forks/sorcery

git clone https://git.gae.moe/forks/sorcery

Grace Yodertypst: gate behind feature flag37af77a2026-10-20

use std::collections::BTreeMap;
use std::fmt::Write as _;
use std::fs;
use std::path::{Path, PathBuf};
use std::sync::Arc;
use std::sync::atomic::{AtomicBool, AtomicUsize, Ordering};

use anyhow::{Context, Result};
// Safe for text and double-quoted attribute contexts.
use html_escape::encode_double_quoted_attribute as escape;
use percent_encoding::{AsciiSet, NON_ALPHANUMERIC, PercentEncode, percent_encode};

use crate::catalog::Repository;
use crate::highlight::{Cache as GrammarCache, Highlighter};

/// Blobs larger than this get a raw download instead of a rendered page.
const MAX_RENDER_BYTES: u64 = 1 << 20;

/// Cooperative cancellation for background builds; set when a
/// human-triggered build wants the machine instead.
pub(crate) type BuildCancel = Arc<AtomicBool>;

pub(crate) fn bail_if_cancelled(cancel: &Option<BuildCancel>) -> Result<()> {
    if cancel
        .as_ref()
        .is_some_and(|flag| flag.load(Ordering::Relaxed))
    {
        anyhow::bail!("site build cancelled");
    }
    Ok(())
}

const PATH: &AsciiSet = &NON_ALPHANUMERIC
    .remove(b'-')
    .remove(b'.')
    .remove(b'_')
    .remove(b'~')
    .remove(b'/');

/// Percent-encodes a slash-separated path for use in a URI.
pub(crate) fn encode_path(path: &str) -> PercentEncode<'_> {
    percent_encode(path.as_bytes(), PATH)
}

/// `tip` names the page's ref and the commit it points at; the client reads
/// them from `<main data-ref data-tip>` to resolve hash routes relative to
/// the page.
fn page(
    instance_name: &str,
    title: &str,
    description: &str,
    tip: Option<(&str, gix::ObjectId)>,
    body: &str,
) -> String {
    format!(
        "<!doctype html>\n\
         <html lang=\"en\">\n\
         <head>\n\
         <meta charset=\"utf-8\">\n\
         <meta name=\"viewport\" content=\"width=device-width, initial-scale=1\">\n\
         <title>{title}</title>\n\
         <meta name=\"description\" content=\"{description}\">\n\
         <meta property=\"og:title\" content=\"{title}\">\n\
         <meta property=\"og:description\" content=\"{description}\">\n\
         <meta property=\"og:type\" content=\"website\">\n\
         <meta property=\"og:site_name\" content=\"{instance_name}\">\n\
         <link rel=\"stylesheet\" href=\"/css/style.css?v={format_version}\">\n\
         <script type=\"module\" src=\"/js/main.js?v={format_version}\"></script>\n\
         </head>\n\
         <body>\n<main{tip}>\n{noscript}{body}</main>\n</body>\n\
         </html>\n",
        instance_name = escape(instance_name),
        title = escape(title),
        description = escape(description),
        format_version = crate::generate::OUTPUT_FORMAT_VERSION,
        tip = tip
            .map(|(name, oid)| format!(" data-ref=\"{}\" data-tip=\"{oid}\"", escape(name)))
            .unwrap_or_default(),
        // only repo pages lose anything to a missing client
        noscript = if tip.is_some() { NOSCRIPT } else { "" },
    )
}

const NOSCRIPT: &str = r#"<noscript>
  <div class="callout">
    <p>
      hey! this is a static git repo viewer. commit diffs, file history,
      older commits and other revisions won't load without javascript.
    </p>
    <p>
      you are still able to view source files on branch tips,
      but the experience is degraded.
    </p>
  </div>
</noscript>
"#;

pub(crate) fn catalog_index(instance_name: &str, repositories: &[Repository]) -> String {
    let mut body = String::from("<h1>repositories</h1>\n");
    if repositories.is_empty() {
        body.push_str("<p class=\"meta\">no repositories found</p>\n");
    }

    for repos in repositories.chunk_by(|a, b| a.user == b.user) {
        let user = &repos[0].user;
        let _ = writeln!(
            body,
            "<h2><a href=\"/{}/\">{}</a></h2>",
            encode_path(user),
            escape(user),
        );
        catalog_list(&mut body, repos);
    }

    let title = format!("repositories - {instance_name}");
    let description = format!("Git repositories hosted on {instance_name}.");
    page(instance_name, &title, &description, None, &body)
}

/// `repositories` must all belong to `user`.
pub(crate) fn user_index(instance_name: &str, user: &str, repositories: &[Repository]) -> String {
    let mut body = format!("<h1>{}</h1>\n", escape(user));
    catalog_list(&mut body, repositories);
    let title = format!("{user} - {instance_name}");
    let description = format!("Git repositories by {user} on {instance_name}.");
    page(instance_name, &title, &description, None, &body)
}

fn catalog_list(body: &mut String, repositories: &[Repository]) {
    body.push_str("<ul class=\"catalog\">\n");
    for repo in repositories {
        let _ = writeln!(
            body,
            "<li><a href=\"/{user}/{repo}/\"><span class=\"repo-name\">{repo_name}</span> <span class=\"msg\">{description}</span>{updated}</a></li>",
            user = encode_path(&repo.user),
            repo = encode_path(&repo.name),
            repo_name = escape(&repo.name),
            description = escape(repo.description.as_deref().unwrap_or_default()),
            updated = repo
                .updated
                .map(|t| {
                    format!(
                        " <time datetime=\"{}\">{}</time>",
                        t.format_or_unix(gix::date::time::format::ISO8601_STRICT),
                        t.format_or_unix(gix::date::time::format::SHORT),
                    )
                })
                .unwrap_or_default(),
        );
    }
    body.push_str("</ul>\n");
}

#[derive(Clone, Copy, Debug, PartialEq, Eq)]
pub enum RefKind {
    Branch,
    Tag,
}

#[derive(Clone, Copy, PartialEq, Eq)]
pub enum StaticMode {
    Highlighted,
    /// Blob pages carry escaped plain text; the client highlights them.
    Plain,
}

#[derive(Clone, PartialEq, Eq, serde::Deserialize, serde::Serialize)]
pub(crate) struct BlobVersion {
    pub oid: String,
    pub mode: u16,
    pub touched: String,
}

#[derive(Clone, Copy)]
pub(crate) struct BlobReuse<'a> {
    pub versions: &'a BTreeMap<String, BlobVersion>,
    pub previous: Option<(&'a Path, &'a BTreeMap<String, BlobVersion>)>,
}

pub struct Site<'a> {
    pub repo: &'a gix::Repository,
    pub instance_name: String,
    pub name: String,
    pub base_url: String,
    pub description: Option<String>,
    pub clone_url: Option<String>,
}

impl Site<'_> {
    /// Display names like `user/repo` shorten to the repo part in crumbs.
    fn crumb_name(&self) -> &str {
        self.name.rsplit('/').next().unwrap_or(&self.name)
    }

    fn ref_href(&self, tip: &RefTip) -> String {
        match tip.static_mode {
            None => format!("{}#{}", self.base_url, tip.commit_id),
            Some(_) => {
                format!("{}ref/{}/", self.base_url, encode_path(&tip.name))
            }
        }
    }

    /// Name, description and clone command; the same on every page of the repo.
    fn repo_header(&self) -> String {
        let mut html = format!("<header class=\"repo\"><div>\n<h1>{}</h1>\n", escape(&self.name));
        if let Some(desc) = &self.description {
            let _ = writeln!(html, "<p class=\"desc\">{}</p>", escape(desc));
        }
        html.push_str("</div>\n");
        if let Some(url) = &self.clone_url {
            let _ = writeln!(html, "<pre class=\"clone\">git clone {}</pre>", escape(url));
        }
        html.push_str("</header>\n");
        html
    }
}

