fallow_engine/

viz.rs

//! Typed data contract and builder for `fallow viz`.
//!
//! The CLI runs one project analysis (dead code + duplication + complexity)
//! through [`crate::session::AnalysisSession`] and hands the retained
//! artifacts to [`build_viz_data`]. The resulting [`VizData`] is embedded as
//! JSON in the self-contained interactive HTML the `viz` command writes.
//!
//! The contract is engine-owned so the graph internals never leak past the
//! engine boundary: everything the frontend needs is resolved to file
//! indices, relative paths, and plain counts here.

use std::path::Path;

use rustc_hash::FxHashMap;
use serde::Serialize;

use fallow_config::{ResolvedConfig, WorkspaceInfo};
use fallow_types::discover::DiscoveredFile;
use fallow_types::duplicates::{CloneInstance, DuplicationReport};
use fallow_types::extract::{FunctionComplexity, ModuleInfo};
use fallow_types::results::AnalysisResults;

use crate::module_graph::RetainedModuleGraph;

/// A file counts as a complexity hotspot at or above this cyclomatic score.
const HOTSPOT_CYCLOMATIC_FLOOR: u16 = 10;
/// Maximum bytes of clone-fragment preview shipped per clone group. The
/// budget is measured in bytes, not characters; truncation only ever cuts
/// at a line boundary, so multi-byte source cannot be sliced mid-character.
const CLONE_PREVIEW_MAX_BYTES: usize = 2000;
/// Maximum lines of clone-fragment preview shipped per clone group. The
/// preview grows to its content in the panel (no inner scroll), so this can
/// be generous; big blocks still truncate, keeping the leading context.
const CLONE_PREVIEW_MAX_LINES: usize = 32;
/// Source lines of context included on each side of the duplicated block
/// in a clone preview. A fixed window is universal: clones are frequently
/// not functions (interface fields, object literals, type aliases), so no
/// enclosing-scope detection is attempted.
const CLONE_PREVIEW_CONTEXT: usize = 4;
/// Maximum clone groups serialized into the payload. Far above any
/// legitimate report; a guardrail against multi-MB HTML on monorepos.
/// Groups keep the detector's report order, so the cap keeps the first N.
const MAX_CLONE_GROUPS: usize = 500;
/// Edge flag bit: every import of this edge is type-only.
const EDGE_FLAG_TYPE_ONLY: u32 = 1;

/// Everything [`build_viz_data`] needs from one project analysis run.
pub struct VizBuildInput<'a> {
    /// Dead-code analysis results (unused files/exports, cycles, boundaries).
    pub results: &'a AnalysisResults,
    /// Retained module graph for edges, entry points, and export counts.
    pub graph: &'a RetainedModuleGraph,
    /// Parsed modules with complexity data, when retained.
    pub modules: Option<&'a [ModuleInfo]>,
    /// Discovered source files, in `FileId` order.
    pub files: &'a [DiscoveredFile],
    /// Duplication report from the same session.
    pub duplication: &'a DuplicationReport,
    /// Discovered monorepo workspaces.
    pub workspaces: &'a [WorkspaceInfo],
    /// Resolved config (project root + boundary zones).
    pub config: &'a ResolvedConfig,
}

/// Serialized payload embedded in the viz HTML.
#[derive(Serialize)]
pub struct VizData {
    /// Project display name (root directory basename).
    pub root: String,
    /// One entry per analyzed source file, indexed by position.
    pub files: Vec<VizFile>,
    /// Import edges as `[from, to, flags]` file-index pairs.
    /// `flags` bit 0 marks an edge whose imports are all type-only.
    pub edges: Vec<[u32; 3]>,
    /// Project-wide totals for the header stat boxes.
    pub summary: VizSummary,
    /// Discovered workspaces; `VizFile.workspace` indexes into this.
    pub workspaces: Vec<VizWorkspace>,
    /// Boundary zones; `VizFile.zone` and violations index into this.
    pub zones: Vec<VizZone>,
    /// Circular-dependency cycles as file-index lists.
    pub cycles: Vec<Vec<u32>>,
    /// Clone groups; `VizFile.clone_groups` indexes into this.
    pub clones: Vec<VizCloneGroup>,
    /// Boundary violations resolved to file indices.
    pub violations: Vec<VizViolation>,
}

/// One analyzed source file.
#[derive(Serialize)]
pub struct VizFile {
    /// Root-relative path with forward slashes.
    pub path: String,
    /// File size in bytes (treemap area).
    pub size: u64,
    /// Dead-code status classification.
    pub status: VizFileStatus,
    /// Number of exports declared by the file.
    pub export_count: u16,
    /// Number of exports (values + types) reported unused.
    pub unused_export_count: u16,
    /// Whether the file is an entry point.
    pub is_entry: bool,
    /// Number of files importing this file.
    pub importer_count: u16,
    /// Number of files this file imports.
    pub import_count: u16,
    /// Index into `VizData.workspaces`, if the file belongs to one.
    #[serde(skip_serializing_if = "Option::is_none")]
    pub workspace: Option<u16>,
    /// Index into `VizData.zones`, if the file matches a boundary zone.
    #[serde(skip_serializing_if = "Option::is_none")]
    pub zone: Option<u16>,
    /// Names of unused exports (for actionable tooltips).
    #[serde(skip_serializing_if = "Vec::is_empty")]
    pub unused_exports: Vec<String>,
    /// Number of functions parsed in the file.
    pub fn_count: u16,
    /// Highest cyclomatic complexity of any function in the file.
    pub max_cyclomatic: u16,
    /// Highest cognitive complexity of any function in the file.
    pub max_cognitive: u16,
    /// Total React hook calls across the file's functions.
    pub react_hooks: u16,
    /// Deepest JSX nesting across the file's functions.
    pub jsx_depth: u16,
    /// Every function in the file, sorted hardest-first.
    #[serde(skip_serializing_if = "Vec::is_empty")]
    pub functions: Vec<VizFunction>,
    /// Duplicated lines in this file across all clone groups.
    pub dup_lines: u32,
    /// Indices into `VizData.clones` this file participates in.
    #[serde(skip_serializing_if = "Vec::is_empty")]
    pub clone_groups: Vec<u32>,
    /// Whether the file participates in any circular dependency.
    pub in_cycle: bool,
}

