links: a rename or a move no longer breaks them

A document links to another by its path, `[the plan](/n/work/plan)`, or
now by its title, `[[Plan]]` or `[[Plan|the plan]]`. What each document
links to is kept in a new `links` table, rebuilt whenever its text is
written: from the browser, over the mount, or by an import. A title is
looked for in the linking document's space first, then in every other.

Renaming or moving a document, or a folder above it, rewrites the links
to it in every document that has one, in the same transaction: the path
to the new one, a title to the new name, with the label left as it was.
Each rewritten document keeps a revision of what it said before. A link
to something in the trash keeps its row, so it follows the thing when it
is restored somewhere else. A document created, or renamed, under a title
that a link was waiting for is found by that link from then on.

The reader draws `[[titles]]` as links, and a title no document has yet
as a dimmed one that offers to write that document. `gf` follows either
from the editor, and `[[` there completes titles from the search. `gb`
and `:backlinks` list the documents that link to the open one, with the
line each does it on; the line under the title counts them. A share page
shows these links as plain text, as it does every /n/ link.

The documents written before this are indexed once, when the server
starts and finds the table empty.

README.md

@@ -86,6 +86,7 @@ it; in the document pane it is that document. In the browse pane:
 | `gO` | the document's headings as a list — `j`/`k`, `Enter` jumps there (`:outline`) |
 | `gx` | open the file under the cursor in a tab of its own |
 | `gh` | the open document's earlier versions, as diffs — `j`/`k`, `Enter` brings one back (`:history`) |
+| `gb` | the documents that link to the open one — `j`/`k`, `Enter` opens one (`:backlinks`) |
 | `F` | a letter over every link and button in view; typing it follows it — links out of the workspace open in a tab of their own |
 | `Enter`, `l` | open the highlighted entry; on an open keystore or sealed file, `Enter` asks for its password |
 | `o` | a new document here, straight into insert mode |
@@ -117,7 +118,8 @@ mode, macros, `:w` to write. `:help` lists every ex command; the useful ones are
 `:set novim`. `:name` on its own moves up into the title; `Enter`, `Esc` or `↓`
 there goes back down into the text. In the text, `]]`, `[[` and `gO` walk the
 headings as they do in the reader. `gf` follows the link under the cursor, saving
-first, and `gx` opens it in a new tab. The rest: `:today` opens today's entry, `:edit [id]` (or `:12`)
+first, and `gx` opens it in a new tab. `[[` in insert mode offers the titles of
+documents to link to. The rest: `:today` opens today's entry, `:edit [id]` (or `:12`)
 opens a node by id, `:next`, `:prev`, `:first`, `:last` and `:random` walk the
 list (`:bn` and `:bp` walk the tiles), `:delete [id]` deletes, `:copy` puts the whole document on the clipboard,
 `:clear` drops the search, `:trash` lists what was deleted, `:reload` re-reads the folder, `:stats` counts words
@@ -313,6 +315,31 @@ the tree is never one of them: it is deleted from its folder, not from
 highlighted one at the cursor while writing, `y` copies its address and `x`
 deletes it, asking first.
 
+## Links
+
+A document links to another by its address, `[the plan](/n/work/notes/plan)`,
+or by its title, `[[Plan]]` or `[[Plan|the plan]]`. A title is looked for in
+the space the link is written in first, then in every other; it is a name in
+any case, or the slug that name has. `[[` in the editor offers the titles as
+the search finds them. A title no document has yet is drawn dimmed, and
+following it — a click, `F`, `gf` — offers to write that document in the
+folder being browsed.
+
+Links keep working when what they link to moves. Renaming a document, or a
+folder above it, or moving either somewhere else — in the browser or over the
+mount — rewrites the links to it in every document that has one, in the same
+write: the path to the new one, a `[[title]]` to the new name, the label left
+as it was. A rewritten document keeps a version of what it said before, like
+any other save. A link to something in the trash stays as it is and follows
+it when it is restored; once it is deleted for good the text of the link
+stays, and leads nowhere.
+
+`gb` or `:backlinks` lists the documents that link to the open one, each with
+the line it does so on, and the line under the title says how many there are.
+A link inside a card, or one written into prose rather than as a markdown
+link, is not counted or rewritten. The links of the documents written before
+there were any are indexed once, when the server first starts with this.
+
 ## Files
 
 A file is a node like a document, only with bytes instead of text: it sits in
@@ -692,6 +719,7 @@ sealed or not.
 | `GET` | `/api/resolve?path=` | the node behind a slug path, `work/notes/monday` |
 | `GET` | `/api/nodes/{id}/revisions` | a document's earlier versions, the newest first |
 | `GET` | `/api/nodes/{id}/revisions/{rev}` | one earlier version, with its text |
+| `GET` | `/api/nodes/{id}/backlinks` | the documents that link to this one, with the line each does it on |
 | `POST`/`DELETE` | `/api/nodes/{id}/share` | mint or rotate / revoke a share link |
 | `GET` | `/api/export` | the whole workspace, boards and dates included, as a zip |
 | `GET`/`POST` | `/api/backup` | Proton Drive mirror: status / run now |

migrations/0021_links.sql

@@ -0,0 +1,17 @@
+-- The links between documents, as their text writes them: a `/n/...` path,
+-- or a `[[title]]`. Rebuilt from a document's text whenever it is written,
+-- and read the other way round: for what links to a document, and for the
+-- texts to rewrite when what they link to is renamed or moved.
+--
+-- Nothing fills it for the documents already written; the server indexes
+-- them all when it starts and finds it empty (see `links::backfill`).
+CREATE TABLE links (
+    from_id INTEGER NOT NULL REFERENCES nodes (id) ON DELETE CASCADE,
+    kind    TEXT    NOT NULL CHECK (kind IN ('path', 'wiki')),
+    -- What the link says: the path after `/n/`, or the title in the brackets.
+    target  TEXT    NOT NULL,
+    to_id   INTEGER NOT NULL REFERENCES nodes (id) ON DELETE CASCADE,
+    PRIMARY KEY (from_id, kind, target)
+) WITHOUT ROWID;
+
+CREATE INDEX idx_links_to ON links (to_id);

src/main.rs

@@ -143,6 +143,13 @@ async fn main() -> Result<()> {
         sessions,
     };
     routes::seal::spawn_rekey(state.clone());
+    // Before the first request, so no document reads as linked from nowhere.
+    match routes::links::backfill(&state.db).await {
+        Ok(0) => {}
+        Ok(1) => tracing::info!("indexed the links of 1 document"),
+        Ok(linking) => tracing::info!("indexed the links of {linking} documents"),
+        Err(e) => tracing::warn!("could not index the links between documents: {e:?}"),
+    }
 
     if let Some(sftp_bind) = config.sftp.bind {
         let sftp_state = state.clone();

src/routes/import.rs

@@ -23,7 +23,7 @@ use crate::{
     auth::Session,
     error::{AppError, AppResult},
     routes::{
-        boards,
+        boards, links,
         manifest::{self, Manifest, NodeMeta},
         media,
         nodes::{self, next_slug, slugify, DOCUMENT, FILE, FOLDER, SPACE},
@@ -968,6 +968,7 @@ async fn build(
                 .await?;
         }
         media::link_to_node(&mut tx, *id, &rewritten).await?;
+        links::index(&mut tx, *id, &rewritten).await?;
     }
 
     // The boards come back only where spaces do.