/// Author, summary, short sha and date of one commit, as a status line.
fn commit_panel(repo: &gix::Repository, oid: gix::ObjectId) -> String {
    let Ok(commit) = repo.find_object(oid).map(|o| o.try_into_commit()) else {
        return String::new();
    };
    let Ok(commit) = commit else {
        return String::new();
    };
    let summary = commit.message().map(|m| m.summary().to_string()).unwrap_or_default();
    let author = commit.author().map(|a| a.name.to_string()).unwrap_or_default();
    format!(
        "<p class=\"snapshot-commit\">\
         <span class=\"commit-author\">{author}</span>\
         <span class=\"commit-message\" data-commit=\"{oid}\">{summary}</span>\
         <span class=\"snapshot-sha sha\" data-commit=\"{oid}\">{short}</span>\
         {time}</p>\n",
        author = escape(&author),
        summary = escape(&summary),
        short = commit.id().shorten_or_id(),
        time = time_of(&commit),
    )
}

pub struct RefTip {
    pub kind: RefKind,
    pub name: String,
    pub commit_id: gix::ObjectId,
    pub static_mode: Option<StaticMode>,
}

/// All local branches and tags, peeled to commits. Refs that don't peel to a
/// commit (e.g. tags of blobs) are skipped.
pub fn list_refs(repo: &gix::Repository) -> Result<Vec<RefTip>> {
    let platform = repo.references()?;
    let mut tips: Vec<RefTip> = Vec::new();
    for (kind, iter) in [
        (RefKind::Branch, platform.local_branches()?),
        (RefKind::Tag, platform.tags()?),
    ] {
        for r in iter {
            let mut r = r.map_err(|e| anyhow::anyhow!("failed to iterate refs: {e}"))?;
            let name = r.name().shorten().to_string();
            // A branch and tag may share a short name. The flat /ref/<name>
            // namespace gives the branch precedence.
            if kind == RefKind::Tag
                && tips.iter().any(|t| t.kind == RefKind::Branch && t.name == name)
            {
                continue;
            }
            if let Ok(commit) = r.peel_to_commit() {
                tips.push(RefTip {
                    kind,
                    name,
                    commit_id: commit.id,
                    static_mode: None,
                });
            }
        }
    }
    Ok(tips)
}

/// The branch HEAD points at, if it exists among `tips` (falling back to the
/// first branch, matching forge behaviour for detached/unborn HEADs).
pub fn head_tip<'a>(repo: &gix::Repository, tips: &'a [RefTip]) -> Option<&'a RefTip> {
    let head = repo.head_name().ok().flatten().map(|n| n.shorten().to_string());
    tips.iter()
        .find(|t| t.kind == RefKind::Branch && Some(&t.name) == head.as_ref())
        .or_else(|| tips.iter().find(|t| t.kind == RefKind::Branch))
}

pub fn write_file(path: &Path, contents: impl AsRef<[u8]>) -> Result<()> {
    if let Some(parent) = path.parent() {
        fs::create_dir_all(parent)?;
    }
    fs::write(path, contents).with_context(|| format!("writing {}", path.display()))
}

/// Renders the full static view of one ref (tree pages, blob pages, raw files
/// for binaries) into `out`, which is typically a staging dir later swapped
/// into place at `<site>/ref/<name>/`.
pub fn render_ref(
    site: &Site,
    tips: &[RefTip],
    tip: &RefTip,
    commit: &gix::Commit,
    out: &Path,
    reuse: Option<BlobReuse<'_>>,
    highlights: Arc<GrammarCache>,
    cancel: Option<BuildCancel>,
) -> Result<()> {
    let mut r = RefRenderer::new(site, tips, tip, commit, out, highlights.clone(), cancel.clone());
    let tree = commit.tree()?;
    let mut blobs = Vec::new();
    r.walk(&tree, &mut Vec::new(), &mut blobs)?;

    // Blob pages dominate build time and are independent of each other, so
    // fan out; the repository is only ever read, so each worker gets its own
    // thread-local handle and reusable highlighter. Page costs vary wildly
    // (and cluster by directory), so workers pull from a shared counter
    // rather than taking fixed shares.
    let ctx = BlobContext {
        instance_name: site.instance_name.clone(),
        name: site.name.clone(),
        base_url: site.base_url.clone(),
        urlbase: r.urlbase.clone(),
        repo_header: site.repo_header(),
        label: tip.name.clone(),
        tip_id: commit.id,
        highlight: tip.static_mode != Some(StaticMode::Plain),
        out: out.to_owned(),
        reuse,
        highlights,
        cancel,
    };
    let repo = site.repo.clone().into_sync();
    let next = AtomicUsize::new(0);
    rayon::broadcast(|_| {
        let repo = repo.to_thread_local();
        let mut highlighter = Highlighter::new(ctx.highlights.clone());
        while let Some(job) = blobs.get(next.fetch_add(1, Ordering::Relaxed)) {
            if let Err(error) = bail_if_cancelled(&ctx.cancel)
                .and_then(|()| blob_page(&ctx, &repo, &mut highlighter, job))
            {
                next.store(blobs.len(), Ordering::Relaxed);
                return Err(error);
            }
        }
        Ok(())
    })
    .into_iter()
    .collect()
}

/// Everything a blob page needs besides the per-worker repository handle and
/// highlighter.
struct BlobContext<'a> {
    instance_name: String,
    name: String,
    base_url: String,
    urlbase: String,
    repo_header: String,
    label: String,
    tip_id: gix::ObjectId,
    highlight: bool,
    out: PathBuf,
    reuse: Option<BlobReuse<'a>>,
    highlights: Arc<GrammarCache>,
    cancel: Option<BuildCancel>,
}

struct BlobJob {
    path: Vec<String>,
    oid: gix::ObjectId,
    is_link: bool,
    /// Switcher and crumbs, prebuilt during the walk: they need ref-wide
    /// state that the workers don't carry.
    chrome: String,
}

fn blob_page(
    ctx: &BlobContext,
    repo: &gix::Repository,
    hl: &mut Highlighter,
    job: &BlobJob,
) -> Result<()> {
    let joined = job.path.join("/");
    let rel = format!("blob/{joined}");
    // Embedded images may change without the document blob changing.
    if !is_markdown(&joined)
        && !is_typst(&joined)
        && let Some(reuse) = ctx.reuse
        && let Some((previous, previous_versions)) = reuse.previous
        && reuse_blob(
            previous,
            previous_versions,
            reuse.versions,
            &joined,
            &ctx.out,
        )
    {
        return Ok(());
    }
    let data = &repo.find_object(job.oid)?.data.to_vec();

    let size = data.len() as u64;
    let image = !job.is_link && image_mime(&joined).is_some();
    let raw_url = format!("{}raw/{}/{}", ctx.base_url, job.oid, encode_path(&joined));
    let looks_binary = data[..data.len().min(8192)].contains(&0);
    let text = (!job.is_link && !image && !looks_binary && size <= MAX_RENDER_BYTES)
        .then(|| String::from_utf8_lossy(data));
    let lines = text.as_ref().map_or(0, |text| text.lines().count());
    let stats = if job.is_link {
        format!("<span>symlink \u{2192} {}</span>", escape(&String::from_utf8_lossy(data)))
    } else if text.is_some() {
        format!(
            "<span>{}</span><span>{lines} line{}</span>",
            human_size(size),
            if lines == 1 { "" } else { "s" },
        )
    } else {
        format!(
            "<span>{}</span><span>{}</span>",
            if image {
                "image"
            } else if looks_binary {
                "binary file"
            } else {
                "large file"
            },
            human_size(size),
        )
    };
    let touched = ctx
        .reuse
        .and_then(|reuse| reuse.versions.get(&joined))
        .and_then(|version| gix::ObjectId::from_hex(version.touched.as_bytes()).ok())
        .unwrap_or(ctx.tip_id);
    let mut body = format!(
        "{repo_header}{panel}<header class=\"topbar\">{chrome}<span class=\"view-stats\">{stats}</span>\
         <span class=\"actions\"><a class=\"raw\" href=\"{raw_url}\">raw</a></span></header>\n",
        repo_header = ctx.repo_header,
        panel = commit_panel(repo, touched),
        chrome = job.chrome,
    );