/// Dead-code status of a file, ordered by severity in the frontend.
#[derive(Serialize, Clone, Copy, PartialEq, Eq)]
#[serde(rename_all = "camelCase")]
pub enum VizFileStatus {
    /// No findings.
    Clean,
    /// Live file with one or more unused exports.
    HasUnusedExports,
    /// Entire file is unreachable.
    Unused,
    /// Configured or detected entry point.
    EntryPoint,
}

/// One function inside a file, with its complexity metrics.
#[derive(Serialize)]
pub struct VizFunction {
    /// Function name, or `<anonymous>`.
    name: String,
    /// 1-based start line.
    line: u32,
    /// McCabe cyclomatic complexity.
    cyclomatic: u16,
    /// SonarSource cognitive complexity.
    cognitive: u16,
    /// Body line count.
    lines: u32,
    /// React hook calls made directly in the body.
    hooks: u16,
    /// Deepest JSX nesting in the body.
    jsx_depth: u16,
    /// Props destructured from the first parameter.
    props: u16,
}

/// Project-wide totals for the header stat boxes.
#[derive(Serialize)]
pub struct VizSummary {
    /// Total analyzed files.
    pub total_files: usize,
    /// Total bytes across analyzed files.
    pub total_size: u64,
    /// Total import edges.
    pub total_edges: usize,
    /// Fully unused files.
    pub unused_files: usize,
    /// Unused exports (values + types).
    pub unused_exports: usize,
    /// Unused exported types.
    pub unused_types: usize,
    /// Unused dependencies (prod + dev + optional).
    pub unused_deps: usize,
    /// Imports that resolve to nothing.
    pub unresolved_imports: usize,
    /// Circular dependency cycles.
    pub circular_deps: usize,
    /// Clone groups detected.
    pub clone_groups: usize,
    /// Total duplicated lines across clone groups.
    pub duplicated_lines: usize,
    /// Boundary violations.
    pub boundary_violations: usize,
    /// Files at or above the complexity hotspot floor.
    pub hotspot_files: usize,
    /// Kept clone groups dropped by the `MAX_CLONE_GROUPS` payload cap.
    /// Present only when the clone payload was truncated.
    #[serde(skip_serializing_if = "Option::is_none")]
    pub clone_groups_truncated: Option<u32>,
}

/// One discovered workspace.
#[derive(Serialize)]
pub struct VizWorkspace {
    /// Package name.
    name: String,
    /// Root-relative workspace root.
    root: String,
}

/// One configured boundary zone.
#[derive(Serialize)]
pub struct VizZone {
    /// Zone name from the boundaries config.
    name: String,
    /// Number of files classified into this zone.
    files: u32,
}

/// One clone group resolved to file indices.
#[derive(Serialize)]
pub struct VizCloneGroup {
    /// Lines per duplicated block.
    lines: usize,
    /// Tokens per duplicated block.
    tokens: usize,
    /// Where the duplicated block appears.
    instances: Vec<VizCloneInstance>,
    /// Source preview: a context window around the duplicated block, the
    /// copied lines flanked by up to `CLONE_PREVIEW_CONTEXT` surrounding
    /// source lines on each side.
    preview: String,
    /// 0-based index, among the lines of `preview`, of the first copied
    /// line. Lines before it are dimmed context.
    highlight_start: u32,
    /// Number of copied lines present in `preview`. The frontend highlights
    /// `preview` lines `[highlight_start, highlight_start + highlight_lines)`
    /// and dims the rest.
    highlight_lines: u32,
}

/// One location of a duplicated block.
#[derive(Serialize)]
pub struct VizCloneInstance {
    /// File index into `VizData.files`.
    file: u32,
    /// 1-based start line.
    start_line: u32,
    /// 1-based end line.
    end_line: u32,
}

/// One boundary violation resolved to file indices.
#[derive(Serialize)]
pub struct VizViolation {
    /// Importing file index.
    from: u32,
    /// Imported file index.
    to: u32,
    /// Index into `VizData.zones` for the importing file's zone.
    from_zone: u16,
    /// Index into `VizData.zones` for the imported file's zone.
    to_zone: u16,
    /// 1-based line of the offending import.
    line: u32,
    /// Raw import specifier.
    specifier: String,
}

/// Build the viz payload from one project analysis run.
#[must_use]
pub fn build_viz_data(input: &VizBuildInput<'_>) -> VizData {
    let root = &input.config.root;
    let index = FileIndex::new(input.files);
    let workspaces = build_workspaces(input.workspaces, root);
    let (zones, zone_by_file) = classify_zones(input, &index);
    let (clones, clone_groups_by_file, dup_lines_by_file, clone_groups_truncated) =
        build_clones(input.duplication, &index, MAX_CLONE_GROUPS);
    let cycles = build_cycles(input.results, &index);
    let violations = build_violations(input.results, &zones, &index);

    let files = build_files(
        input,
        &index,
        &FilePropertyMaps {
            zone_by_file: &zone_by_file,
            clone_groups_by_file: &clone_groups_by_file,
            dup_lines_by_file: &dup_lines_by_file,
            cycles: &cycles,
        },
    );

    let summary = build_summary(
        input,
        &files,
        &clones,
        &cycles,
        &violations,
        clone_groups_truncated,
    );

    VizData {
        root: display_root(root),
        files,
        edges: build_edges(input.graph, &index),
        summary,
        workspaces,
        zones,
        cycles,
        clones,
        violations,
    }
}

/// Maps absolute paths to dense viz file indices in `FileId` order.
struct FileIndex<'a> {
    ordered: Vec<&'a DiscoveredFile>,
    by_path: FxHashMap<&'a Path, u32>,
    by_file_id: FxHashMap<u32, u32>,
}

impl<'a> FileIndex<'a> {
    fn new(files: &'a [DiscoveredFile]) -> Self {
        let mut ordered: Vec<&DiscoveredFile> = files.iter().collect();
        ordered.sort_by_key(|f| f.id.0);
        let mut by_path = FxHashMap::default();
        let mut by_file_id = FxHashMap::default();
        for (i, f) in ordered.iter().enumerate() {
            let idx = clamp_u32(i);
            by_path.insert(f.path.as_path(), idx);
            by_file_id.insert(f.id.0, idx);
        }
        Self {
            ordered,
            by_path,
            by_file_id,
        }
    }

    fn index_of_path(&self, path: &Path) -> Option<u32> {
        self.by_path.get(path).copied()
    }

    fn index_of_file_id(&self, file_id: u32) -> Option<u32> {
        self.by_file_id.get(&file_id).copied()
    }
}