src/routes/links.rs

@@ -0,0 +1,797 @@
+//! Links between documents. A document links to another by the path the
+//! address bar shows, `[the plan](/n/work/plan)`, or by its title,
+//! `[[Plan]]` or `[[Plan|the plan]]`.
+//!
+//! What a document links to is kept in `links`, rebuilt whenever its text is
+//! written. Read the other way round, that is what links to a document, and
+//! the texts a rename or a move would break: those are rewritten, in the same
+//! transaction, so a link keeps working however its target moves around.
+
+use std::{
+    cmp::Reverse,
+    collections::{BTreeMap, BTreeSet, HashMap},
+    sync::LazyLock,
+};
+
+use axum::{
+    extract::{Path, State},
+    Json,
+};
+use regex::Regex;
+use serde::Serialize;
+use sqlx::{Row, SqliteConnection, SqlitePool};
+
+use crate::{
+    auth::Session,
+    error::{AppError, AppResult},
+    routes::{nodes, revisions, trash},
+    state::AppState,
+};
+
+/// A markdown link to a path in the workspace, inline or by reference:
+/// `](/n/work/plan)` or `[plan]: /n/work/plan`. A path anywhere else in the
+/// prose is text about one, not a link.
+static PATH_LINK: LazyLock<Regex> = LazyLock::new(|| {
+    Regex::new(r"(?m)(?:\]\(\s*<?|^[ \t]{0,3}\[[^\]\n]+\]:[ \t]*<?)/n/([a-z0-9-]+(?:/[a-z0-9-]+)*)")
+        .expect("static regex is valid")
+});
+
+/// `[[title]]`, or `[[title|what the link reads]]`.
+static WIKILINK: LazyLock<Regex> = LazyLock::new(|| {
+    Regex::new(r"\[\[([^\[\]|\n]+)(\|[^\[\]\n]*)?\]\]").expect("static regex is valid")
+});
+
+/// How a link names what it links to.
+#[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)]
+enum Kind {
+    /// By its path, `work/plan`.
+    Path,
+    /// By its title, `Plan`.
+    Wiki,
+}
+
+impl Kind {
+    fn as_str(self) -> &'static str {
+        match self {
+            Kind::Path => "path",
+            Kind::Wiki => "wiki",
+        }
+    }
+
+    fn parse(kind: &str) -> Self {
+        if kind == "wiki" {
+            Kind::Wiki
+        } else {
+            Kind::Path
+        }
+    }
+}
+
+/// The path links in `body`, each as the span of the path after `/n/`.
+fn paths(body: &str) -> impl Iterator<Item = regex::Match<'_>> {
+    PATH_LINK.captures_iter(body).filter_map(|caps| {
+        let path = caps.get(1)?;
+        // The path has to end where the link does: `/n/work/Plan` or
+        // `/n/work/plan.md` is not a link to `work`.
+        let rest = &body[path.end()..];
+        let rest = rest.strip_prefix('/').unwrap_or(rest);
+        let ends = rest
+            .chars()
+            .next()
+            .is_none_or(|c| c.is_whitespace() || matches!(c, ')' | '#' | '?' | '>' | '"' | '\''));
+        ends.then_some(path)
+    })
+}
+
+/// What `body` links to, as it is written.
+fn written(body: &str) -> BTreeSet<(Kind, String)> {
+    let mut found: BTreeSet<(Kind, String)> = paths(body)
+        .map(|path| (Kind::Path, path.as_str().to_string()))
+        .collect();
+    for caps in WIKILINK.captures_iter(body) {
+        let title = caps[1].trim();
+        if !title.is_empty() {
+            found.insert((Kind::Wiki, title.to_string()));
+        }
+    }
+    found
+}
+
+/// `body` with the links named in `to` pointed elsewhere: each target written
+/// there becomes the one it maps to. A label stays as it was.
+fn rewrite(body: &str, to: &HashMap<(Kind, String), String>) -> String {
+    let mut out = String::with_capacity(body.len());
+    let mut last = 0;
+    for path in paths(body) {
+        if let Some(new) = to.get(&(Kind::Path, path.as_str().to_string())) {
+            out.push_str(&body[last..path.start()]);
+            out.push_str(new);
+            last = path.end();
+        }
+    }
+    out.push_str(&body[last..]);
+
+    WIKILINK
+        .replace_all(&out, |caps: &regex::Captures| {
+            match to.get(&(Kind::Wiki, caps[1].trim().to_string())) {
+                Some(new) => format!("[[{new}{}]]", caps.get(2).map_or("", |m| m.as_str())),
+                None => caps[0].to_string(),
+            }
+        })
+        .into_owned()
+}
+
+/// The node a slug path names, from a space down, as an `/n/` address
+/// names it. Never one in the trash, which is no space.
+pub async fn resolve_path(db: &mut SqliteConnection, path: &str) -> AppResult<Option<i64>> {
+    let mut current: Option<i64> = None;
+    for segment in path.split('/').filter(|s| !s.is_empty()) {
+        let found: Option<i64> = sqlx::query_scalar(
+            "SELECT id FROM nodes
+             WHERE parent_id IS ?1 AND slug = ?2 AND (?1 IS NOT NULL OR kind = 'space')",
+        )
+        .bind(current)
+        .bind(segment)
+        .fetch_optional(&mut *db)
+        .await?;
+        let Some(id) = found else {
+            return Ok(None);
+        };
+        current = Some(id);
+    }
+    Ok(current)
+}
+
+/// The space a node is in, or the node itself when it is one. `None` in the
+/// trash.
+async fn space_of(db: &mut SqliteConnection, id: i64) -> AppResult<Option<i64>> {
+    Ok(sqlx::query_scalar(
+        "WITH RECURSIVE up (id, parent_id) AS (
+             SELECT id, parent_id FROM nodes WHERE id = ?1
+             UNION ALL
+             SELECT n.id, n.parent_id FROM nodes n JOIN up ON n.id = up.parent_id
+         )
+         SELECT n.id FROM up JOIN nodes n ON n.id = up.id
+         WHERE up.parent_id IS NULL AND n.kind = 'space'",
+    )
+    .bind(id)
+    .fetch_optional(&mut *db)
+    .await?)
+}
+
+/// The document or file a `[[title]]` names: one called that, whatever the
+/// case, or one whose slug it is. In the space it is written in first, then
+/// in any; one called that before one only slugged so; the oldest of what is
+/// still left. Never one in the trash.
+async fn resolve_title(
+    db: &mut SqliteConnection,
+    title: &str,
+    space: Option<i64>,
+) -> AppResult<Option<i64>> {
+    // `[[???]]` slugs to `untitled`, which is not what it names.
+    let slug = title
+        .chars()
+        .any(char::is_alphanumeric)
+        .then(|| nodes::slugify(title));
+    let rows = sqlx::query(
+        "WITH RECURSIVE up (origin, id, parent_id) AS (
+             SELECT id, id, parent_id FROM nodes
+             WHERE kind IN ('document', 'file') AND (name = ?1 COLLATE NOCASE OR slug = ?2)
+             UNION ALL
+             SELECT up.origin, n.id, n.parent_id FROM nodes n JOIN up ON n.id = up.parent_id
+         )
+         SELECT up.origin, up.id AS space, (o.name = ?1 COLLATE NOCASE) AS named
+         FROM up JOIN nodes s ON s.id = up.id JOIN nodes o ON o.id = up.origin
+         WHERE up.parent_id IS NULL AND s.kind = 'space'",
+    )
+    .bind(title)
+    .bind(slug)
+    .fetch_all(&mut *db)
+    .await?;
+
+    Ok(rows
+        .iter()
+        .map(|row| {
+            let origin: i64 = row.get("origin");
+            let here = space.is_some_and(|space| row.get::<i64, _>("space") == space);
+            (here, row.get::<bool, _>("named"), Reverse(origin))
+        })
+        .max()
+        .map(|(_, _, Reverse(origin))| origin))
+}
+
+async fn resolve(
+    db: &mut SqliteConnection,
+    kind: Kind,
+    target: &str,
+    space: Option<i64>,
+) -> AppResult<Option<i64>> {
+    match kind {
+        Kind::Path => resolve_path(db, target).await,
+        Kind::Wiki => resolve_title(db, target, space).await,
+    }
+}
+
+/// A node's path from its space, `work/notes/plan`, as `/n/` writes it.
+async fn path_text(db: &mut SqliteConnection, id: i64) -> AppResult<String> {
+    let slugs: Vec<String> = sqlx::query_scalar(
+        "WITH RECURSIVE up (id, parent_id, slug, depth) AS (
+             SELECT id, parent_id, slug, 0 FROM nodes WHERE id = ?1
+             UNION ALL
+             SELECT n.id, n.parent_id, n.slug, up.depth + 1
+             FROM nodes n JOIN up ON n.id = up.parent_id
+         )
+         SELECT slug FROM up ORDER BY depth DESC",
+    )
+    .bind(id)
+    .fetch_all(&mut *db)
+    .await?;
+    Ok(slugs.join("/"))
+}
+
+/// What a `[[link]]` to a node says: its name, or its slug when the name
+/// could not stand between the brackets.
+async fn title_of(db: &mut SqliteConnection, id: i64) -> AppResult<String> {
+    let row = sqlx::query("SELECT name, slug FROM nodes WHERE id = ?1")
+        .bind(id)
+        .fetch_one(&mut *db)
+        .await?;
+    let name: String = row.get("name");
+    let name = name.trim();
+    Ok(if name.is_empty() || name.contains(['[', ']', '|', '\n']) {
+        row.get("slug")
+    } else {
+        name.to_string()
+    })
+}
+
+/// Record what `from` links to. Called by every write of a document's text,
+/// after it, in the same transaction.
+pub async fn index(db: &mut SqliteConnection, from: i64, body: &str) -> AppResult<()> {
+    let space = space_of(db, from).await?;
+    let before: HashMap<(Kind, String), i64> =
+        sqlx::query("SELECT kind, target, to_id FROM links WHERE from_id = ?1")
+            .bind(from)
+            .fetch_all(&mut *db)
+            .await?
+            .iter()
+            .map(|row| {
+                (
+                    (Kind::parse(row.get("kind")), row.get("target")),
+                    row.get("to_id"),
+                )
+            })
+            .collect();
+
+    let mut rows = Vec::new();
+    for (kind, target) in written(body) {
+        let to = match resolve(db, kind, &target, space).await? {
+            Some(to) => Some(to),
+            // What it linked to is in the trash. The row stays, so a restore
+            // that puts it somewhere else still finds the link to rewrite.
+            None => match before.get(&(kind, target.clone())) {