    if let Some(text) = text {
        let name = job.path.last().expect("blob path is never empty");
        // Plain mode defers highlighting to the client, which detects the
        // language from the path carried in data-hl.
        let (src_attrs, code) = if ctx.highlight {
            let lang = crate::grammar_manifest::detect(name);
            let code = lang
                .and_then(|lang| hl.highlight_lines(lang, &text).ok())
                .unwrap_or_else(|| text.lines().map(|line| escape(line).to_string()).collect());
            let class = lang.map(|l| format!(" language-{l}")).unwrap_or_default();
            (format!("class=\"code src{class}\""), code)
        } else {
            (
                format!("class=\"code src\" data-hl=\"{}\"", escape(&joined)),
                text.lines().map(|line| escape(line).to_string()).collect(),
            )
        };
        let mut source = format!("<pre {src_attrs}>");
        let line_count = code.len();
        for (i, line) in code.into_iter().enumerate() {
            let number = i + 1;
            let _ = write!(
                source,
                "<span class=\"code-line\"><a class=\"ln\" id=\"L{number}\" href=\"#L{number}\">{number}</a><span class=\"code-text\">{line}</span>"
            );
            if number < line_count {
                source.push_str("<span class=\"code-break\">\n</span>");
            }
            source.push_str("</span>");
        }
        source.push_str("</pre>");
        if is_markdown(name) {
            let urls = RelativeUrls {
                repo,
                base_url: &ctx.base_url,
                urlbase: &ctx.urlbase,
                commit: ctx.tip_id,
                directory: &job.path[..job.path.len() - 1],
            };
            let rendered = markdown(hl, &text, |url, image| urls.resolve(url, image));
            let _ = writeln!(
                body,
                "<div class=\"markdown-view\">\
                 <input type=\"radio\" name=\"markdown-view\" id=\"markdown-rendered\" checked>\
                 <label for=\"markdown-rendered\">rendered</label>\
                 <input type=\"radio\" name=\"markdown-view\" id=\"markdown-code\">\
                 <label for=\"markdown-code\">code</label>\
                 <section class=\"readme rendered\">{rendered}</section>\
                 <div class=\"code-panel\">{source}</div>\
                 </div>",
            );
        } else if is_typst(name) {
            let urls = RelativeUrls {
                repo,
                base_url: &ctx.base_url,
                urlbase: &ctx.urlbase,
                commit: ctx.tip_id,
                directory: &job.path[..job.path.len() - 1],
            };
            let rendered = typst(hl, &text, |url, image| urls.resolve(url, image));
            let _ = writeln!(
                body,
                "<div class=\"markdown-view\">\
                 <input type=\"radio\" name=\"markdown-view\" id=\"markdown-rendered\" checked>\
                 <label for=\"markdown-rendered\">rendered</label>\
                 <input type=\"radio\" name=\"markdown-view\" id=\"markdown-code\">\
                 <label for=\"markdown-code\">code</label>\
                 <section class=\"readme rendered\">{rendered}</section>\
                 <div class=\"code-panel\">{source}</div>\
                 </div>",
            );
        } else {
            let _ = writeln!(body, "{source}");
        }
    } else if image {
        let _ = writeln!(
            body,
            "<div class=\"image-preview\"><img src=\"{}\" alt=\"{}\"></div>",
            escape(&raw_url),
            escape(&joined)
        );
    } else if !job.is_link {
        let _ = writeln!(
            body,
            "<div class=\"file-download\"><p>This file cannot be previewed.</p><a href=\"{}\" download=\"{}\">download raw file</a></div>",
            escape(&raw_url),
            escape(job.path.last().expect("blob path is never empty"))
        );
    }

    let title = format!("{} - {} @ {}", joined, ctx.name, ctx.label);
    let description = format!("{joined} in {} at {}.", ctx.name, ctx.label);
    write_file(
        &ctx.out.join(&rel),
        page(&ctx.instance_name, &title, &description, Some((&ctx.label, ctx.tip_id)), &body),
    )
}

fn reuse_blob(
    previous: &Path,
    previous_versions: &BTreeMap<String, BlobVersion>,
    versions: &BTreeMap<String, BlobVersion>,
    path: &str,
    out: &Path,
) -> bool {
    let Some(version) = versions.get(path) else {
        return false;
    };
    if previous_versions.get(path) != Some(version) {
        return false;
    }
    let previous = previous.join("blob").join(path);
    let out = out.join("blob").join(path);
    if !previous.is_file()
        || out.parent().is_none_or(|parent| fs::create_dir_all(parent).is_err())
    {
        return false;
    }
    if fs::hard_link(&previous, &out).is_ok() || fs::copy(&previous, &out).is_ok() {
        return true;
    }
    let _ = fs::remove_file(out);
    false
}

/// The landing page: repo header plus the HEAD branch's root tree view, with
/// entry links pointing into the branch's own pages under `ref/`.
pub fn render_site_index(
    site: &Site,
    tips: &[RefTip],
    out: &Path,
    highlights: Arc<GrammarCache>,
    cancel: Option<BuildCancel>,
) -> Result<()> {
    let mut body = format!("<a class=\"back\" href=\"/\">← back</a>\n{}", site.repo_header());

    let head = head_tip(site.repo, tips);
    match head {
        None => body.push_str("<p class=\"meta\">no branches yet</p>\n"),
        Some(tip) => {
            let commit = site.repo.find_object(tip.commit_id)?.try_into_commit()?;
            let tree = commit.tree()?;
            let mut r = RefRenderer::new(site, tips, tip, &commit, out, highlights, cancel);
            let entries = collect_entries(&tree)?;
            body.push_str(&commit_panel(site.repo, tip.commit_id));
            body.push_str(&r.tree_topbar(&[], &entries));
            if tip.static_mode == Some(StaticMode::Highlighted) {
                body.push_str(&render_languages(
                    &crate::languages::analyze(site.repo, &tree)?,
                    tip.commit_id,
                ));
            }
            body.push_str(&r.overview(&entries));
            body.push_str(&r.readme_section(&[], &entries));
        }
    }

    let description = site
        .description
        .clone()
        .unwrap_or_else(|| format!("{} repository on {}.", site.name, site.instance_name));
    write_file(
        &out.join("index.html"),
        page(&site.instance_name, &site.name, &description, head.map(|tip| (tip.name.as_str(), tip.commit_id)), &body),
    )
}

/// `data-tip` and `data-language` let the client turn each label into a
/// link to that language's files at the tip.
fn render_languages(stats: &[crate::languages::Stat], tip: gix::ObjectId) -> String {
    let total = stats.iter().map(|stat| stat.bytes).sum::<u64>();
    if total == 0 {
        return String::new();
    }

    let mut html = format!(
        "<details class=\"languages\" data-tip=\"{tip}\">\n<summary><span class=\"language-label\">languages</span>",
    );
    for stat in stats {
        let percentage = stat.bytes as f64 * 100.0 / total as f64;
        let _ = write!(
            html,
            "<span title=\"{} {:.1}%\" style=\"background:{};width:{percentage:.6}%\"></span>",
            escape(stat.name),
            percentage,
            stat.color,
        );
    }
    html.push_str("</summary>\n<ul>\n");
    for stat in stats {
        let percentage = stat.bytes as f64 * 100.0 / total as f64;
        let _ = writeln!(
            html,
            "<li data-language=\"{}\"><i style=\"background:{}\"></i><span>{} <small>{:.1}%</small></span></li>",
            stat.id,
            stat.color,
            escape(stat.name),
            percentage,
        );
    }
    html.push_str("</ul>\n</details>\n");
    html
}