/// Per-file lookup maps threaded into [`build_files`].
struct FilePropertyMaps<'a> {
    zone_by_file: &'a FxHashMap<u32, u16>,
    clone_groups_by_file: &'a FxHashMap<u32, Vec<u32>>,
    dup_lines_by_file: &'a FxHashMap<u32, u32>,
    cycles: &'a [Vec<u32>],
}

fn display_root(root: &Path) -> String {
    root.file_name().map_or_else(
        || root.to_string_lossy().into_owned(),
        |n| n.to_string_lossy().into_owned(),
    )
}

fn relative_path(path: &Path, root: &Path) -> String {
    path.strip_prefix(root)
        .unwrap_or(path)
        .to_string_lossy()
        .replace('\\', "/")
}

fn build_workspaces(workspaces: &[WorkspaceInfo], root: &Path) -> Vec<VizWorkspace> {
    workspaces
        .iter()
        .map(|ws| VizWorkspace {
            name: ws.name.clone(),
            root: relative_path(&ws.root, root),
        })
        .collect()
}

fn workspace_index_for(path: &Path, workspaces: &[WorkspaceInfo]) -> Option<u16> {
    let mut best: Option<(usize, usize)> = None;
    for (i, ws) in workspaces.iter().enumerate() {
        if path.starts_with(&ws.root) {
            let depth = ws.root.components().count();
            if best.is_none_or(|(_, d)| depth > d) {
                best = Some((i, depth));
            }
        }
    }
    best.map(|(i, _)| clamp_u16(i))
}

fn classify_zones(
    input: &VizBuildInput<'_>,
    index: &FileIndex<'_>,
) -> (Vec<VizZone>, FxHashMap<u32, u16>) {
    let boundaries = &input.config.boundaries;
    let mut zones: Vec<VizZone> = boundaries
        .zones
        .iter()
        .map(|z| VizZone {
            name: z.name.clone(),
            files: 0,
        })
        .collect();
    let name_to_index: FxHashMap<&str, u16> = boundaries
        .zones
        .iter()
        .enumerate()
        .map(|(i, z)| (z.name.as_str(), clamp_u16(i)))
        .collect();

    let mut zone_by_file = FxHashMap::default();
    if zones.is_empty() {
        return (zones, zone_by_file);
    }

    for (i, file) in index.ordered.iter().enumerate() {
        let rel = relative_path(&file.path, &input.config.root);
        if let Some(zone_name) = boundaries.classify_zone(&rel)
            && let Some(&zone_idx) = name_to_index.get(zone_name)
        {
            zone_by_file.insert(clamp_u32(i), zone_idx);
            zones[zone_idx as usize].files += 1;
        }
    }

    (zones, zone_by_file)
}

/// Clone payload maps: kept groups, per-file group ids, per-file duplicated
/// lines, and how many kept-groups the payload cap dropped.
type CloneMaps = (
    Vec<VizCloneGroup>,
    FxHashMap<u32, Vec<u32>>,
    FxHashMap<u32, u32>,
    u32,
);

fn build_clones(
    duplication: &DuplicationReport,
    index: &FileIndex<'_>,
    max_groups: usize,
) -> CloneMaps {
    let mut clones = Vec::new();
    let mut groups_by_file: FxHashMap<u32, Vec<u32>> = FxHashMap::default();
    let mut dup_lines_by_file: FxHashMap<u32, u32> = FxHashMap::default();
    let mut truncated: usize = 0;

    for group in &duplication.clone_groups {
        let instances: Vec<VizCloneInstance> = group
            .instances
            .iter()
            .filter_map(|inst| {
                index
                    .index_of_path(&inst.file)
                    .map(|file| VizCloneInstance {
                        file,
                        start_line: clamp_u32(inst.start_line),
                        end_line: clamp_u32(inst.end_line),
                    })
            })
            .collect();
        if instances.len() < 2 {
            continue;
        }
        if clones.len() >= max_groups {
            truncated += 1;
            continue;
        }

        let group_idx = clamp_u32(clones.len());
        for inst in &instances {
            let entry = groups_by_file.entry(inst.file).or_default();
            if entry.last() != Some(&group_idx) {
                entry.push(group_idx);
            }
            *dup_lines_by_file.entry(inst.file).or_default() +=
                inst.end_line.saturating_sub(inst.start_line) + 1;
        }

        let (preview, highlight_start, highlight_lines) = group
            .instances
            .first()
            .map(build_clone_preview)
            .unwrap_or_default();

        clones.push(VizCloneGroup {
            lines: group.line_count,
            tokens: group.token_count,
            instances,
            preview,
            highlight_start,
            highlight_lines,
        });
    }

    (
        clones,
        groups_by_file,
        dup_lines_by_file,
        clamp_u32(truncated),
    )
}

fn truncate_preview(fragment: &str) -> String {
    let mut out = String::new();
    for (i, line) in fragment.lines().enumerate() {
        if i >= CLONE_PREVIEW_MAX_LINES || out.len() + line.len() > CLONE_PREVIEW_MAX_BYTES {
            out.push('\u{2026}');
            break;
        }
        if i > 0 {
            out.push('\n');
        }
        out.push_str(line);
    }
    out
}

/// Build the representative clone preview: a context window around the
/// duplicated block, with the highlight range located within it. Returns
/// `(preview, highlight_start, highlight_lines)` where `highlight_start`
/// is the 0-based index of the first copied line among the preview lines
/// and `highlight_lines` is the copied line count present in `preview`.
///
/// Falls back to the bare fragment with the whole block highlighted on
/// any read failure, empty source, or an out-of-range line span. Never
/// panics.
fn build_clone_preview(inst: &CloneInstance) -> (String, u32, u32) {
    let Ok(source) = std::fs::read_to_string(&inst.file) else {
        return fragment_fallback(&inst.fragment);
    };
    let lines: Vec<&str> = source.lines().collect();
    let total = lines.len();
    if total == 0 || inst.start_line == 0 || inst.start_line > total {
        return fragment_fallback(&inst.fragment);
    }