+                Some(&to) if trash::in_trash(db, to).await? => Some(to),
+                _ => None,
+            },
+        };
+        if let Some(to) = to.filter(|&to| to != from) {
+            rows.push((kind, target, to));
+        }
+    }
+
+    sqlx::query("DELETE FROM links WHERE from_id = ?1")
+        .bind(from)
+        .execute(&mut *db)
+        .await?;
+    for (kind, target, to) in rows {
+        sqlx::query("INSERT INTO links (from_id, kind, target, to_id) VALUES (?1, ?2, ?3, ?4)")
+            .bind(from)
+            .bind(kind.as_str())
+            .bind(&target)
+            .bind(to)
+            .execute(&mut *db)
+            .await?;
+    }
+    Ok(())
+}
+
+/// After `id` was created, renamed or moved, in the same transaction. Every
+/// link to it, or to anything under it, that no longer finds what it linked
+/// to is rewritten to find it again; a link that found nothing until now and
+/// names it now is recorded. Returns how many documents were rewritten.
+pub async fn follow(db: &mut SqliteConnection, id: i64) -> AppResult<usize> {
+    // Nothing in the trash is followed there: its links stay as they were
+    // written, for whatever the restore brings it back as.
+    if trash::in_trash(db, id).await? {
+        return Ok(0);
+    }
+    let rows = sqlx::query(
+        "WITH RECURSIVE sub (id) AS (
+             SELECT ?1
+             UNION ALL
+             SELECT n.id FROM nodes n JOIN sub ON n.parent_id = sub.id
+         )
+         SELECT from_id, kind, target, to_id FROM links
+         WHERE to_id IN (SELECT id FROM sub)",
+    )
+    .bind(id)
+    .fetch_all(&mut *db)
+    .await?;
+
+    let mut linking: BTreeMap<i64, Vec<(Kind, String, i64)>> = BTreeMap::new();
+    for row in &rows {
+        linking.entry(row.get("from_id")).or_default().push((
+            Kind::parse(row.get("kind")),
+            row.get("target"),
+            row.get("to_id"),
+        ));
+    }
+
+    let mut rewritten = 0;
+    for (from, links) in linking {
+        let space = space_of(db, from).await?;
+        let mut to = HashMap::new();
+        for (kind, target, node) in links {
+            if resolve(db, kind, &target, space).await? == Some(node) {
+                continue;
+            }
+            let now = match kind {
+                Kind::Path => path_text(db, node).await?,
+                Kind::Wiki => title_of(db, node).await?,
+            };
+            to.insert((kind, target), now);
+        }
+        if to.is_empty() {
+            continue;
+        }
+
+        let body: String = sqlx::query_scalar("SELECT body FROM nodes WHERE id = ?1")
+            .bind(from)
+            .fetch_one(&mut *db)
+            .await?;
+        let body_now = rewrite(&body, &to);
+        if body_now == body {
+            continue;
+        }
+        revisions::keep(db, from, &body_now, revisions::Reason::Edit).await?;
+        // `updated_at` moves, so an editor open on the old text is told it
+        // changed rather than saving the broken links back over it.
+        sqlx::query("UPDATE nodes SET body = ?1, updated_at = ?2 WHERE id = ?3")
+            .bind(&body_now)
+            .bind(crate::now())
+            .bind(from)
+            .execute(&mut *db)
+            .await?;
+        index(db, from, &body_now).await?;
+        rewritten += 1;
+    }
+
+    claim(db, id).await?;
+    Ok(rewritten)
+}
+
+/// Index again the documents that may have been waiting for `id`: whose
+/// `[[title]]` or path named nothing until it was called that, or put there.
+/// A rough cut by text, as the index then decides what each link finds.
+async fn claim(db: &mut SqliteConnection, id: i64) -> AppResult<()> {
+    let row = sqlx::query("SELECT name, slug FROM nodes WHERE id = ?1")
+        .bind(id)
+        .fetch_one(&mut *db)
+        .await?;
+    // `lower` in SQLite folds ASCII only, so the name is folded the same way.
+    let name = row.get::<String, _>("name").trim().to_ascii_lowercase();
+    let slug: String = row.get("slug");
+    let path = path_text(db, id).await?;
+    let waiting: Vec<(i64, String)> = sqlx::query_as(
+        "SELECT id, body FROM nodes
+         WHERE kind = 'document' AND id <> ?1
+           AND (instr(body, '/n/' || ?4) > 0
+                OR (instr(body, '[[') > 0
+                    AND ((?2 <> '' AND instr(lower(body), ?2) > 0) OR instr(body, ?3) > 0)))",
+    )
+    .bind(id)
+    .bind(&name)
+    .bind(&slug)
+    .bind(&path)
+    .fetch_all(&mut *db)
+    .await?;
+    for (from, body) in waiting {
+        index(db, from, &body).await?;
+    }
+    Ok(())
+}
+
+/// The documents that link to `id`, the last written first; none in the
+/// trash.
+async fn linking(db: &mut SqliteConnection, id: i64) -> AppResult<Vec<i64>> {
+    let from: Vec<i64> = sqlx::query_scalar(
+        "SELECT DISTINCT l.from_id FROM links l JOIN nodes n ON n.id = l.from_id
+         WHERE l.to_id = ?1 ORDER BY n.updated_at DESC, n.id",
+    )
+    .bind(id)
+    .fetch_all(&mut *db)
+    .await?;
+    let mut live = Vec::with_capacity(from.len());
+    for from in from {
+        if !trash::in_trash(db, from).await? {
+            live.push(from);
+        }
+    }
+    Ok(live)
+}
+
+/// How many documents link to `id`, for the line under its title.
+pub async fn linked_from(db: &mut SqliteConnection, id: i64) -> AppResult<i64> {
+    Ok(linking(db, id).await?.len() as i64)
+}
+
+/// Where the `[[titles]]` in `body` lead, as `/n/` paths, for the reader to
+/// draw them as links. A title that leads nowhere is left out.
+pub async fn wikilinks(
+    db: &mut SqliteConnection,
+    from: i64,
+    body: &str,
+) -> AppResult<BTreeMap<String, String>> {
+    let mut found = BTreeMap::new();
+    let titles: Vec<String> = written(body)
+        .into_iter()
+        .filter(|(kind, _)| *kind == Kind::Wiki)
+        .map(|(_, title)| title)
+        .collect();
+    if titles.is_empty() {
+        return Ok(found);
+    }
+    let space = space_of(db, from).await?;
+    for title in titles {
+        if let Some(to) = resolve_title(db, &title, space).await? {
+            found.insert(title, path_text(db, to).await?);
+        }
+    }
+    Ok(found)
+}
+
+#[derive(Serialize)]
+pub struct Backlink {
+    pub id: i64,
+    pub name: String,
+    pub slug: String,
+    /// Where it sits, `work/notes/plan`.
+    pub path: String,
+    /// The line the link is on.
+    pub context: String,
+}
+
+/// How much of the line a link is on a backlink shows.
+const CONTEXT_CHARS: usize = 200;
+
+/// `GET /api/nodes/{id}/backlinks` — the documents that link here.
+pub async fn backlinks(
+    _: Session,
+    State(state): State<AppState>,
+    Path(id): Path<i64>,
+) -> AppResult<Json<Vec<Backlink>>> {
+    let mut conn = state.db.acquire().await?;
+    if trash::in_trash(&mut conn, id).await? {
+        return Err(AppError::NotFound);
+    }
+    let mut found = Vec::new();
+    for from in linking(&mut conn, id).await? {
+        let row = sqlx::query("SELECT name, slug, body FROM nodes WHERE id = ?1")
+            .bind(from)
+            .fetch_one(&mut *conn)
+            .await?;
+        let spellings: Vec<String> =
+            sqlx::query("SELECT kind, target FROM links WHERE from_id = ?1 AND to_id = ?2")
+                .bind(from)
+                .bind(id)
+                .fetch_all(&mut *conn)
+                .await?
+                .iter()
+                .map(|link| match Kind::parse(link.get("kind")) {
+                    Kind::Path => format!("/n/{}", link.get::<String, _>("target")),
+                    Kind::Wiki => format!("[[{}", link.get::<String, _>("target")),
+                })
+                .collect();
+        let body: String = row.get("body");
+        let context = body
+            .lines()
+            .find(|line| {
+                spellings
+                    .iter()
+                    .any(|spelling| line.contains(spelling.as_str()))
+            })
+            .unwrap_or_default()
+            .trim();
+        let mut context: String = context.chars().take(CONTEXT_CHARS).collect();
+        if context.chars().count() == CONTEXT_CHARS {
+            context.push('…');
+        }
+        found.push(Backlink {
+            id: from,
+            name: row.get("name"),
+            slug: row.get("slug"),
+            path: path_text(&mut conn, from).await?,
+            context,
+        });
+    }
+    Ok(Json(found))
+}
+
+/// Index every document, when there is no link recorded at all: the ones
+/// written before links were kept. Returns how many documents link anywhere.
+pub async fn backfill(db: &SqlitePool) -> AppResult<usize> {
+    let any: bool = sqlx::query_scalar("SELECT EXISTS (SELECT 1 FROM links)")
+        .fetch_one(db)
+        .await?;
+    if any {
+        return Ok(0);
+    }
+    let documents = sqlx::query(&format!(
+        "WITH RECURSIVE {} SELECT id, body FROM nodes
+         WHERE kind = 'document' AND id IN (SELECT id FROM live)
+           AND (instr(body, '/n/') > 0 OR instr(body, '[[') > 0)",
+        nodes::LIVE
+    ))
+    .fetch_all(db)
+    .await?;
+
+    let mut tx = crate::db::begin_write(db).await?;
+    for row in &documents {
+        index(&mut tx, row.get("id"), row.get("body")).await?;
+    }
+    let linking: i64 = sqlx::query_scalar("SELECT COUNT(DISTINCT from_id) FROM links")
+        .fetch_one(&mut *tx)
+        .await?;
+    tx.commit().await?;
+    Ok(linking as usize)
+}
+
+#[cfg(test)]
+mod tests {
+    use std::collections::HashMap;
+