/// `/refs`: branch and tag tables with tip summaries and dates.
pub fn render_refs_page(site: &Site, tips: &[RefTip], out: &Path) -> Result<()> {
    let head = head_tip(site.repo, tips);
    let mut body = format!(
        "{}<header class=\"topbar\"><nav class=\"crumbs\"><a href=\"{}\">{}</a> / <span class=\"cur\">refs</span></nav></header>\n",
        site.repo_header(),
        site.base_url,
        escape(site.crumb_name()),
    );

    for (kind, heading) in [(RefKind::Branch, "branches"), (RefKind::Tag, "tags")] {
        // HEAD branch first, then alphabetical; tags newest first
        let mut dated: Vec<_> = tips
            .iter()
            .filter(|t| t.kind == kind)
            .map(|t| {
                let commit = site
                    .repo
                    .find_object(t.commit_id)
                    .map_err(anyhow::Error::from)
                    .and_then(|o| Ok(o.try_into_commit()?));
                let (summary, time, secs) = match &commit {
                    Ok(c) => (
                        c.message().map(|m| m.summary().to_string()).unwrap_or_default(),
                        time_of(c),
                        author_time(c).map(|t| t.seconds).unwrap_or(0),
                    ),
                    Err(_) => (String::new(), String::new(), 0),
                };
                (t, summary, time, secs)
            })
            .collect();
        if dated.is_empty() {
            continue;
        }
        match kind {
            RefKind::Branch => dated.sort_by(|a, b| {
                let is_head = |t: &RefTip| head.is_some_and(|head| head.name == t.name);
                is_head(b.0).cmp(&is_head(a.0)).then_with(|| a.0.name.cmp(&b.0.name))
            }),
            RefKind::Tag => dated.sort_by_key(|d| std::cmp::Reverse(d.3)),
        }

        let _ = write!(body, "<h2>{heading}</h2>\n<table class=\"list\">\n");
        for (tip, summary, time, _) in dated {
            let _ = writeln!(
                body,
                "<tr><td><a href=\"{href}\">{name}</a></td><td class=\"msg\">{summary}</td><td class=\"date\">{time}</td></tr>",
                href = site.ref_href(tip),
                name = escape(&tip.name),
                summary = escape(&summary),
                time = time,
            );
        }
        body.push_str("</table>\n");
    }

    let title = format!("refs - {}", site.name);
    let description = format!("Branches and tags for {}.", site.name);
    write_file(
        &out.join("refs"),
        page(&site.instance_name, &title, &description, head.map(|tip| (tip.name.as_str(), tip.commit_id)), &body),
    )
}

struct RefRenderer<'a> {
    site: &'a Site<'a>,
    tips: &'a [RefTip],
    out: &'a Path,
    label: String,
    /// Absolute URL prefix of this ref's pages, e.g. `/user/repo/ref/main/`.
    urlbase: String,
    tip_id: gix::ObjectId,
    hl: Highlighter,
    cancel: Option<BuildCancel>,
}

enum EntryKind {
    Dir,
    File,
    Link,
    Submodule,
}

type Entry = (String, EntryKind, gix::ObjectId);

fn collect_entries(tree: &gix::Tree<'_>) -> Result<Vec<Entry>> {
    let mut entries = Vec::new();
    for entry in tree.iter() {
        let entry = entry?;
        let mode = entry.mode();
        let kind = if mode.is_tree() {
            EntryKind::Dir
        } else if mode.is_link() {
            EntryKind::Link
        } else if mode.is_commit() {
            EntryKind::Submodule
        } else {
            EntryKind::File
        };
        entries.push((entry.filename().to_string(), kind, entry.oid().to_owned()));
    }
    entries.sort_by(|a, b| {
        let rank = |k: &EntryKind| !matches!(k, EntryKind::Dir);
        rank(&a.1).cmp(&rank(&b.1)).then_with(|| a.0.cmp(&b.0))
    });
    Ok(entries)
}

/// `a/b/c` when `a` holds only `b`, which holds only `c`: a chain of lone
/// directories reads better as one link than as three clicks.
fn collapse_lone_dirs(repo: &gix::Repository, name: &str, mut oid: gix::ObjectId) -> String {
    let mut path = name.to_string();
    while let Ok(tree) = repo.find_tree(oid) {
        let mut entries = tree.iter();
        match (entries.next(), entries.next()) {
            (Some(Ok(only)), None) if only.mode().is_tree() => {
                let _ = write!(path, "/{}", only.filename());
                oid = only.oid().to_owned();
            }
            _ => break,
        }
    }
    path
}

impl<'a> RefRenderer<'a> {
    fn new(
        site: &'a Site<'a>,
        tips: &'a [RefTip],
        tip: &RefTip,
        commit: &gix::Commit,
        out: &'a Path,
        highlights: Arc<GrammarCache>,
        cancel: Option<BuildCancel>,
    ) -> Self {
        RefRenderer {
            site,
            tips,
            out,
            label: tip.name.clone(),
            urlbase: format!("{}ref/{}/", site.base_url, encode_path(&tip.name)),
            tip_id: commit.id,
            hl: Highlighter::new(highlights),
            cancel,
        }
    }

    fn walk(
        &mut self,
        tree: &gix::Tree<'_>,
        path: &mut Vec<String>,
        blobs: &mut Vec<BlobJob>,
    ) -> Result<()> {
        bail_if_cancelled(&self.cancel)?;
        let entries = collect_entries(tree)?;
        self.tree_page(path, &entries)?;

        for (name, kind, oid) in entries {
            path.push(name);
            match kind {
                EntryKind::Dir => {
                    let subtree = self.site.repo.find_object(oid)?.try_into_tree()?;
                    self.walk(&subtree, path, blobs)?;
                }
                EntryKind::File | EntryKind::Link => {
                    blobs.push(BlobJob {
                        chrome: format!("{}{}", self.switcher(), self.crumbs(path, true)),
                        path: path.clone(),
                        oid,
                        is_link: matches!(kind, EntryKind::Link),
                    });
                }
                EntryKind::Submodule => {}
            }
            path.pop();
        }
        Ok(())
    }

    fn tree_page(&mut self, path: &[String], entries: &[Entry]) -> Result<()> {
        let rel = if path.is_empty() {
            "index.html".to_string()
        } else {
            format!("tree/{}/index.html", path.join("/"))
        };
        let mut body = if path.is_empty() {
            "<a class=\"back\" href=\"/\">← back</a>\n".to_string()
        } else {
            String::new()
        };
        body.push_str(&self.site.repo_header());
        body.push_str(&commit_panel(self.site.repo, self.tip_id));
        body.push_str(&self.tree_topbar(path, entries));
        if path.is_empty() {
            body.push_str(&self.overview(entries));
        } else {
            body.push_str("<h2 class=\"file-heading\">files</h2>\n");
            body.push_str(&self.listing(path, entries));
        }
        body.push_str(&self.readme_section(path, entries));

        let title = if path.is_empty() {
            format!("{} @ {}", self.site.name, self.label)
        } else {
            format!("{}/ - {} @ {}", path.join("/"), self.site.name, self.label)
        };
        let description = if path.is_empty() {
            format!("Source tree for {} at {}.", self.site.name, self.label)
        } else {
            format!(
                "{} in {} at {}.",
                path.join("/"),
                self.site.name,
                self.label,
            )
        };
        write_file(
            &self.out.join(&rel),
            page(&self.site.instance_name, &title, &description, Some((&self.label, self.tip_id)), &body),
        )
    }

    /// Root-page layout: file listing beside a recent-commits log.
    fn overview(&self, entries: &[Entry]) -> String {
        format!(
            "<div class=\"overview\"><section class=\"files\"><h2>files</h2>\n{}</section><aside class=\"commits\">{}</aside></div>\n",
            self.listing(&[], entries),
            self.log_section(),
        )
    }

    /// Recent commits from this ref's tip, jj-style: change id (from the
    /// `change-id` commit header jj can write), short sha, summary.
    fn log_section(&self) -> String {
        const SHOWN: usize = 10;
        let mut s = String::from("<h2 class=\"log-heading\">recent commits</h2>\n<ol class=\"log\">\n");
        let mut shown = Vec::with_capacity(SHOWN);
        let mut parents = Vec::new();
        let mut truncated = false;
        let Ok(walk) = self.site.repo.rev_walk([self.tip_id]).all() else {
            return String::new();
        };
        for (n, info) in walk.flatten().enumerate() {
            if n == SHOWN {
                truncated = true;
                break;
            }
            let Ok(commit) = self
                .site
                .repo
                .find_object(info.id)
                .map_err(anyhow::Error::from)
                .and_then(|o| Ok(o.try_into_commit()?))
            else {
                continue;
            };
            shown.push(info.id);
            parents.extend(commit.parent_ids().map(|id| id.detach()));
            let change_id = commit
                .decode()
                .ok()
                .and_then(|c| c.extra_headers().find("change-id").map(|v| v.to_string()));
            let cid = change_id
                .as_deref()
                .map(|c| {
                    format!(
                        "<span class=\"cid\" data-commit=\"{}\">{}</span> ",
                        info.id,
                        escape(&c[..c.len().min(8)]),
                    )
                })
                .unwrap_or_default();
            let summary = commit.message().map(|m| m.summary().to_string()).unwrap_or_default();
            let author = commit.author().map(|a| a.name.to_string()).unwrap_or_default();
            let _ = writeln!(
                s,
                "<li data-oid=\"{full}\">{cid}<span class=\"sha\" data-commit=\"{full}\">{sha}</span> <span class=\"who\"><span>{author}</span>{time}</span><span class=\"msg\">{summary}</span></li>",
                full = info.id,
                sha = commit.id().shorten_or_id(),
                author = escape(&author),
                time = time_of(&commit),
                summary = escape(&summary),
            );
        }
        s.push_str("</ol>\n");
        if truncated {
            let mut frontier = Vec::new();
            for parent in parents {
                if !shown.contains(&parent) && !frontier.contains(&parent) {
                    frontier.push(parent);
                }
            }
            let frontier = frontier
                .iter()
                .map(ToString::to_string)
                .collect::<Vec<_>>()
                .join(" ");
            let _ = writeln!(
                s,
                "<p class=\"meta log-pagination\" data-log-frontier=\"{frontier}\">⋯ older commits not shown</p>",
            );
        }
        s
    }