    // Block bounds as a 0-based `[block_start, block_end)` range, clamped
    // to the file and guaranteed to hold at least one line.
    let block_start = inst.start_line - 1;
    let block_end = inst.end_line.min(total).max(inst.start_line);
    let mut block_lines = block_end - block_start;
    let mut before = block_start.min(CLONE_PREVIEW_CONTEXT);
    let mut after = (total - block_end).min(CLONE_PREVIEW_CONTEXT);

    // Line cap: when the block plus its context fits, trim context
    // symmetrically to fit. When the block alone fills the cap, keep the
    // leading context (so the highlight always reads against some dimmed
    // lines) and truncate the block's tail, always keeping >= 1 block line.
    if before + block_lines + after > CLONE_PREVIEW_MAX_LINES {
        if before + block_lines >= CLONE_PREVIEW_MAX_LINES {
            after = 0;
            block_lines = CLONE_PREVIEW_MAX_LINES.saturating_sub(before).max(1);
        } else {
            trim_context(
                &mut before,
                &mut after,
                CLONE_PREVIEW_MAX_LINES - block_lines,
            );
        }
    }

    enforce_byte_cap(
        &lines,
        block_start,
        &mut before,
        &mut after,
        &mut block_lines,
    );

    let win_start = block_start - before;
    let win_end = win_start + before + block_lines + after;
    let preview = lines[win_start..win_end].join("\n");
    (preview, clamp_u32(before), clamp_u32(block_lines))
}

/// Fallback preview: the bare fragment, capped, with the whole block
/// highlighted (nothing dimmed).
fn fragment_fallback(fragment: &str) -> (String, u32, u32) {
    let preview = truncate_preview(fragment);
    let highlight_lines = if preview.is_empty() {
        0
    } else {
        preview.lines().count()
    };
    (preview, 0, clamp_u32(highlight_lines))
}

/// Reduce `before`/`after` so their sum fits `budget`, dropping from the
/// larger side first (ties favor keeping `after`) so the two flanks stay
/// balanced. Deterministic.
fn trim_context(before: &mut usize, after: &mut usize, budget: usize) {
    while *before + *after > budget {
        if *before >= *after {
            *before -= 1;
        } else {
            *after -= 1;
        }
    }
}

/// Trim the preview window to `CLONE_PREVIEW_MAX_BYTES`, dropping context
/// lines (larger side first) before ever cutting into the highlighted
/// block. If the block alone still overflows, its tail lines are dropped,
/// but at least one line is always kept.
fn enforce_byte_cap(
    lines: &[&str],
    block_start: usize,
    before: &mut usize,
    after: &mut usize,
    block_lines: &mut usize,
) {
    let window_bytes = |before: usize, after: usize, block_lines: usize| -> usize {
        let start = block_start - before;
        let end = start + before + block_lines + after;
        let separators = (end - start).saturating_sub(1);
        lines[start..end].iter().map(|l| l.len()).sum::<usize>() + separators
    };
    while window_bytes(*before, *after, *block_lines) > CLONE_PREVIEW_MAX_BYTES {
        if *before + *after > 0 {
            if *before >= *after {
                *before -= 1;
            } else {
                *after -= 1;
            }
        } else if *block_lines > 1 {
            *block_lines -= 1;
        } else {
            break;
        }
    }
}

fn build_cycles(results: &AnalysisResults, index: &FileIndex<'_>) -> Vec<Vec<u32>> {
    results
        .circular_dependencies
        .iter()
        .filter_map(|cd| {
            let ids: Vec<u32> = cd
                .cycle
                .files
                .iter()
                .filter_map(|p| index.index_of_path(p))
                .collect();
            (ids.len() == cd.cycle.files.len()).then_some(ids)
        })
        .collect()
}

fn build_violations(
    results: &AnalysisResults,
    zones: &[VizZone],
    index: &FileIndex<'_>,
) -> Vec<VizViolation> {
    let name_to_index: FxHashMap<&str, u16> = zones
        .iter()
        .enumerate()
        .map(|(i, z)| (z.name.as_str(), clamp_u16(i)))
        .collect();

    results
        .boundary_violations
        .iter()
        .filter_map(|finding| {
            let v = &finding.violation;
            let from = index.index_of_path(&v.from_path)?;
            let to = index.index_of_path(&v.to_path)?;
            let from_zone = *name_to_index.get(v.from_zone.as_str())?;
            let to_zone = *name_to_index.get(v.to_zone.as_str())?;
            Some(VizViolation {
                from,
                to,
                from_zone,
                to_zone,
                line: v.line,
                specifier: v.import_specifier.clone(),
            })
        })
        .collect()
}

fn build_edges(graph: &RetainedModuleGraph, index: &FileIndex<'_>) -> Vec<[u32; 3]> {
    let graph = graph.as_graph();
    let mut edges = Vec::with_capacity(graph.edge_count());
    for node in &graph.modules {
        let Some(source) = index.index_of_file_id(node.file_id.0) else {
            continue;
        };
        for (target_id, all_type_only, _span) in graph.outgoing_edge_summaries(node.file_id) {
            let Some(target) = index.index_of_file_id(target_id.0) else {
                continue;
            };
            let flags = if all_type_only {
                EDGE_FLAG_TYPE_ONLY
            } else {
                0
            };
            edges.push([source, target, flags]);
        }
    }
    edges
}

/// Complexity aggregates for one file, folded from its parsed functions.
#[derive(Default)]
struct ComplexityRollup {
    fn_count: u16,
    max_cyclomatic: u16,
    max_cognitive: u16,
    react_hooks: u16,
    jsx_depth: u16,
    functions: Vec<VizFunction>,
}

fn rollup_complexity(functions: &[FunctionComplexity]) -> ComplexityRollup {
    let mut rollup = ComplexityRollup {
        fn_count: clamp_u16(functions.len()),
        ..ComplexityRollup::default()
    };
    for f in functions {
        rollup.max_cyclomatic = rollup.max_cyclomatic.max(f.cyclomatic);
        rollup.max_cognitive = rollup.max_cognitive.max(f.cognitive);
        rollup.react_hooks = rollup.react_hooks.saturating_add(f.react_hook_count);
        rollup.jsx_depth = rollup.jsx_depth.max(f.react_jsx_max_depth);
    }