+    use axum::extract::{Path, State};
+
+    use super::{rewrite, written, Kind};
+    use crate::{routes::nodes, state::AppState};
+
+    #[test]
+    fn what_counts_as_a_link() {
+        let body = "[plan](/n/work/plan) and [notes](</n/work/a/notes#goals>)\n\
+                    [ref]: /n/uni/thesis\n\
+                    not a link: /n/work/prose, [cased](/n/work/Plan), [file](/n/work/a.md)\n\
+                    [[Plan]], [[ Other plan |the other]], [[]], [[a|b|c]] links to a";
+        let written = written(body);
+        let found: Vec<(Kind, &str)> = written
+            .iter()
+            .map(|(kind, target)| (*kind, target.as_str()))
+            .collect();
+        assert_eq!(
+            found,
+            [
+                (Kind::Path, "uni/thesis"),
+                (Kind::Path, "work/a/notes"),
+                (Kind::Path, "work/plan"),
+                (Kind::Wiki, "Other plan"),
+                (Kind::Wiki, "Plan"),
+                (Kind::Wiki, "a"),
+            ]
+        );
+    }
+
+    #[test]
+    fn a_rewrite_moves_only_the_link_it_names() {
+        let to = HashMap::from([
+            ((Kind::Path, "work/a".to_string()), "work/b".to_string()),
+            ((Kind::Wiki, "Plan".to_string()), "Roadmap".to_string()),
+        ]);
+        let body = "[a](/n/work/a) [deeper](/n/work/a/x) [longer](/n/work/ab) [[Plan]] [[Plan|the plan]] [[Plans]]";
+        assert_eq!(
+            rewrite(body, &to),
+            "[a](/n/work/b) [deeper](/n/work/a/x) [longer](/n/work/ab) [[Roadmap]] [[Roadmap|the plan]] [[Plans]]"
+        );
+    }
+
+    async fn space(state: &AppState, slug: &str) -> i64 {
+        sqlx::query_scalar("SELECT id FROM nodes WHERE slug = ?1 AND parent_id IS NULL")
+            .bind(slug)
+            .fetch_one(&state.db)
+            .await
+            .unwrap()
+    }
+
+    async fn create(state: &AppState, parent: i64, kind: &str, name: &str, body: &str) -> i64 {
+        let input = serde_json::from_value(serde_json::json!({
+            "parent_id": parent, "kind": kind, "name": name, "body": body,
+        }))
+        .unwrap();
+        nodes::create(session(), State(state.clone()), axum::Json(input))
+            .await
+            .unwrap()
+            .0
+            .id
+    }
+
+    async fn rename(state: &AppState, id: i64, name: &str) {
+        let body: String = sqlx::query_scalar("SELECT body FROM nodes WHERE id = ?1")
+            .bind(id)
+            .fetch_one(&state.db)
+            .await
+            .unwrap();
+        let input =
+            serde_json::from_value(serde_json::json!({ "name": name, "body": body })).unwrap();
+        let _renamed = nodes::update(session(), State(state.clone()), Path(id), axum::Json(input))
+            .await
+            .unwrap();
+    }
+
+    async fn body(state: &AppState, id: i64) -> String {
+        sqlx::query_scalar("SELECT body FROM nodes WHERE id = ?1")
+            .bind(id)
+            .fetch_one(&state.db)
+            .await
+            .unwrap()
+    }
+
+    async fn targets(state: &AppState, from: i64) -> Vec<(String, String, i64)> {
+        sqlx::query_as("SELECT kind, target, to_id FROM links WHERE from_id = ?1 ORDER BY kind")
+            .bind(from)
+            .fetch_all(&state.db)
+            .await
+            .unwrap()
+    }
+
+    fn session() -> crate::auth::Session {
+        crate::auth::Session { id: "test".into() }
+    }
+
+    /// A rename of the target rewrites the path and the title that link to
+    /// it; a rename of a folder above it rewrites the path.
+    #[tokio::test]
+    async fn a_rename_rewrites_what_links_to_it() {
+        let state = crate::routes::tests::state_in(".".into()).await;
+        let work = space(&state, "work").await;
+        let folder = create(&state, work, "folder", "Notes", "").await;
+        let plan = create(&state, folder, "document", "Plan", "the plan").await;
+        let linker = create(
+            &state,
+            work,
+            "document",
+            "Linker",
+            "see [it](/n/work/notes/plan#goals) and [[Plan]]",
+        )
+        .await;
+        assert_eq!(
+            targets(&state, linker).await,
+            [
+                ("path".to_string(), "work/notes/plan".to_string(), plan),
+                ("wiki".to_string(), "Plan".to_string(), plan),
+            ]
+        );
+
+        rename(&state, plan, "Roadmap").await;
+        assert_eq!(
+            body(&state, linker).await,
+            "see [it](/n/work/notes/roadmap#goals) and [[Roadmap]]"
+        );
+        rename(&state, folder, "Ideas").await;
+        assert_eq!(
+            body(&state, linker).await,
+            "see [it](/n/work/ideas/roadmap#goals) and [[Roadmap]]"
+        );
+        assert_eq!(targets(&state, linker).await.len(), 2);
+
+        // Each rewrite kept what it replaced, as any other write of the text.
+        let kept: i64 = sqlx::query_scalar("SELECT COUNT(*) FROM revisions WHERE node_id = ?1")
+            .bind(linker)
+            .fetch_one(&state.db)
+            .await
+            .unwrap();
+        assert_eq!(kept, 1, "the second rewrite came within ten minutes");
+    }
+
+    /// A move into another folder rewrites the path; the title still finds it.
+    #[tokio::test]
+    async fn a_move_rewrites_the_path() {
+        let state = crate::routes::tests::state_in(".".into()).await;
+        let work = space(&state, "work").await;
+        let elsewhere = create(&state, work, "folder", "Elsewhere", "").await;
+        let plan = create(&state, work, "document", "Plan", "").await;
+        let linker = create(
+            &state,
+            work,
+            "document",
+            "Linker",
+            "[p](/n/work/plan) [[plan]]",
+        )
+        .await;
+
+        nodes::move_nodes(&state, &[plan], elsewhere, None)
+            .await
+            .unwrap();
+        assert_eq!(
+            body(&state, linker).await,
+            "[p](/n/work/elsewhere/plan) [[plan]]"
+        );
+    }
+
+    /// A title is looked for in the linking document's space first.
+    #[tokio::test]
+    async fn a_title_resolves_in_its_own_space_first() {
+        let state = crate::routes::tests::state_in(".".into()).await;
+        let work = space(&state, "work").await;
+        let uni = space(&state, "uni").await;
+        let at_work = create(&state, work, "document", "Plan", "").await;
+        let at_uni = create(&state, uni, "document", "Plan", "").await;
+        let from_uni = create(&state, uni, "document", "a", "[[Plan]]").await;
+        let from_work = create(&state, work, "document", "b", "[[plan]]").await;
+        let only_uni = create(&state, work, "document", "c", "[[Thesis]]").await;
+        let thesis = create(&state, uni, "document", "Thesis", "").await;
+
+        assert_eq!(targets(&state, from_uni).await[0].2, at_uni);
+        assert_eq!(targets(&state, from_work).await[0].2, at_work);
+        // Written before the thesis was, and claimed by it when it came.
+        assert_eq!(targets(&state, only_uni).await[0].2, thesis);
+    }
+
+    /// The trash keeps the link; a restore elsewhere rewrites it, a purge
+    /// leaves the text as it was written.
+    #[tokio::test]
+    async fn the_trash_keeps_a_link_and_a_purge_drops_it() {
+        let state = crate::routes::tests::state_in(".".into()).await;
+        let work = space(&state, "work").await;
+        let folder = create(&state, work, "folder", "Notes", "").await;
+        let plan = create(&state, folder, "document", "Plan", "").await;
+        let linker = create(
+            &state,
+            work,
+            "document",
+            "Linker",
+            "[p](/n/work/notes/plan)",
+        )
+        .await;
+
+        crate::routes::trash::trash_subtree(&state.db, plan)
+            .await
+            .unwrap();
+        // Written again while the plan is in the trash: the link stays known.
+        rename(&state, linker, "Linker").await;
+        assert_eq!(targets(&state, linker).await.len(), 1);
+
+        nodes::purge_subtree(&state.db, std::path::Path::new("."), folder)
+            .await
+            .unwrap();
+        // The folder is gone for good, so the plan goes back to the space.
+        crate::routes::trash::restore_node(&state, plan)
+            .await
+            .unwrap();
+        assert_eq!(body(&state, linker).await, "[p](/n/work/plan)");
+
+        crate::routes::trash::trash_subtree(&state.db, plan)
+            .await
+            .unwrap();
+        crate::routes::trash::purge_older(&state.db, std::path::Path::new("."), i64::MAX)
+            .await
+            .unwrap();
+        assert!(targets(&state, linker).await.is_empty());
+        assert_eq!(body(&state, linker).await, "[p](/n/work/plan)");
+    }
+
+    /// The documents written before links were kept are indexed once, and
+    /// only while nothing is.
+    #[tokio::test]
+    async fn the_backfill_indexes_what_came_before() {
+        let state = crate::routes::tests::state_in(".".into()).await;
+        let work = space(&state, "work").await;
+        let plan = create(&state, work, "document", "Plan", "").await;
+        let linker = create(&state, work, "document", "Linker", "[[Plan]]").await;
+        sqlx::query("DELETE FROM links")
+            .execute(&state.db)
+            .await
+            .unwrap();
+
+        assert_eq!(super::backfill(&state.db).await.unwrap(), 1);
+        assert_eq!(targets(&state, linker).await[0].2, plan);
+        assert_eq!(super::backfill(&state.db).await.unwrap(), 0);
+    }
+}