    fn listing(&self, path: &[String], entries: &[Entry]) -> String {
        let mut sub = path.join("/");
        if !sub.is_empty() {
            sub.push('/');
        }
        let mut body = "<table class=\"list\">\n".to_string();
        for (name, kind, oid) in entries {
            match kind {
                EntryKind::Dir => {
                    let name = collapse_lone_dirs(self.site.repo, name, *oid);
                    let _ = writeln!(
                        body,
                        "<tr><td><a href=\"{base}tree/{href}/\">{name}/</a></td><td class=\"size\"></td></tr>",
                        base = self.urlbase,
                        href = encode_path(&format!("{sub}{name}")),
                        name = escape(&name),
                    );
                }
                EntryKind::File | EntryKind::Link => {
                    let size = self.site.repo.find_header(*oid).map(|h| h.size()).unwrap_or(0);
                    let _ = writeln!(
                        body,
                        "<tr><td><a href=\"{base}blob/{href}\">{name}{sigil}</a></td><td class=\"size\">{size}</td></tr>",
                        base = self.urlbase,
                        href = encode_path(&format!("{sub}{name}")),
                        name = escape(name),
                        sigil = if matches!(kind, EntryKind::Link) { "@" } else { "" },
                        size = human_size(size),
                    );
                }
                EntryKind::Submodule => {
                    let _ = writeln!(
                        body,
                        "<tr><td>{name} @ {oid:.8}</td><td class=\"size\"></td></tr>",
                        name = escape(name),
                    );
                }
            }
        }
        body.push_str("</table>\n");
        body
    }

    fn readme_section(&mut self, path: &[String], entries: &[Entry]) -> String {
        match find_readme(self.site.repo, entries) {
            Some((name, data)) => {
                let text = String::from_utf8_lossy(&data);
                let readme = if is_markdown(&name) {
                    let urls = RelativeUrls {
                        repo: self.site.repo,
                        base_url: &self.site.base_url,
                        urlbase: &self.urlbase,
                        commit: self.tip_id,
                        directory: path,
                    };
                    markdown(&mut self.hl, &text, |url, image| urls.resolve(url, image))
                } else if is_typst(&name) {
                    let urls = RelativeUrls {
                        repo: self.site.repo,
                        base_url: &self.site.base_url,
                        urlbase: &self.urlbase,
                        commit: self.tip_id,
                        directory: path,
                    };
                    typst(&mut self.hl, &text, |url, image| urls.resolve(url, image))
                } else {
                    format!("<pre>{}</pre>\n", escape(&text))
                };
                format!("<section class=\"readme\">\n{readme}</section>\n")
            }
            None => String::new(),
        }
    }

    fn tree_topbar(&self, path: &[String], entries: &[Entry]) -> String {
        let folders = entries.iter().filter(|(_, kind, _)| matches!(kind, EntryKind::Dir)).count();
        let files = entries.len() - folders;
        let mut stats = String::from("<span class=\"view-stats\">");
        if folders > 0 {
            let _ = write!(stats, "<span>{folders} folder{}</span>", if folders == 1 { "" } else { "s" });
        }
        if files > 0 {
            let _ = write!(stats, "<span>{files} file{}</span>", if files == 1 { "" } else { "s" });
        }
        stats.push_str("</span>");
        // the empty actions slot is where the client puts the history link
        format!(
            "<header class=\"topbar\">{}{}{stats}<span class=\"actions\"></span></header>\n",
            self.switcher(),
            self.crumbs(path, false),
        )
    }

    /// A no-JS `<details>` dropdown listing all refs. Baked at build time, so
    /// pages of refs untouched since a ref was added/deleted list it stale;
    /// the always-rebuilt index and refs pages stay fresh.
    fn switcher(&self) -> String {
        let mut s = format!(
            "<details class=\"switcher\"><summary>{}</summary><div class=\"menu\">",
            escape(&self.label),
        );
        for (kind, heading) in [(RefKind::Branch, "branches"), (RefKind::Tag, "tags")] {
            let mut group = self
                .tips
                .iter()
                .filter(|t| t.kind == kind && t.static_mode.is_some())
                .peekable();
            if group.peek().is_none() {
                continue;
            }
            let _ = write!(s, "<strong>{heading}</strong>");
            for t in group {
                let current = t.name == self.label;
                let _ = write!(
                    s,
                    "<a{class} href=\"{href}\">{name}</a>",
                    class = if current { " class=\"current\"" } else { "" },
                    href = self.site.ref_href(t),
                    name = escape(&t.name),
                );
            }
        }
        let _ = write!(
            s,
            "<a class=\"all\" href=\"{}refs\">all refs →</a></div></details>",
            self.site.base_url,
        );
        s
    }

    /// Breadcrumb nav: site name / path components, all but the last linked.
    fn crumbs(&self, path: &[String], last_is_file: bool) -> String {
        let mut nav = format!(
            "<nav class=\"crumbs\"><a href=\"{href}\">{site}</a>",
            href = self.site.base_url,
            site = escape(self.site.crumb_name()),
        );
        for (i, comp) in path.iter().enumerate() {
            let is_last = i + 1 == path.len();
            if is_last && last_is_file {
                let _ = write!(nav, " / <span class=\"cur\">{}</span>", escape(comp));
            } else if is_last {
                let _ = write!(nav, " / <span class=\"cur\">{}</span> /", escape(comp));
            } else {
                let _ = write!(
                    nav,
                    " / <a href=\"{base}tree/{href}/\">{name}</a>",
                    base = self.urlbase,
                    href = encode_path(&path[..=i].join("/")),
                    name = escape(comp),
                );
            }
        }
        nav.push_str("</nav>\n");
        nav
    }
}

/// Author time as a `<time>`: absolute in `datetime` and, on hover, in
/// `title`; date-only as text, which the client relativizes while recent.
/// The in-browser client also renders author time, so dates match site-wide.
fn time_of(commit: &gix::Commit<'_>) -> String {
    author_time(commit)
        .map(|t| {
            format!(
                "<time datetime=\"{}\" title=\"{}\">{}</time>",
                t.format_or_unix(gix::date::time::format::ISO8601_STRICT),
                t.format_or_unix(gix::date::time::format::ISO8601),
                t.format_or_unix(gix::date::time::format::SHORT),
            )
        })
        .unwrap_or_default()
}

pub(crate) fn author_time(commit: &gix::Commit<'_>) -> Option<gix::date::Time> {
    commit.author().ok()?.time().ok()
}

fn find_readme(repo: &gix::Repository, entries: &[Entry]) -> Option<(String, Vec<u8>)> {
    ["readme.md", "readme", "readme.txt", "readme.typ"].iter().find_map(|want| {
        if *want == "readme.typ" && !cfg!(feature = "typst") { return None; }
        entries.iter().find_map(|(name, kind, oid)| {
            (matches!(kind, EntryKind::File) && name.to_lowercase() == *want)
                .then(|| Some((name.clone(), repo.find_object(*oid).ok()?.data.to_vec())))
                .flatten()
        })
    })
}

fn is_markdown(name: &str) -> bool {
    let name = name.to_lowercase();
    name.ends_with(".md") || name.ends_with(".markdown")
}

fn is_typst(name: &str) -> bool {
    let name = name.to_lowercase();
    cfg!(feature = "typst") && name.ends_with(".typ")
}

pub(crate) fn image_mime(path: &str) -> Option<&'static str> {
    match Path::new(path)
        .extension()?
        .to_str()?
        .to_ascii_lowercase()
        .as_str()
    {
        "png" => Some("image/png"),
        "jpg" | "jpeg" => Some("image/jpeg"),
        "gif" => Some("image/gif"),
        "webp" => Some("image/webp"),
        "avif" => Some("image/avif"),
        "bmp" => Some("image/bmp"),
        "ico" => Some("image/x-icon"),
        _ => None,
    }
}