    // Named functions only, hardest-first: the panel lists these and folds the
    // (often many) anonymous arrow/callback functions into a single count via
    // `fn_count`. Placeholder names for unnamed functions are `<arrow>` /
    // `<anonymous>`, so a leading `<` marks the ones to fold away.
    let mut named: Vec<&FunctionComplexity> = functions
        .iter()
        .filter(|f| !f.name.starts_with('<'))
        .collect();
    named.sort_by(|a, b| {
        b.cyclomatic
            .cmp(&a.cyclomatic)
            .then(b.cognitive.cmp(&a.cognitive))
    });
    rollup.functions = named
        .into_iter()
        .map(|f| VizFunction {
            name: f.name.clone(),
            line: f.line,
            cyclomatic: f.cyclomatic,
            cognitive: f.cognitive,
            lines: f.line_count,
            hooks: f.react_hook_count,
            jsx_depth: f.react_jsx_max_depth,
            props: f.react_prop_count,
        })
        .collect();
    rollup
}

fn build_files(
    input: &VizBuildInput<'_>,
    index: &FileIndex<'_>,
    maps: &FilePropertyMaps<'_>,
) -> Vec<VizFile> {
    let graph = input.graph.as_graph();
    let unused_file_paths: rustc_hash::FxHashSet<&Path> = input
        .results
        .unused_files
        .iter()
        .map(|f| f.file.path.as_path())
        .collect();

    let mut unused_exports_by_file: FxHashMap<&Path, Vec<String>> = FxHashMap::default();
    for export in &input.results.unused_exports {
        unused_exports_by_file
            .entry(export.export.path.as_path())
            .or_default()
            .push(export.export.export_name.clone());
    }
    for export in &input.results.unused_types {
        unused_exports_by_file
            .entry(export.export.path.as_path())
            .or_default()
            .push(export.export.export_name.clone());
    }

    let mut complexity_by_file_id: FxHashMap<u32, ComplexityRollup> = FxHashMap::default();
    if let Some(modules) = input.modules {
        for module in modules {
            if !module.complexity.is_empty() {
                complexity_by_file_id
                    .insert(module.file_id.0, rollup_complexity(&module.complexity));
            }
        }
    }

    let mut in_cycle = vec![false; index.ordered.len()];
    for cycle in maps.cycles {
        for &idx in cycle {
            if let Some(slot) = in_cycle.get_mut(idx as usize) {
                *slot = true;
            }
        }
    }

    index
        .ordered
        .iter()
        .enumerate()
        .map(|(i, file)| {
            let viz_idx = clamp_u32(i);
            let node_idx = file.id.0 as usize;
            let node = graph.modules.get(node_idx);
            let is_entry = node.is_some_and(|n| n.is_entry_point());
            let export_count = node.map_or(0, |n| clamp_u16(n.exports.len()));
            let import_count = clamp_u16(graph.edges_for(file.id).len());
            let importer_count = clamp_u16(input.graph.direct_importer_count(file.id));

            let unused_export_names = unused_exports_by_file
                .remove(file.path.as_path())
                .unwrap_or_default();
            let unused_export_count = clamp_u16(unused_export_names.len());

            let status = if unused_file_paths.contains(file.path.as_path()) {
                VizFileStatus::Unused
            } else if unused_export_count > 0 {
                VizFileStatus::HasUnusedExports
            } else if is_entry {
                VizFileStatus::EntryPoint
            } else {
                VizFileStatus::Clean
            };

            let complexity = complexity_by_file_id.remove(&file.id.0).unwrap_or_default();

            VizFile {
                path: relative_path(&file.path, &input.config.root),
                size: file.size_bytes,
                status,
                export_count,
                unused_export_count,
                is_entry,
                importer_count,
                import_count,
                workspace: workspace_index_for(&file.path, input.workspaces),
                zone: maps.zone_by_file.get(&viz_idx).copied(),
                unused_exports: unused_export_names,
                fn_count: complexity.fn_count,
                max_cyclomatic: complexity.max_cyclomatic,
                max_cognitive: complexity.max_cognitive,
                react_hooks: complexity.react_hooks,
                jsx_depth: complexity.jsx_depth,
                functions: complexity.functions,
                dup_lines: maps.dup_lines_by_file.get(&viz_idx).copied().unwrap_or(0),
                clone_groups: maps
                    .clone_groups_by_file
                    .get(&viz_idx)
                    .cloned()
                    .unwrap_or_default(),
                in_cycle: in_cycle[i],
            }
        })
        .collect()
}

fn build_summary(
    input: &VizBuildInput<'_>,
    files: &[VizFile],
    clones: &[VizCloneGroup],
    cycles: &[Vec<u32>],
    violations: &[VizViolation],
    clone_groups_truncated: u32,
) -> VizSummary {
    let results = input.results;
    VizSummary {
        total_files: files.len(),
        total_size: files.iter().map(|f| f.size).sum(),
        total_edges: input.graph.edge_count(),
        unused_files: results.unused_files.len(),
        unused_exports: results.unused_exports.len() + results.unused_types.len(),
        unused_types: results.unused_types.len(),
        unused_deps: results.unused_dependencies.len()
            + results.unused_dev_dependencies.len()
            + results.unused_optional_dependencies.len(),
        unresolved_imports: results.unresolved_imports.len(),
        circular_deps: cycles.len(),
        clone_groups: clones.len(),
        duplicated_lines: clones.iter().map(|c| c.lines * c.instances.len()).sum(),
        boundary_violations: violations.len(),
        hotspot_files: files
            .iter()
            .filter(|f| f.max_cyclomatic >= HOTSPOT_CYCLOMATIC_FLOOR)
            .count(),
        clone_groups_truncated: (clone_groups_truncated > 0).then_some(clone_groups_truncated),
    }
}

fn clamp_u16(value: usize) -> u16 {
    u16::try_from(value).unwrap_or(u16::MAX)
}

fn clamp_u32(value: usize) -> u32 {
    u32::try_from(value).unwrap_or(u32::MAX)
}