src/routes/mod.rs

@@ -3,6 +3,7 @@ pub mod boards;
 pub mod export;
 pub mod import;
 pub mod inspect;
+pub mod links;
 pub mod manifest;
 pub mod media;
 pub mod nodes;
@@ -71,6 +72,7 @@ pub fn api_router(max_upload: usize) -> Router<AppState> {
         .route("/nodes/{id}", put(nodes::update))
         .route("/nodes/{id}", delete(nodes::remove))
         .route("/nodes/{id}/revisions", get(revisions::list))
+        .route("/nodes/{id}/backlinks", get(links::backlinks))
         .route("/nodes/{id}/revisions/{rev}", get(revisions::get))
         .route("/nodes/move", post(nodes::move_many))
         .route("/nodes/{id}/move", post(nodes::move_node))

src/routes/nodes.rs

@@ -1,4 +1,4 @@
-use std::collections::{BTreeSet, HashSet};
+use std::collections::{BTreeMap, BTreeSet, HashSet};
 
 use axum::{
     extract::{Path, Query, State},
@@ -10,7 +10,7 @@ use sqlx::{Row, SqliteConnection, SqlitePool};
 use crate::{
     auth::Session,
     error::{AppError, AppResult},
-    routes::{media, revisions, trash},
+    routes::{links, media, revisions, trash},
     state::AppState,
 };
 
@@ -104,6 +104,11 @@ pub struct Node {
     /// This node's ancestors, the space first and the node itself last — the
     /// breadcrumb, and everything needed to build its /n/... path.
     pub path: Vec<Crumb>,
+    /// How many documents link here.
+    pub backlinks: i64,
+    /// Where each `[[title]]` in the body leads, as an /n/... path. A title
+    /// that leads nowhere is not in it.
+    pub wikilinks: BTreeMap<String, String>,
 }
 