/// Corrects a markdown file's relative URLs :D
struct RelativeUrls<'a> {
    repo: &'a gix::Repository,
    base_url: &'a str,
    urlbase: &'a str,
    commit: gix::ObjectId,
    directory: &'a [String],
}

impl RelativeUrls<'_> {
    fn resolve(&self, url: &str, image: bool) -> Option<String> {
        if url.starts_with(['/', '#']) || url.contains(':') {
            return None;
        }
        let end = url.find(['?', '#']).unwrap_or(url.len());
        let suffix = &url[end..];
        let decoded = percent_encoding::percent_decode_str(&url[..end])
            .decode_utf8()
            .ok()?;
        let mut path = self.directory.to_vec();
        for part in decoded.split('/') {
            match part {
                "" | "." => {}
                ".." => {
                    path.pop()?;
                }
                part => path.push(part.to_string()),
            }
        }
        if path.is_empty() {
            return (!image).then(|| format!("{}{suffix}", self.urlbase));
        }
        let joined = path.join("/");
        let tree = self.repo.find_commit(self.commit).ok()?.tree().ok()?;
        let entry = tree.lookup_entry_by_path(&joined).ok()??;
        let mode = entry.mode();
        let href = encode_path(&joined);
        if image {
            mode.is_blob()
                .then(|| format!("{}raw/{}/{href}{suffix}", self.base_url, entry.id()))
        } else if mode.is_tree() {
            Some(format!("{}tree/{href}/{suffix}", self.urlbase))
        } else if mode.is_blob_or_symlink() {
            Some(format!("{}blob/{href}{suffix}", self.urlbase))
        } else {
            None
        }
    }
}

fn markdown(
    hl: &mut Highlighter,
    src: &str,
    mut resolve: impl FnMut(&str, bool) -> Option<String>,
) -> String {
    use pulldown_cmark::{html, CodeBlockKind, Event, Options, Parser, Tag, TagEnd};
    let opts = Options::ENABLE_TABLES
        | Options::ENABLE_STRIKETHROUGH
        | Options::ENABLE_FOOTNOTES
        | Options::ENABLE_TASKLISTS;
    // Buffer fenced code blocks so they get the same arborium treatment as
    // blob pages; escape raw HTML rather than pulling in a sanitizer.
    let mut fence: Option<(Option<&'static str>, String)> = None;
    let events = Parser::new_ext(src, opts).filter_map(|ev| match ev {
        Event::Start(Tag::CodeBlock(CodeBlockKind::Fenced(info))) => {
            // only the first word names the language: ```rust,ignore
            let lang = info
                .split(|c: char| c == ',' || c.is_whitespace())
                .next()
                .and_then(crate::grammar_manifest::canonical);
            fence = Some((lang, String::new()));
            None
        }
        Event::Text(t) if fence.is_some() => {
            fence.as_mut().expect("checked in guard").1.push_str(&t);
            None
        }
        Event::End(TagEnd::CodeBlock) if fence.is_some() => {
            let (lang, code) = fence.take().expect("checked in guard");
            let class = lang.map(|l| format!(" class=\"language-{l}\"")).unwrap_or_default();
            let code = lang
                .and_then(|lang| hl.highlight(lang, &code).ok())
                .unwrap_or_else(|| escape(&code).to_string());
            Some(Event::Html(format!("<pre{class}>{code}</pre>\n").into()))
        }
        Event::Html(s) => Some(Event::Text(s)),
        Event::Start(Tag::Link {
            link_type,
            dest_url,
            title,
            id,
        }) => {
            let dest_url = resolve(&dest_url, false).map(Into::into).unwrap_or(dest_url);
            Some(Event::Start(Tag::Link {
                link_type,
                dest_url,
                title,
                id,
            }))
        }
        Event::Start(Tag::Image {
            link_type,
            dest_url,
            title,
            id,
        }) => {
            let dest_url = resolve(&dest_url, true).map(Into::into).unwrap_or(dest_url);
            Some(Event::Start(Tag::Image {
                link_type,
                dest_url,
                title,
                id,
            }))
        }
        Event::InlineHtml(s) => Some(Event::Text(s)),
        ev => Some(ev),
    });
    let mut out = String::new();
    html::push_html(&mut out, events);
    out
}

#[cfg(not(feature = "typst"))]
fn typst(
    _hl: &mut Highlighter,
    src: &str,
    _resolve: impl FnMut(&str, bool) -> Option<String>,
) -> String {
    format!("<pre>{}</pre>\n", escape(src))
}

#[cfg(feature = "typst")]
fn typst(
    hl: &mut Highlighter,
    src: &str,
    mut resolve: impl FnMut(&str, bool) -> Option<String>,
) -> String {
    use typst_syntax::{ast::{self, AstNode, Expr}, parse};

    fn image(call: ast::FuncCall<'_>) -> Option<(String, String)> {
        let Expr::Ident(name) = call.callee() else { return None };
        if name.get() != "image" { return None; }
        let mut path = None;
        let mut alt = String::new();
        for arg in call.args().items() {
            match arg {
                ast::Arg::Pos(Expr::Str(value)) if path.is_none() => path = Some(value.get().to_string()),
                ast::Arg::Named(named) if named.name().get() == "alt" => {
                    let Expr::Str(value) = named.expr() else { return None };
                    alt = value.get().to_string();
                }
                // Layout options do not affect the HTML preview.
                ast::Arg::Named(named) if matches!(named.name().get().as_str(), "width" | "height" | "fit") => {}
                _ => return None,
            }
        }
        Some((path?, alt))
    }

