]| {
            ids.iter()
                .copied()
                .collect::<std::collections::HashSet<_>>()
        };

        assert!(all_reviewed(&hunks, &marked(&[0, 1])));
        assert!(!all_reviewed(&hunks, &marked(&[0])), "one hunk unmarked");
        assert!(!all_reviewed(&hunks, &marked(&[])), "nothing marked");
        // A binary file or a gitlink has no hunks: "all of nothing" is not done.
        assert!(!all_reviewed(&[], &marked(&[0, 1])));
    }

    fn two_group_doc() -> schema::PlanDocument {
        let mut doc = doc_with(
            &[("C0", &["h0", "h1"], "h0"), ("C1", &["h2"], "h2")],
            &[("src/a.rs", &["h0", "h1"]), ("src/b.rs", &["h2"])],
        );
        doc.groups = Some(vec![
            group("g0", schema::Effort::Focus, &["C0"]),
            group("g1", schema::Effort::Skim, &["C1"]),
        ]);
        doc
    }

    #[test]
    fn groups_carry_their_totals_and_distinct_file_count() {
        let doc = two_group_doc();
        let view = ReviewView::project(&doc).unwrap();

        assert_eq!(hunk_ids(&view.groups[0].hunks), ["h0", "h1"]);
        assert_eq!(
            view.groups[0].n_files, 2,
            "the fixture puts each hunk in its own file"
        );
        // The fixture's hunks are +2/-1 each.
        assert_eq!(view.groups[0].counts, LineCounts { adds: 4, dels: 2 });
        assert_eq!(view.files[0].counts, LineCounts { adds: 4, dels: 2 });
    }

    /// A count beside a file, printed while the reader is inside one group,
    /// is about that group's part of the file. Two surfaces asked the question
    /// and both answered it with `files[i].counts`, which is the rest of the
    /// file as much as what is on screen.
    #[test]
    fn a_group_is_sized_within_one_file() {
        // `src/a.rs` holds a hunk from each group; `src/b.rs` holds one of g0's.
        let mut doc = doc_with(
            &[("C0", &["h0", "h1"], "h0"), ("C1", &["h2"], "h2")],
            &[("src/a.rs", &["h0", "h2"]), ("src/b.rs", &["h1"])],
        );
        doc.groups = Some(vec![
            group("g0", schema::Effort::Focus, &["C0"]),
            group("g1", schema::Effort::Skim, &["C1"]),
        ]);
        let view = ReviewView::project(&doc).unwrap();

        // The fixture's hunks are +2/-1 each, so the shared file totals +4/-2.
        assert_eq!(view.files[0].counts, LineCounts { adds: 4, dels: 2 });
        for (g, id) in [(0, "h0"), (1, "h2")] {
            let part = view.hunks_in(g, 0);
            assert_eq!(hunk_ids(&part), [id], "group {g}'s part of src/a.rs");
            assert_eq!(view.counts(&part), LineCounts { adds: 2, dels: 1 });
        }

        // A group that never enters the file has no part of it — and no size,
        // which is not the same number as the file's.
        assert!(view.hunks_in(1, 1).is_empty());
        assert_eq!(view.counts(&view.hunks_in(1, 1)), LineCounts::default());
        assert_eq!(view.files[1].counts, LineCounts { adds: 2, dels: 1 });
    }

    #[test]
    fn reviewed_keys_are_the_documents_own_hunk_digests() {
        let doc = two_group_doc();
        let view = ReviewView::project(&doc).unwrap();

        // One key per hunk, not one per class: two hunks of the same class
        // carry different keys, so changing one cannot unmark the other.
        assert_eq!(view.digest(HunkId::from_index(0)), "digest0");
        assert_eq!(view.digest(HunkId::from_index(1)), "digest1");
        assert_eq!(view.digest(HunkId::from_index(2)), "digest2");
    }

    #[test]
    fn a_dependency_listed_later_is_flagged_unsatisfied() {
        let mut doc = two_group_doc();
        // g0 (rank 0) depends on g1 (rank 1): the order could not honour it.
        let edge = |on: &str| schema::Edge {
            on: on.to_string(),
            via: vec!["Config".to_string()],
            cycle: Some(schema::Cycle::Artefact),
        };
        doc.groups.as_mut().unwrap()[0].depends_on = vec![edge("g1")];
        doc.groups.as_mut().unwrap()[1].depends_on = vec![edge("g0")];
        let view = ReviewView::project(&doc).unwrap();

        assert_eq!(
            view.groups[0].depends_on,
            [Dependency {
                via: vec!["Config".to_string()],
                cycle: Some(schema::Cycle::Artefact),
                id: "g1".into(),
                label: "g1 label".into(),
                unsatisfied: true
            }]
        );
        assert!(
            !view.groups[1].depends_on[0].unsatisfied,
            "a dependency earlier in the plan is honoured"
        );
    }

    #[test]
    fn hunks_resolve_to_their_owning_group_and_their_digest() {
        let doc = two_group_doc();
        let view = ReviewView::project(&doc).unwrap();

        assert_eq!(view.group_of_hunk(HunkId::from_index(1)).unwrap().id, "g0");
        assert_eq!(view.hunk_by_digest("digest2"), Some(HunkId::from_index(2)));
        assert_eq!(view.hunk_by_digest("nope"), None);
    }

    /// The asymmetry this projection exists to remove: one flag, so a renderer
    /// cannot decide for itself that a back-filled group is ordinary.
    #[test]
    fn the_trailing_backfill_group_is_marked_unclassified() {
        let mut doc = two_group_doc();
        assert!(
            ReviewView::project(&doc)
                .unwrap()
                .groups
                .iter()
                .all(|g| !g.unclassified),
            "no back-fill recorded in the audit"
        );

        doc.audit.classes_missing = Some(1);
        let view = ReviewView::project(&doc).unwrap();
        assert!(!view.groups[0].unclassified);
        assert!(
            view.groups[1].unclassified,
            "the back-fill is assembled last"
        );
        assert_eq!(view.tier_name(&view.groups[1]), "unclassified");
        assert_eq!(view.tier_name(&view.groups[0]), "focus");
    }
}