 #[derive(Serialize)]
@@ -595,21 +600,9 @@ pub async fn resolve(
     State(state): State<AppState>,
     Query(params): Query<ResolveQuery>,
 ) -> AppResult<Json<Node>> {
-    let mut current: Option<i64> = None;
+    // A path starts at a space, never at the trash beside them.
-    for segment in params.path.split('/').filter(|s| !s.is_empty()) {
+    let found = links::resolve_path(&mut *state.db.acquire().await?, &params.path).await?;
-        // A path starts at a space, never at the trash beside them.
+    load(&state, found.ok_or(AppError::NotFound)?)
-        let row = sqlx::query(
-            "SELECT id FROM nodes
-             WHERE parent_id IS ?1 AND slug = ?2 AND (?1 IS NOT NULL OR kind = 'space')",
-        )
-        .bind(current)
-        .bind(segment)
-        .fetch_optional(&state.db)
-        .await?
-        .ok_or(AppError::NotFound)?;
-        current = Some(row.get("id"));
-    }
-    load(&state, current.ok_or(AppError::NotFound)?)
         .await
         .map(Json)
 }
@@ -650,6 +643,9 @@ pub async fn create(
     .get("id");
 
     media::link_to_node(&mut tx, id, &input.body).await?;
+    links::index(&mut tx, id, &input.body).await?;
+    // A `[[link]]` written before the document it names now finds it.
+    links::follow(&mut tx, id).await?;
     tx.commit().await?;
     state.backup.signal();
     load(&state, id).await.map(Json)
@@ -743,6 +739,12 @@ pub async fn update(
 
     if kind != FILE {
         media::link_to_node(&mut tx, id, body).await?;
+        links::index(&mut tx, id, body).await?;
+    }
+    // A new name is a new path, for this and everything under it, and a new
+    // title: the links written with the old ones are rewritten to find it.
+    if name != previous_name {
+        links::follow(&mut tx, id).await?;
     }
     tx.commit().await?;
     state.backup.signal();
@@ -878,11 +880,13 @@ pub async fn move_nodes(
         moved: 0,
         unchanged: 0,
     };
+    let mut moved = Vec::with_capacity(nodes.len());
     for (id, name, from) in nodes {
         if from == Some(parent) && position.is_none() {
             done.unchanged += 1;
             continue;
         }
+        moved.push(id);
         // The slug has to be free where it lands, not where it came from —
         // and it sees the ones moved in before it, so two `notes` stay apart.
         let slug = unique_slug(&mut tx, Some(parent), &slugify(&name), Some(id)).await?;
@@ -900,6 +904,11 @@ pub async fn move_nodes(
         .await?;
         done.moved += 1;
     }
+    // Once all of them are where they go, so a link between two of them is
+    // rewritten to where both now are.
+    for id in moved {
+        links::follow(&mut tx, id).await?;
+    }
     tx.commit().await?;
 
     if done.moved > 0 {
@@ -1023,6 +1032,13 @@ pub async fn load(state: &AppState, id: i64) -> AppResult<Node> {
 
     let body: String = row.get("body");
     let summary = row_to_summary(&row, &body.chars().take(HEAD_CHARS).collect::<String>());
+    let (backlinks, wikilinks) = {
+        let mut conn = state.db.acquire().await?;
+        (
+            links::linked_from(&mut conn, id).await?,
+            links::wikilinks(&mut conn, id, &body).await?,
+        )
+    };
 
     Ok(Node {
         id: summary.id,
@@ -1043,6 +1059,8 @@ pub async fn load(state: &AppState, id: i64) -> AppResult<Node> {
         sealed: summary.sealed,
         file: summary.file,
         path: path_of(&state.db, id).await?,
+        backlinks,
+        wikilinks,
     })
 }
 

src/routes/trash.rs

@@ -248,6 +248,8 @@ pub async fn restore_node(state: &AppState, id: i64) -> AppResult<bool> {
             .bind(id)
             .execute(&mut *tx)
             .await?;
+        // Under another slug when its own was taken meanwhile.
+        crate::routes::links::follow(&mut tx, id).await?;
         tx.commit().await?;
         return Ok(false);
     }

src/sftp/fs.rs

@@ -35,7 +35,7 @@ use crate::{
         Backup,
     },
     error::{AppError, AppResult},
-    routes::{media, nodes, revisions, seal, trash},
+    routes::{links, media, nodes, revisions, seal, trash},
     state::AppState,
     vault,
 };
@@ -778,6 +778,14 @@ impl SpaceFs {
         media::link_to_node(&mut tx, id, &body)
             .await
             .map_err(|_| StatusCode::Failure)?;
+        links::index(&mut tx, id, &body)
+            .await
+            .map_err(|_| StatusCode::Failure)?;
+        // A heading may have renamed it, and a `[[link]]` to the old title
+        // then follows it; a new document is what some link was waiting for.
+        links::follow(&mut tx, id)
+            .await
+            .map_err(|_| StatusCode::Failure)?;
         tx.commit().await.map_err(|_| StatusCode::Failure)?;
         self.changed();
         Ok(())
@@ -1097,6 +1105,9 @@ impl SpaceFs {
             .execute(&mut *tx)
             .await
             .map_err(|_| StatusCode::Failure)?;
+        links::follow(&mut tx, file.id)
+            .await
+            .map_err(|_| StatusCode::Failure)?;
         tx.commit().await.map_err(|_| StatusCode::Failure)?;
         self.changed();
         Ok(())
@@ -1715,6 +1726,9 @@ impl russh_sftp::server::Handler for SpaceFs {
                 .map_err(|_| StatusCode::Failure)?;
                 // A move that keeps the name keeps the title too.
                 let name = (new_leaf != slug).then_some(new_leaf);
+                let mut tx = crate::db::begin_write(&self.db)
+                    .await
+                    .map_err(|_| StatusCode::Failure)?;
                 sqlx::query(
                     "UPDATE nodes SET parent_id = ?1, name = COALESCE(?2, name), slug = ?3,
                      updated_at = ?4 WHERE id = ?5",
@@ -1724,9 +1738,13 @@ impl russh_sftp::server::Handler for SpaceFs {
                 .bind(&new_slug)
                 .bind(now)
                 .bind(node)
-                .execute(&self.db)
+                .execute(&mut *tx)
                 .await
                 .map_err(|_| StatusCode::Failure)?;
+                links::follow(&mut tx, node)
+                    .await
+                    .map_err(|_| StatusCode::Failure)?;
+                tx.commit().await.map_err(|_| StatusCode::Failure)?;
                 self.changed();
             }
 
@@ -1754,6 +1772,9 @@ impl russh_sftp::server::Handler for SpaceFs {
                         )
                         .await
                         .map_err(|_| StatusCode::Failure)?;
+                        let mut tx = crate::db::begin_write(&self.db)
+                            .await
+                            .map_err(|_| StatusCode::Failure)?;
                         sqlx::query(
                             "UPDATE nodes SET parent_id = ?1, slug = ?2, updated_at = ?3 WHERE id = ?4",
                         )
@@ -1761,9 +1782,13 @@ impl russh_sftp::server::Handler for SpaceFs {
                         .bind(&new_slug)
                         .bind(now)
                         .bind(doc.id)
-                        .execute(&self.db)
+                        .execute(&mut *tx)
                         .await
                         .map_err(|_| StatusCode::Failure)?;
+                        links::follow(&mut tx, doc.id)
+                            .await
+                            .map_err(|_| StatusCode::Failure)?;
+                        tx.commit().await.map_err(|_| StatusCode::Failure)?;
                         self.changed();
                     }
                     // Onto a name that is not a document: an editor moving the
@@ -2503,6 +2528,36 @@ mod tests {
         assert_eq!(kept, [("v1\n".to_string(), "sftp".to_string())]);
     }
 