    fn markup(value: ast::Markup<'_>, hl: &mut Highlighter, resolve: &mut impl FnMut(&str, bool) -> Option<String>, paragraphs: bool) -> String {
        let mut out = String::new();
        let mut paragraph = false;
        let mut list = "";
        for expr in value.exprs() {
            let list_kind = match expr {
                Expr::ListItem(_) => "ul",
                Expr::EnumItem(_) => "ol",
                Expr::TermItem(_) => "dl",
                _ => "",
            };
            if matches!(expr, Expr::Space(_) | Expr::Parbreak(_)) && !list.is_empty() {
                continue;
            }
            if list != list_kind && !list.is_empty() {
                let _ = writeln!(out, "</{list}>");
                list = "";
            }
            let block = !list_kind.is_empty() || matches!(expr, Expr::Heading(_) | Expr::Parbreak(_))
                || matches!(expr, Expr::Raw(raw) if raw.block());
            if block && paragraph {
                out.push_str("</p>\n");
                paragraph = false;
            }
            if !list_kind.is_empty() && list.is_empty() {
                list = list_kind;
                let _ = writeln!(out, "<{list}>");
            }
            if !block && !paragraph && paragraphs {
                if matches!(expr, Expr::Space(_)) { continue; }
                out.push_str("<p>");
                paragraph = true;
            }
            match expr {
                Expr::Text(text) => out.push_str(&escape(text.get())),
                Expr::Space(_) => out.push(' '),
                Expr::Parbreak(_) => {},
                Expr::Linebreak(_) => out.push_str("<br>\n"),
                Expr::Escape(value) => out.push_str(&escape(&value.get().to_string())),
                Expr::Shorthand(value) => out.push_str(&escape(&value.get().to_string())),
                Expr::Strong(value) => {
                    let body = markup_inline(value.body(), hl, resolve);
                    let _ = write!(out, "<strong>{body}</strong>");
                }
                Expr::Emph(value) => {
                    let body = markup_inline(value.body(), hl, resolve);
                    let _ = write!(out, "<em>{body}</em>");
                }
                Expr::Heading(value) => {
                    let level = value.depth().get().min(6);
                    let body = markup_inline(value.body(), hl, resolve);
                    let _ = writeln!(out, "<h{level}>{body}</h{level}>");
                }
                Expr::ListItem(value) => {
                    let body = markup(value.body(), hl, resolve, true);
                    let _ = writeln!(out, "<li>{body}</li>");
                }
                Expr::EnumItem(value) => {
                    let number = value.number().map(|n| format!(" value=\"{n}\"")).unwrap_or_default();
                    let body = markup(value.body(), hl, resolve, true);
                    let _ = writeln!(out, "<li{number}>{body}</li>");
                }
                Expr::TermItem(value) => {
                    let term = markup_inline(value.term(), hl, resolve);
                    let body = markup(value.description(), hl, resolve, true);
                    let _ = writeln!(out, "<dt>{term}</dt><dd>{body}</dd>");
                }
                Expr::Link(value) => {
                    let target = value.get().as_str();
                    let href = resolve(target, false).unwrap_or_else(|| target.to_owned());
                    let _ = write!(out, "<a href=\"{}\">{}</a>", escape(&href), escape(target));
                }
                Expr::Raw(raw) => {
                    let code = raw.lines().map(|line| line.get().as_str()).collect::<Vec<_>>().join("\n");
                    if raw.block() {
                        let lang = raw.lang().and_then(|lang| crate::grammar_manifest::canonical(lang.get()));
                        let class = lang.map(|lang| format!(" class=\"language-{lang}\"")).unwrap_or_default();
                        let code = lang.and_then(|lang| hl.highlight(lang, &code).ok()).unwrap_or_else(|| escape(&code).to_string());
                        let _ = writeln!(out, "<pre{class}>{code}</pre>");
                    } else {
                        let _ = write!(out, "<code>{}</code>", escape(&code));
                    }
                }
                Expr::FuncCall(call) if image(call).is_some() => {
                    let (path, alt) = image(call).unwrap();
                    let url = resolve(&path, true).unwrap_or(path);
                    let _ = write!(out, "<img src=\"{}\" alt=\"{}\">", escape(&url), escape(&alt));
                }
                Expr::SmartQuote(_) => out.push_str(&escape(&expr.to_untyped().full_text())),
                _ => {
                    let source = expr.to_untyped().full_text();
                    let prefix = if matches!(expr, Expr::Equation(_) | Expr::Label(_) | Expr::Ref(_)) { "" } else { "#" };
                    let _ = write!(out, "<code>{prefix}{}</code>", escape(&source));
                }
            }
        }
        if paragraph { out.push_str("</p>\n"); }
        if !list.is_empty() { let _ = writeln!(out, "</{list}>"); }
        out
    }

    fn markup_inline(value: ast::Markup<'_>, hl: &mut Highlighter, resolve: &mut impl FnMut(&str, bool) -> Option<String>) -> String {
        markup(value, hl, resolve, false)
    }

    let tree = parse(src);
    if tree.errors_and_warnings().0.is_empty() {
        if let Some(root) = tree.cast::<ast::Markup>() {
            return markup(root, hl, &mut resolve, true);
        }
    }
    format!("<pre>{}</pre>\n", escape(src))
}

fn human_size(bytes: u64) -> String {
    const UNITS: &[&str] = &["KiB", "MiB", "GiB", "TiB"];
    if bytes < 1024 {
        return format!("{bytes} B");
    }
    let mut size = bytes as f64;
    let mut unit = "";
    for u in UNITS {
        size /= 1024.0;
        unit = u;
        if size < 1024.0 {
            break;
        }
    }
    format!("{size:.1} {unit}")
}

#[cfg(test)]
mod tests {
    use std::collections::BTreeMap;
    use std::fs;
    use std::os::unix::fs::symlink;

    use anyhow::Result;

    use crate::highlight::Highlighter;
    use crate::testutil::{TempDir, commit, git, grammar_cache, init_repo};

    use super::{
        BlobVersion, EntryKind, RefKind, collapse_lone_dirs, collect_entries, encode_path,
        find_readme, head_tip, human_size, list_refs, markdown, reuse_blob, time_of, typst,
    };

    #[test]
    fn encodes_url_paths() {
        assert_eq!(encode_path("a b/c?d#\u{e9}").to_string(), "a%20b/c%3Fd%23%C3%A9");
    }

    #[test]
    fn human_sizes_switch_units_at_1024() {
        assert_eq!(human_size(0), "0 B");
        assert_eq!(human_size(1023), "1023 B");
        assert_eq!(human_size(1024), "1.0 KiB");
        assert_eq!(human_size(1536), "1.5 KiB");
        assert_eq!(human_size(1 << 20), "1.0 MiB");
        assert_eq!(human_size(1 << 40), "1.0 TiB");
        assert_eq!(human_size(1 << 50), "1024.0 TiB");
    }

    #[test]
    fn commit_time_carries_absolute_timestamp() -> Result<()> {
        let root = TempDir::new("time-of");
        let repo_path = root.join("repo");
        init_repo(&repo_path)?;
        fs::write(repo_path.join("file"), "")?;
        git(&repo_path, &["add", "-A"])?;
        git(&repo_path, &["commit", "-qm", "first", "--date=@978307200 +0000"])?;
        let repo = gix::open(&repo_path)?;
        let commit = repo.head_commit()?;
        assert_eq!(
            time_of(&commit),
            "<time datetime=\"2001-01-01T00:00:00+00:00\" title=\"2001-01-01 00:00:00 +0000\">2001-01-01</time>",
        );
        Ok(())
    }

    #[test]
    fn tree_entries_list_directories_first_then_by_name() -> Result<()> {
        let root = TempDir::new("entries");
        let repo_path = root.join("repo");
        init_repo(&repo_path)?;
        for dir in ["zeta", "alpha"] {
            fs::create_dir(repo_path.join(dir))?;
            fs::write(repo_path.join(dir).join("keep"), "")?;
        }
        fs::write(repo_path.join("beta"), "")?;
        fs::write(repo_path.join("aardvark"), "")?;
        symlink("beta", repo_path.join("gamma"))?;
        commit(&repo_path, "first")?;

        let repo = gix::open(&repo_path)?;
        let entries = collect_entries(&repo.head_commit()?.tree()?)?;
        let names: Vec<_> = entries.iter().map(|(name, ..)| name.as_str()).collect();
        assert_eq!(names, ["alpha", "zeta", "aardvark", "beta", "gamma"]);
        assert!(matches!(entries[1].1, EntryKind::Dir));
        assert!(matches!(entries[2].1, EntryKind::File));
        assert!(matches!(entries[4].1, EntryKind::Link));
        Ok(())
    }

    #[test]
    fn lone_directory_chains_collapse_into_one_name() -> Result<()> {
        let root = TempDir::new("collapse");
        let repo_path = root.join("repo");
        init_repo(&repo_path)?;
        fs::create_dir_all(repo_path.join("a/b/c"))?;
        fs::write(repo_path.join("a/b/c/keep"), "")?;
        fs::create_dir_all(repo_path.join("x/y"))?;
        fs::write(repo_path.join("x/y/keep"), "")?;
        fs::write(repo_path.join("x/stop"), "")?;
        commit(&repo_path, "first")?;

        let repo = gix::open(&repo_path)?;
        let entries = collect_entries(&repo.head_commit()?.tree()?)?;
        let collapsed: Vec<_> = entries
            .iter()
            .map(|(name, _, oid)| collapse_lone_dirs(&repo, name, *oid))
            .collect();
        assert_eq!(collapsed, ["a/b/c", "x"]);
        Ok(())
    }