#[cfg(test)]
mod tests {
    use std::path::PathBuf;

    use fallow_config::{BoundaryConfig, BoundaryZone, FallowConfig};
    use fallow_graph::graph::ModuleGraph;
    use fallow_graph::resolve::{ResolveResult, ResolvedImport, ResolvedModule};
    use fallow_types::duplicates::{CloneGroup, CloneInstance};
    use fallow_types::extract::{ImportInfo, ImportedName};
    use fallow_types::output_dead_code::{BoundaryViolationFinding, CircularDependencyFinding};
    use fallow_types::output_format::OutputFormat;
    use fallow_types::results::{BoundaryViolation, CircularDependency};

    use super::*;
    use crate::discover::{EntryPoint, EntryPointSource, FileId};

    /// Owned fixture parts backing one [`VizBuildInput`].
    struct Fixture {
        config: ResolvedConfig,
        files: Vec<DiscoveredFile>,
        results: AnalysisResults,
        graph: crate::module_graph::RetainedModuleGraph,
        duplication: DuplicationReport,
        workspaces: Vec<WorkspaceInfo>,
    }

    impl Fixture {
        fn input(&self) -> VizBuildInput<'_> {
            VizBuildInput {
                results: &self.results,
                graph: &self.graph,
                modules: None,
                files: &self.files,
                duplication: &self.duplication,
                workspaces: &self.workspaces,
                config: &self.config,
            }
        }
    }

    fn project_root() -> PathBuf {
        PathBuf::from("/viz-project")
    }

    fn discovered(id: u32, path: PathBuf, size_bytes: u64) -> DiscoveredFile {
        DiscoveredFile {
            id: FileId(id),
            path,
            size_bytes,
        }
    }

    fn import_of(target: FileId, specifier: &str) -> ResolvedImport {
        ResolvedImport {
            info: ImportInfo {
                source: specifier.to_owned(),
                imported_name: ImportedName::Named("value".to_owned()),
                local_name: "value".to_owned(),
                is_type_only: false,
                from_style: false,
                span: oxc_span::Span::new(0, 0),
                source_span: oxc_span::Span::new(0, 0),
            },
            target: ResolveResult::InternalModule(target),
        }
    }

    fn zone(name: &str, pattern: &str) -> BoundaryZone {
        BoundaryZone {
            name: name.to_owned(),
            patterns: vec![pattern.to_owned()],
            auto_discover: Vec::new(),
            root: None,
        }
    }

    fn resolved_config(root: &Path) -> ResolvedConfig {
        let config = FallowConfig {
            boundaries: BoundaryConfig {
                zones: vec![zone("app", "src/**"), zone("shared", "lib/**")],
                ..BoundaryConfig::default()
            },
            ..FallowConfig::default()
        };
        config.resolve(root.to_path_buf(), OutputFormat::Json, 1, false, true, None)
    }

    fn cycle_finding(files: Vec<PathBuf>) -> CircularDependencyFinding {
        let length = files.len();
        CircularDependencyFinding::with_actions(CircularDependency {
            files,
            length,
            line: 1,
            col: 0,
            edges: Vec::new(),
            is_cross_package: false,
        })
    }

    fn violation_finding(from_path: PathBuf, to_path: PathBuf) -> BoundaryViolationFinding {
        BoundaryViolationFinding::with_actions(BoundaryViolation {
            from_path,
            to_path,
            from_zone: "app".to_owned(),
            to_zone: "shared".to_owned(),
            import_specifier: "../lib/c".to_owned(),
            line: 2,
            col: 0,
        })
    }

    fn clone_instance(file: PathBuf, start_line: usize, end_line: usize) -> CloneInstance {
        CloneInstance {
            file,
            start_line,
            end_line,
            start_col: 0,
            end_col: 0,
            fragment: "const shared = 1;\nconst repeated = 2;\nconst block = 3;".to_owned(),
        }
    }

    fn clone_group(instances: Vec<CloneInstance>) -> CloneGroup {
        CloneGroup {
            instances,
            token_count: 12,
            line_count: 3,
            similarity: None,
        }
    }

    /// Synthetic project: 3 files, one import edge a to b, one resolvable
    /// cycle (a, b) plus one unresolvable, one clone group over (a, c) plus a
    /// dropped and a same-file group, one resolvable boundary violation a to
    /// c plus one unresolvable, two zones, one workspace over `lib/`.
    fn fixture_with(extra_graph_file: bool) -> Fixture {
        let root = project_root();
        let a = root.join("src/a.ts");
        let b = root.join("src/b.ts");
        let c = root.join("lib/c.ts");
        let missing = root.join("src/missing.ts");

        let files = vec![
            discovered(0, a.clone(), 100),
            discovered(1, b.clone(), 50),
            discovered(2, c.clone(), 25),
        ];

        let mut graph_files = files.clone();
        let mut imports = vec![import_of(FileId(1), "./b")];
        if extra_graph_file {
            graph_files.push(discovered(3, root.join("src/d.ts"), 10));
            imports.push(import_of(FileId(3), "./d"));
        }
        let resolved = vec![ResolvedModule {
            file_id: FileId(0),
            path: a.clone(),
            resolved_imports: imports,
            ..ResolvedModule::default()
        }];
        let entry_points = vec![EntryPoint {
            path: a.clone(),
            source: EntryPointSource::PackageJsonMain,
        }];
        let graph = crate::module_graph::RetainedModuleGraph::from(ModuleGraph::build(
            &resolved,
            &entry_points,
            &graph_files,
        ));

        let results = AnalysisResults {
            circular_dependencies: vec![
                cycle_finding(vec![a.clone(), b]),
                cycle_finding(vec![a.clone(), missing.clone()]),
            ],
            boundary_violations: vec![
                violation_finding(a.clone(), c.clone()),
                violation_finding(a.clone(), missing),
            ],
            ..AnalysisResults::default()
        };

        let duplication = DuplicationReport {
            clone_groups: vec![
                clone_group(vec![
                    clone_instance(a.clone(), 1, 3),
                    clone_instance(c, 10, 12),
                ]),
                clone_group(vec![
                    clone_instance(a.clone(), 20, 22),
                    clone_instance(root.join("outside.ts"), 1, 3),
                ]),
                clone_group(vec![
                    clone_instance(a.clone(), 30, 32),
                    clone_instance(a, 40, 42),
                ]),
            ],
            ..DuplicationReport::default()
        };

        let workspaces = vec![WorkspaceInfo {
            root: root.join("lib"),
            name: "shared-lib".to_owned(),
            is_internal_dependency: false,
        }];