+    /// A move over the mount, and a heading that renames, rewrite the links
+    /// to the document as a move and a rename in the browser do.
+    #[tokio::test]
+    async fn links_follow_a_document_moved_over_the_mount() {
+        let mut fs = fs().await;
+        fs.mkdir(0, "/proj".into(), FileAttributes::empty())
+            .await
+            .expect("mkdir");
+        fs.mkdir(0, "/proj/old".into(), FileAttributes::empty())
+            .await
+            .expect("mkdir");
+        put(&mut fs, "/proj/plan.md", "# Plan\n\nv1\n").await;
+        put(
+            &mut fs,
+            "/proj/notes.md",
+            "# Notes\n\nsee [it](/n/proj/plan) and [[Plan]]\n",
+        )
+        .await;
+
+        fs.rename(0, "/proj/plan.md".into(), "/proj/old/roadmap.md".into())
+            .await
+            .expect("rename");
+        let (.., body) = node(&fs, "notes").await;
+        assert_eq!(body, "see [it](/n/proj/old/roadmap) and [[Plan]]\n");
+
+        put(&mut fs, "/proj/old/roadmap.md", "# Roadmap\n\nv2\n").await;
+        let (.., body) = node(&fs, "notes").await;
+        assert_eq!(body, "see [it](/n/proj/old/roadmap) and [[Roadmap]]\n");
+    }
+
     #[tokio::test]
     async fn spaces_rename_at_the_root_and_folders_cannot_swallow_themselves() {
         let mut fs = fs().await;

web/bun.lock

@@ -5,6 +5,7 @@
     "": {
       "name": "narl-diary-web",
       "dependencies": {
+        "@codemirror/autocomplete": "^6.20.3",
         "@codemirror/commands": "^6.7.1",
         "@codemirror/lang-markdown": "^6.3.1",
         "@codemirror/language": "^6.10.8",

web/package.json

@@ -17,6 +17,7 @@
     "vite": "^6.0.11"
   },
   "dependencies": {
+    "@codemirror/autocomplete": "^6.20.3",
     "@codemirror/commands": "^6.7.1",
     "@codemirror/lang-markdown": "^6.3.1",
     "@codemirror/language": "^6.10.8",

web/src/App.svelte

@@ -15,6 +15,7 @@
   import Overview from './lib/Overview.svelte'
   import MediaBrowser, { copyMedia, insertMedia, removeMedia } from './lib/MediaBrowser.svelte'
   import Trash, { purgeTrashed, restoreTrashed } from './lib/Trash.svelte'
+  import Backlinks, { openBacklink } from './lib/Backlinks.svelte'
   import History, { moveHistory, restoreShown } from './lib/History.svelte'
   import Actions from './lib/Actions.svelte'
   import Prompt from './lib/Prompt.svelte'
@@ -422,6 +423,31 @@
     }
   }
 
+  /** The backlinks' keys: j/k through the documents that link here, Enter opens one. */
+  function backlinksKey(event: KeyboardEvent): boolean {
+    const last = Math.max(workspace.backlinks.length - 1, 0)
+    switch (event.key) {
+      case 'j':
+      case 'ArrowDown':
+        workspace.backlinksCursor = Math.min(workspace.backlinksCursor + 1, last)
+        return true
+      case 'k':
+      case 'ArrowUp':
+        workspace.backlinksCursor = Math.max(workspace.backlinksCursor - 1, 0)
+        return true
+      case 'G':
+        workspace.backlinksCursor = last
+        return true
+      case 'Enter': {
+        const link = workspace.backlinks[workspace.backlinksCursor]
+        if (link) void openBacklink(link)
+        return true
+      }
+      default:
+        return false
+    }
+  }
+
   /** The history's keys: j/k through the versions, Enter brings the one shown back. */
   function historyKey(event: KeyboardEvent): boolean {
     switch (event.key) {
@@ -637,6 +663,12 @@
         return
       }
     }
+    if (workspace.overlay === 'backlinks' && listening && !event.ctrlKey && !event.metaKey && !event.altKey) {
+      if (backlinksKey(event)) {
+        event.preventDefault()
+        return
+      }
+    }
     if (workspace.overlay === 'trash' && listening && !event.ctrlKey && !event.metaKey && !event.altKey) {
       if (trashKey(event)) {
         event.preventDefault()
@@ -793,6 +825,11 @@
       void runCommand('history')
       return
     }
+    if (previous === 'g' && key === 'b') {
+      event.preventDefault()
+      void runCommand('backlinks')
+      return
+    }
     if (previous === 'g' && key === 'x') {
       event.preventDefault()
       openInTab()
@@ -1075,7 +1112,7 @@
       <!-- svelte-ignore a11y_click_events_have_key_events, a11y_no_static_element_interactions, a11y_no_noninteractive_tabindex -->
       <div class="overlay" tabindex="-1" bind:this={overlayEl} onclick={(e) => e.target === e.currentTarget && (workspace.overlay = 'none')}
       >
-        {#if workspace.overlay === 'help'}<Help />{:else if workspace.overlay === 'overview'}<Overview />{:else if workspace.overlay === 'agenda'}<Agenda />{:else if workspace.overlay === 'actions'}<Actions />{:else if workspace.overlay === 'verify'}<Verify />{:else if workspace.overlay === 'outline'}<Headings />{:else if workspace.overlay === 'trash'}<Trash />{:else if workspace.overlay === 'history'}<History />{:else}<MediaBrowser />{/if}
+        {#if workspace.overlay === 'help'}<Help />{:else if workspace.overlay === 'overview'}<Overview />{:else if workspace.overlay === 'agenda'}<Agenda />{:else if workspace.overlay === 'actions'}<Actions />{:else if workspace.overlay === 'verify'}<Verify />{:else if workspace.overlay === 'outline'}<Headings />{:else if workspace.overlay === 'trash'}<Trash />{:else if workspace.overlay === 'history'}<History />{:else if workspace.overlay === 'backlinks'}<Backlinks />{:else}<MediaBrowser />{/if}
       </div>
     {/if}
   </div>

web/src/app.css

@@ -93,6 +93,8 @@ a {
 }
 a:hover { color: var(--light); text-decoration-color: var(--light); }
 a:focus-visible { background: var(--light); color: var(--ink); text-decoration: none; outline: none; }
+/* A `[[title]]` no document has yet: there, but not written. */
+a.wikilink.missing { color: var(--muted); text-decoration-style: dashed; }
 
 /*
  * A button rests as a hairline and a label. Hover lifts it a step and lights

web/src/lib/Backlinks.svelte

@@ -0,0 +1,101 @@
+<script lang="ts" module>
+  import type { Backlink } from './api'
+  import { workspace } from './store.svelte'
+
+  /** What Enter and a click do: the linking document is opened. */
+  export function openBacklink(link: Backlink) {
+    workspace.overlay = 'none'
+    return workspace.guard(() => workspace.openPath(link.path))
+  }
+</script>
+
+<script lang="ts">
+  let sheet = $state<HTMLDivElement>()
+
+  $effect(() => {
+    const index = workspace.backlinksCursor
+    sheet?.querySelector<HTMLElement>(`[data-index="${index}"]`)?.scrollIntoView({ block: 'nearest' })
+  })
+</script>
+
+<div class="sheet" bind:this={sheet}>
+  <header>
+    <span class="label rule">
+      <span class="marker"></span><span class="title">backlinks</span>
+      {workspace.open?.name.trim() || workspace.open?.slug} · linked from {workspace.backlinks.length}
+    </span>
+    <button onclick={() => (workspace.overlay = 'none')}>esc</button>
+  </header>
+
+  <ul>
+    {#each workspace.backlinks as link, index (link.id)}
+      <!-- svelte-ignore a11y_click_events_have_key_events, a11y_no_noninteractive_element_interactions -->
+      <li
+        class:at={index === workspace.backlinksCursor}
+        data-index={index}
+        onpointerenter={() => (workspace.backlinksCursor = index)}
+        onclick={() => openBacklink(link)}
+      >
+        <span class="name">{link.name.trim() || link.slug}</span>
+        <span class="muted">{link.path}</span>
+        {#if link.context}<span class="context">{link.context}</span>{/if}
+      </li>
+    {:else}
+      <li class="muted">nothing links here yet — a link is [text](/n/…) or [[title]].</li>
+    {/each}
+  </ul>
+
+  <footer class="label">
+    <span class="kbd">j</span><span class="kbd">k</span> move
+    <span class="gap"></span>
+    <span class="kbd">enter</span> open
+    <span class="gap"></span>
+    <span class="kbd">q</span><span class="kbd">esc</span> close
+  </footer>
+</div>
+
+<style>
+  /* A free-standing panel, as the desktop draws one: hull, one rim, the frame's radius. */
+  .sheet {
+    display: flex;
+    flex-direction: column;
+    gap: 18px;
+    max-height: 100%;
+    background: color-mix(in oklab, var(--hull) 97%, transparent);
+    border: 1px solid var(--surface-alt);
+    border-radius: var(--frame-radius);
+    padding: 20px 24px;
+    animation: resolve var(--t-slow) var(--ease-snap) both;
+    width: min(100%, 92ch);
+    overflow: hidden;
+  }
+
+  header { display: flex; align-items: center; gap: 12px; }
+  header .label { flex: 1; }
+  header .marker { background: var(--light); }
+  header .title { color: var(--light); }
+
+  ul { list-style: none; margin: 0; padding: 0; overflow-y: auto; display: flex; flex-direction: column; gap: 2px; }
+
+  li {
+    display: flex;
+    flex-direction: column;
+    padding: 6px 8px;
+    border-radius: var(--radius-m);
+    cursor: pointer;
+    transition: background-color var(--t-base) var(--ease-glide);
+  }
+  li:hover { background: var(--surface); }
+  li.at { background: var(--surface-alt); box-shadow: inset 2px 0 0 var(--light); }
+
+  .name { overflow: hidden; text-overflow: ellipsis; white-space: nowrap; color: var(--fg); font-family: var(--font-sans); font-size: 14px; }
+  .muted { font-size: 11px; }
+  .context { font-size: 12px; color: var(--subtle); overflow: hidden; text-overflow: ellipsis; white-space: nowrap; }
+
+  footer { display: flex; align-items: center; gap: 4px; font-size: 10px; }
+  footer .kbd { margin-right: 2px; }
+  footer .gap { width: 14px; }
+  @media (max-width: 767px), (pointer: coarse) {
+    footer { display: none; }
+  }
+</style>