    #[test]
    fn readme_lookup_prefers_markdown_ignores_case_and_skips_directories() -> Result<()> {
        let root = TempDir::new("readme");
        let repo_path = root.join("repo");
        init_repo(&repo_path)?;
        fs::write(repo_path.join("readme.txt"), "txt")?;
        fs::write(repo_path.join("README"), "bare")?;
        fs::write(repo_path.join("ReadMe.MD"), "md")?;
        commit(&repo_path, "first")?;
        let repo = gix::open(&repo_path)?;
        let entries = collect_entries(&repo.head_commit()?.tree()?)?;
        let (name, data) = find_readme(&repo, &entries).unwrap();
        assert_eq!((name.as_str(), data.as_slice()), ("ReadMe.MD", b"md".as_slice()));

        fs::remove_file(repo_path.join("ReadMe.MD"))?;
        fs::create_dir(repo_path.join("readme.md"))?;
        fs::write(repo_path.join("readme.md/keep"), "")?;
        commit(&repo_path, "second")?;
        let entries = collect_entries(&repo.head_commit()?.tree()?)?;
        let (name, data) = find_readme(&repo, &entries).unwrap();
        assert_eq!((name.as_str(), data.as_slice()), ("README", b"bare".as_slice()));
        Ok(())
    }

    #[test]
    fn refs_skip_non_commit_tags_and_let_branches_shadow_tags() -> Result<()> {
        let root = TempDir::new("refs");
        let repo_path = root.join("repo");
        init_repo(&repo_path)?;
        fs::write(repo_path.join("file"), "")?;
        commit(&repo_path, "first")?;
        git(&repo_path, &["tag", "main"])?;
        git(&repo_path, &["tag", "v1"])?;
        let blob = git(&repo_path, &["rev-parse", "HEAD:file"])?;
        git(&repo_path, &["tag", "blobtag", &blob])?;

        let repo = gix::open(&repo_path)?;
        let mut tips: Vec<_> = list_refs(&repo)?
            .into_iter()
            .map(|tip| (tip.kind, tip.name))
            .collect();
        tips.sort_by(|a, b| a.1.cmp(&b.1));
        assert_eq!(tips, [(RefKind::Branch, "main".into()), (RefKind::Tag, "v1".into())]);
        Ok(())
    }

    #[test]
    fn head_tip_follows_head_and_falls_back_to_any_branch() -> Result<()> {
        let root = TempDir::new("head");
        let repo_path = root.join("repo");
        init_repo(&repo_path)?;
        fs::write(repo_path.join("file"), "")?;
        commit(&repo_path, "first")?;
        git(&repo_path, &["branch", "aaa"])?;
        let repo = gix::open(&repo_path)?;
        let tips = list_refs(&repo)?;
        assert_eq!(head_tip(&repo, &tips).unwrap().name, "main");

        git(&repo_path, &["checkout", "-q", "--detach"])?;
        let repo = gix::open(&repo_path)?;
        assert_eq!(head_tip(&repo, &tips).unwrap().kind, RefKind::Branch);

        git(&repo_path, &["symbolic-ref", "HEAD", "refs/heads/unborn"])?;
        let repo = gix::open(&repo_path)?;
        assert_eq!(head_tip(&repo, &tips).unwrap().kind, RefKind::Branch);
        Ok(())
    }

    #[test]
    fn markdown_never_passes_through_raw_html() {
        let mut hl = Highlighter::new(grammar_cache());
        let html = markdown(
            &mut hl,
            "<script>alert(1)</script>\n\n\
             text <b onclick=\"x\">bold</b>\n\n\
             ```no-such-lang\n<img src=x onerror=alert(1)>\n```\n",
            |_, _| None,
        );
        assert!(html.contains("&lt;script&gt;"), "{html}");
        for forbidden in ["<script", "<b ", "<img"] {
            assert!(!html.contains(forbidden), "{forbidden} leaked into {html}");
        }
    }

    #[test]
    fn typst_readme_requires_feature() -> Result<()> {
        let root = TempDir::new("typst-feature");
        let repo_path = root.join("repo");
        init_repo(&repo_path)?;
        fs::write(repo_path.join("README.TYP"), "= Title")?;
        commit(&repo_path, "typst readme")?;
        let repo = gix::open(&repo_path)?;
        let entries = collect_entries(&repo.head_commit()?.tree()?)?;
        assert_eq!(find_readme(&repo, &entries).is_some(), cfg!(feature = "typst"));
        assert_eq!(super::is_typst("README.TYP"), cfg!(feature = "typst"));
        Ok(())
    }

    #[test]
    #[cfg(not(feature = "typst"))]
    fn typst_disabled_falls_back_to_escaped_source() {
        let mut hl = Highlighter::new(grammar_cache());
        let html = typst(&mut hl, "<script> #image(\"pic.png\")", |_, _| panic!("Typst disabled"));
        assert!(html.starts_with("<pre>&lt;script&gt;"), "{html}");
        assert!(!html.contains("<img"), "{html}");
    }

    #[test]
    #[cfg(feature = "typst")]
    fn typst_renders_markup_and_blocks() {
        let mut hl = Highlighter::new(grammar_cache());
        let html = typst(&mut hl, "= Heading\n\n*bold* _italic_ https://typst.org \\# ~\n\n```\nfirst\nsecond\n```\n\n- one\n  - nested\n- two\n\n+ numbered\n\n/ Term: Description", |_, _| None);
        for expected in ["<h1>Heading</h1>", "<strong>bold</strong>", "<em>italic</em>", "<a href=\"https://typst.org\">https://typst.org</a>", "first\nsecond", "<ul>", "<ol>", "<dl>", "<dt>Term</dt>"] {
            assert!(html.contains(expected), "missing {expected}: {html}");
        }
        for forbidden in ["<p><h", "<p><pre", "<p><ul", "*bold*", "_italic_"] {
            assert!(!html.contains(forbidden), "{forbidden}: {html}");
        }
        assert_eq!(html.matches("<ul>").count(), 2, "{html}");
    }

    #[test]
    #[cfg(feature = "typst")]
    fn typst_images_resolve_and_escape() {
        let mut hl = Highlighter::new(grammar_cache());
        let html = typst(&mut hl, "#image(\"../pic.png\", alt: \"A & B\", width: 50%)", |url, image| {
            assert!(image);
            assert_eq!(url, "../pic.png");
            Some("/raw/revision/pic.png?x=1&y=2".into())
        });
        assert!(html.contains("<img src=\"/raw/revision/pic.png?x=1&amp;y=2\" alt=\"A &amp; B\">"), "{html}");
        let html = typst(&mut hl, "#image(\"https://example.com/pic.png\")", |_, _| None);
        assert!(html.contains("src=\"https://example.com/pic.png\""), "{html}");
    }

    #[test]
    #[cfg(feature = "typst")]
    fn typst_unsupported_and_invalid_input_falls_back_to_source() {
        let mut hl = Highlighter::new(grammar_cache());
        let html = typst(&mut hl, "#image(path) $x^2$ #unknown(\"<script>\")", |_, _| panic!("must not evaluate dynamic expressions"));
        assert!(html.contains("#image(path)"), "{html}");
        assert!(html.contains("$x^2$"), "{html}");
        assert!(html.contains("&lt;script&gt;"), "{html}");
        assert!(!html.contains("<script>"), "{html}");
        let html = typst(&mut hl, "#image(\"unfinished", |_, _| None);
        assert!(html.starts_with("<pre>"), "{html}");
    }

    #[test]
    fn reuses_only_matching_blob_pages() -> Result<()> {
        let root = TempDir::new("blob-reuse");
        let previous = root.join("previous");
        let out = root.join("out");
        fs::create_dir_all(previous.join("blob/src"))?;
        fs::write(previous.join("blob/src/main.rs"), "highlighted")?;
        let old = BlobVersion {
            oid: "abc123".into(),
            mode: 0o100644,
            touched: "first".into(),
        };
        let previous_versions = BTreeMap::from([("src/main.rs".into(), old.clone())]);
        let changed = BlobVersion {
            touched: "second".into(),
            ..old.clone()
        };
        let changed_versions = BTreeMap::from([("src/main.rs".into(), changed)]);
        let versions = BTreeMap::from([("src/main.rs".into(), old)]);

        assert!(!reuse_blob(
            &previous,
            &previous_versions,
            &changed_versions,
            "src/main.rs",
            &out,
        ));
        assert!(reuse_blob(
            &previous,
            &previous_versions,
            &versions,
            "src/main.rs",
            &out,
        ));
        assert_eq!(
            fs::read_to_string(out.join("blob/src/main.rs"))?,
            "highlighted",
        );
        Ok(())
    }
}