        Fixture {
            config: resolved_config(&root),
            files,
            results,
            graph,
            duplication,
            workspaces,
        }
    }

    fn fixture() -> Fixture {
        fixture_with(false)
    }

    #[test]
    fn files_and_edges_use_stable_indices() {
        let fx = fixture();
        let data = build_viz_data(&fx.input());

        let paths: Vec<&str> = data.files.iter().map(|f| f.path.as_str()).collect();
        assert_eq!(paths, ["src/a.ts", "src/b.ts", "lib/c.ts"]);
        assert_eq!(data.edges, vec![[0, 1, 0]]);
        assert!(data.files[0].is_entry);
        assert!(matches!(data.files[0].status, VizFileStatus::EntryPoint));
        assert!(matches!(data.files[1].status, VizFileStatus::Clean));
        assert_eq!(data.files[0].import_count, 1);
        assert_eq!(data.files[1].importer_count, 1);
        assert_eq!(data.files[0].workspace, None);
        assert_eq!(data.files[2].workspace, Some(0));
        assert_eq!(data.workspaces.len(), 1);
        assert_eq!(data.workspaces[0].root, "lib");
    }

    #[test]
    fn edges_to_files_missing_from_input_are_dropped() {
        let fx = fixture_with(true);
        let data = build_viz_data(&fx.input());

        // The graph carries a to b AND a to d, but d is not in `input.files`,
        // so build_edges drops the second edge instead of emitting a
        // dangling index.
        assert_eq!(fx.graph.edge_count(), 2);
        assert_eq!(data.edges, vec![[0, 1, 0]]);
    }

    #[test]
    fn clone_groups_drop_unresolvable_and_dedup_per_file() {
        let fx = fixture();
        let data = build_viz_data(&fx.input());

        // The group whose second instance lives outside `input.files` keeps
        // only 1 resolvable instance and is dropped entirely.
        assert_eq!(data.clones.len(), 2);
        assert_eq!(data.clones[0].instances.len(), 2);
        assert_eq!(data.clones[0].instances[0].file, 0);
        assert_eq!(data.clones[0].instances[1].file, 2);
        assert_eq!(data.clones[0].lines, 3);
        assert_eq!(data.clones[0].tokens, 12);
        // Two same-file instances in one group dedup to a single group id.
        assert_eq!(data.files[0].clone_groups, vec![0, 1]);
        assert_eq!(data.files[2].clone_groups, vec![0]);
        // dup_lines sums (end minus start plus 1) per resolvable instance.
        assert_eq!(data.files[0].dup_lines, 9);
        assert_eq!(data.files[2].dup_lines, 3);
        assert_eq!(data.files[1].dup_lines, 0);
    }

    #[test]
    fn truncate_preview_caps_lines_and_bytes() {
        // Line cap: more lines than the cap in, CLONE_PREVIEW_MAX_LINES out
        // plus the ellipsis appended directly after the last kept line.
        let last_kept = CLONE_PREVIEW_MAX_LINES - 1;
        let many_lines = (0..CLONE_PREVIEW_MAX_LINES + 5)
            .map(|i| format!("line {i}"))
            .collect::<Vec<_>>();
        let out = truncate_preview(&many_lines.join("\n"));
        assert_eq!(out.matches('\n').count(), CLONE_PREVIEW_MAX_LINES - 1);
        assert!(out.contains(&format!("line {last_kept}")));
        assert!(!out.contains(&format!("line {CLONE_PREVIEW_MAX_LINES}")));
        assert!(out.ends_with('\u{2026}'));

        // Byte budget: the second big line would exceed CLONE_PREVIEW_MAX_BYTES,
        // so output stops after the first line.
        let big = CLONE_PREVIEW_MAX_BYTES * 3 / 4;
        let two_long_lines = format!("{}\n{}", "a".repeat(big), "b".repeat(big));
        let out = truncate_preview(&two_long_lines);
        assert_eq!(out, format!("{}\u{2026}", "a".repeat(big)));

        // Multi-byte content over budget truncates at a line boundary and
        // never slices inside a character (4 bytes per emoji, well over budget).
        let emoji_line = "\u{1f389}".repeat(CLONE_PREVIEW_MAX_BYTES);
        let out = truncate_preview(&emoji_line);
        assert_eq!(out, "\u{2026}");
    }

    #[test]
    fn clone_preview_windows_context_around_the_block() {
        use std::io::Write as _;

        // 20 numbered source lines; the copied block covers lines 8..=11.
        let mut file = tempfile::NamedTempFile::new().expect("temp file");
        let body = (1..=20)
            .map(|i| format!("line {i}"))
            .collect::<Vec<_>>()
            .join("\n");
        file.write_all(body.as_bytes()).expect("write source");
        let inst = clone_instance(file.path().to_path_buf(), 8, 11);

        let (preview, highlight_start, highlight_lines) = build_clone_preview(&inst);
        let preview_lines: Vec<&str> = preview.lines().collect();

        // Block (4 lines) plus 4 lines of context each side fits the cap, so
        // the full window is kept: 4 dimmed + 4 highlighted + 4 dimmed.
        assert_eq!(preview_lines.len(), 12);
        assert_eq!(highlight_start, 4);
        assert_eq!(highlight_lines, 4);
        assert_eq!(preview_lines.first(), Some(&"line 4"));
        let start = highlight_start as usize;
        let end = start + highlight_lines as usize;
        assert_eq!(
            &preview_lines[start..end],
            ["line 8", "line 9", "line 10", "line 11"],
        );
        // The line directly above the block is dimmed context, not copied.
        assert_eq!(preview_lines[start - 1], "line 7");
    }

    #[test]
    fn clone_preview_keeps_leading_context_when_the_block_fills_the_cap() {
        use std::io::Write as _;

        // A block far larger than the cap. The old logic zeroed the context
        // and highlighted the whole (truncated) window; the fix keeps the
        // leading context dimmed so the highlight still reads against it.
        let mut file = tempfile::NamedTempFile::new().expect("temp file");
        let body = (1..=200)
            .map(|i| format!("line {i}"))
            .collect::<Vec<_>>()
            .join("\n");
        file.write_all(body.as_bytes()).expect("write source");
        let inst = clone_instance(file.path().to_path_buf(), 50, 150);

        let (preview, highlight_start, highlight_lines) = build_clone_preview(&inst);
        let preview_lines: Vec<&str> = preview.lines().collect();

        assert_eq!(highlight_start, CLONE_PREVIEW_CONTEXT as u32);
        assert!(
            highlight_start > 0,
            "leading context must survive a huge block"
        );
        assert_eq!(preview_lines.len(), CLONE_PREVIEW_MAX_LINES);
        assert_eq!(
            highlight_lines as usize,
            CLONE_PREVIEW_MAX_LINES - CLONE_PREVIEW_CONTEXT,
        );
        assert_eq!(preview_lines[highlight_start as usize - 1], "line 49");
        assert_eq!(preview_lines[highlight_start as usize], "line 50");
    }

    #[test]
    fn clone_preview_clamps_context_at_file_start() {
        use std::io::Write as _;

        let mut file = tempfile::NamedTempFile::new().expect("temp file");
        file.write_all(b"line 1\nline 2\nline 3\nline 4\nline 5")
            .expect("write source");
        // Block at the very top: no context fits above it, so the highlight
        // starts at index 0 and the trailing lines are dimmed context.
        let inst = clone_instance(file.path().to_path_buf(), 1, 2);

        let (preview, highlight_start, highlight_lines) = build_clone_preview(&inst);
        assert_eq!(highlight_start, 0);
        assert_eq!(highlight_lines, 2);
        assert_eq!(preview, "line 1\nline 2\nline 3\nline 4\nline 5");
    }

    #[test]
    fn clone_preview_falls_back_when_source_is_unreadable() {
        // A missing file forces the fragment fallback: the whole block is
        // highlighted so nothing is dimmed.
        let inst = clone_instance(project_root().join("does-not-exist.ts"), 1, 3);
        let (preview, highlight_start, highlight_lines) = build_clone_preview(&inst);
        assert_eq!(preview, inst.fragment);
        assert_eq!(highlight_start, 0);
        assert_eq!(highlight_lines as usize, preview.lines().count());
    }

    #[test]
    fn cycles_drop_when_any_member_unresolved() {
        let fx = fixture();
        let data = build_viz_data(&fx.input());

        // The a/b cycle resolves fully; the cycle referencing the missing
        // file yields no entry at all (not a partial one).
        assert_eq!(data.cycles, vec![vec![0, 1]]);
        assert!(data.files[0].in_cycle);
        assert!(data.files[1].in_cycle);
        assert!(!data.files[2].in_cycle);
        // The summary counts the rendered cycles, not the raw results, so
        // the dropped cycle does not inflate the header number.
        assert_eq!(data.summary.circular_deps, data.cycles.len());
    }

    #[test]
    fn violations_resolve_zone_and_file_indices() {
        let fx = fixture();
        let data = build_viz_data(&fx.input());

        assert_eq!(data.zones.len(), 2);
        assert_eq!(data.zones[0].name, "app");
        assert_eq!(data.zones[0].files, 2);
        assert_eq!(data.zones[1].name, "shared");
        assert_eq!(data.zones[1].files, 1);
        assert_eq!(data.files[0].zone, Some(0));
        assert_eq!(data.files[1].zone, Some(0));
        assert_eq!(data.files[2].zone, Some(1));

        // The violation whose to_path is not in `input.files` is dropped.
        assert_eq!(data.violations.len(), 1);
        let v = &data.violations[0];
        assert_eq!((v.from, v.to), (0, 2));
        assert_eq!((v.from_zone, v.to_zone), (0, 1));
        assert_eq!(v.line, 2);
        assert_eq!(v.specifier, "../lib/c");
    }

    #[test]
    fn clone_group_cap_counts_truncated_groups() {
        let fx = fixture();
        let index = FileIndex::new(&fx.files);

        // The fixture report has two keepable groups plus one dropped for
        // unresolvable instances; a cap of 1 keeps the first keepable group
        // and counts only the second as truncated (the unresolvable drop is
        // not a truncation).
        let (clones, groups_by_file, _dup_lines, truncated) =
            build_clones(&fx.duplication, &index, 1);
        assert_eq!(clones.len(), 1);
        assert_eq!(truncated, 1);
        assert!(
            groups_by_file
                .values()
                .all(|ids| ids.iter().all(|&id| (id as usize) < clones.len()))
        );

        // The default cap leaves a small report untouched and unflagged.
        let data = build_viz_data(&fx.input());
        assert_eq!(data.clones.len(), 2);
        assert_eq!(data.summary.clone_groups_truncated, None);
    }

    #[test]
    fn summary_flags_clone_truncation_only_when_nonzero() {
        let fx = fixture();
        let data = build_viz_data(&fx.input());

        let summary = build_summary(&fx.input(), &data.files, &data.clones, &[], &[], 3);
        assert_eq!(summary.clone_groups_truncated, Some(3));
        let summary = build_summary(&fx.input(), &data.files, &data.clones, &[], &[], 0);
        assert_eq!(summary.clone_groups_truncated, None);
    }

    #[test]
    fn summary_counts_match_rendered_arrays() {
        let fx = fixture();
        let data = build_viz_data(&fx.input());
        let s = &data.summary;

        assert_eq!(s.total_files, data.files.len());
        assert_eq!(s.total_size, 175);
        assert_eq!(s.total_edges, data.edges.len());
        assert_eq!(s.clone_groups, data.clones.len());
        assert_eq!(s.duplicated_lines, 12);
        assert_eq!(s.hotspot_files, 0);
        assert_eq!(s.unused_files, 0);
        assert_eq!(s.unused_exports, 0);
        // The raw results carry one unresolvable cycle and one unresolvable
        // violation; the header counts only what the arrays render.
        assert_eq!(s.circular_deps, data.cycles.len());
        assert_eq!(s.circular_deps, 1);
        assert_eq!(s.boundary_violations, data.violations.len());
        assert_eq!(s.boundary_violations, 1);
    }
}