web/src/lib/DocumentPane.svelte

@@ -21,7 +21,7 @@
 
   const focused = $derived(workspace.tile === tile)
   const doc = $derived(tile.open!)
-  const html = $derived(renderMarkdown(tile.editing ? tile.draft.body : doc.body))
+  const html = $derived(renderMarkdown(tile.editing ? tile.draft.body : doc.body, { wikilinks: doc.wikilinks }))
   const words = $derived(
     (tile.editing ? tile.draft.body : doc.body).split(/\s+/).filter(Boolean).length,
   )
@@ -161,6 +161,10 @@
         <dd>{relative(doc.updated_at)}</dd>
         <dt>words</dt>
         <dd>{words}<span class="unit">words</span></dd>
+        {#if doc.backlinks > 0}
+          <dt>linked</dt>
+          <dd><button class="linked" title="what links here (gb)" onclick={() => runCommand('backlinks')}>from {doc.backlinks}</button></dd>
+        {/if}
       </dl>
 
       <div class="actions">
@@ -298,6 +302,9 @@
   }
   .meta dt { color: var(--muted); letter-spacing: 0.18em; text-transform: uppercase; }
   .meta dd { margin: 0 10px 0 0; color: var(--fg); }
+  /* Read as the rest of the line, and still a way to the backlinks. */
+  .meta .linked { all: unset; cursor: pointer; text-decoration: underline; text-decoration-color: var(--overlay); text-underline-offset: 4px; }
+  .meta .linked:hover, .meta .linked:focus-visible { color: var(--light); }
 
   .actions { display: flex; gap: 4px; flex-wrap: wrap; flex: none; }
   .actions .on { color: var(--light); border-color: var(--light-dim); }

web/src/lib/Editor.svelte

@@ -133,6 +133,7 @@
 </script>
 
 <script lang="ts">
+  import { autocompletion, type CompletionContext, type CompletionResult } from '@codemirror/autocomplete'
   import { defaultKeymap, history, historyKeymap, indentWithTab } from '@codemirror/commands'
   import { markdown, markdownLanguage } from '@codemirror/lang-markdown'
   import { HighlightStyle, indentOnInput, syntaxHighlighting } from '@codemirror/language'
@@ -146,6 +147,7 @@
   } from '@codemirror/view'
   import { tags } from '@lezer/highlight'
   import { onDestroy, onMount } from 'svelte'
+  import { api } from './api'
   import { hooks } from './commands'
   import { embedSnippet } from './markdown'
   import { workspace, type Tile } from './store.svelte'
@@ -200,6 +202,16 @@
         display: 'flex',
         alignItems: 'center',
       },
+      '.cm-tooltip': {
+        backgroundColor: 'var(--hull)',
+        color: 'var(--fg)',
+        border: '1px solid var(--rim)',
+      },
+      '.cm-tooltip-autocomplete > ul > li[aria-selected]': {
+        backgroundColor: 'var(--surface-alt)',
+        color: 'var(--light)',
+      },
+      '.cm-completionDetail': { color: 'var(--muted)', fontStyle: 'normal', marginLeft: '1ch' },
       '.cm-vim-panel input': {
         fontFamily: 'inherit',
         fontSize: 'inherit',
@@ -212,6 +224,29 @@
     { dark: true },
   )
 
+  /*
+   * `[[` offers the titles of documents as the search finds them, each with
+   * where it sits; picking one closes the brackets.
+   */
+  async function titleCompletions(context: CompletionContext): Promise<CompletionResult | null> {
+    const typed = context.matchBefore(/\[\[[^[\]|\n]*/)
+    if (!typed) return null
+    const query = typed.text.slice(2).trim()
+    if (!query) return null
+    const found = await api.listNodes({ q: query }).catch(() => [])
+    if (context.aborted) return null
+    const options = found
+      .filter((node) => node.kind === 'document' || node.kind === 'file')
+      .filter((node) => node.id !== tile.open?.id)
+      .map((node) => {
+        // What a `[[link]]` can hold: a name with no bracket or bar in it, or else the slug.
+        const name = node.name.trim()
+        const title = name && !/[[\]|\n]/.test(name) ? name : node.slug
+        return { label: title, detail: node.location, apply: `${title}]]` }
+      })
+    return { from: typed.from + 2, options }
+  }
+
   /** The `#` lines being written, as `gO` lists them, and the one the cursor is under. */
   function editorHeadings() {
     if (!view) return { headings: [], current: 0 }
@@ -264,6 +299,7 @@
       highlightActiveLine(),
       EditorState.allowMultipleSelections.of(true),
       markdown({ base: markdownLanguage, codeLanguages: [] }),
+      autocompletion({ override: [titleCompletions], icons: false }),
       syntaxHighlighting(markdownHighlight),
       EditorView.lineWrapping,
       placeholder('write…'),

web/src/lib/Help.svelte

@@ -44,6 +44,8 @@
     ['x · dd', 'delete what has focus — the row in the list, the document in its pane — into the trash (asks y/N)'],
     ['u', 'undo the last delete: it comes back out of the trash — :trash has the rest'],
     ['gh', "the open document's earlier versions, as diffs — Enter brings one back"],
+
+    ['gb', 'the documents that link to the open one — [[title]] links by title, [[ completes one while writing'],
     [':', 'command line — Tab / Shift-Tab cycle the matches, then the argument; ↑ ↓ recall what ran before'],
     ['Ctrl-S', 'save, from anywhere'],
     ['?', 'this screen — in the editor too'],

web/src/lib/Shared.svelte

@@ -19,7 +19,7 @@
   let article = $state<HTMLElement>()
   let position = $state('All')
   // Links into the private workspace come back as their text; see renderMarkdown.
-  const html = $derived(doc ? renderMarkdown(doc.body, { token }) : '')
+  const html = $derived(doc ? renderMarkdown(doc.body, { share: { token } }) : '')
   const title = $derived(doc?.name.trim() ?? '')
   const words = $derived(doc ? doc.body.split(/\s+/).filter(Boolean).length : 0)
   // The outline gets a window of its own only when there is something to outline.

web/src/lib/api.ts

@@ -49,6 +49,19 @@ export interface Crumb {
 export interface WorkspaceNode extends NodeSummary {
   body: string
   path: Crumb[]
+  /** How many documents link here. */
+  backlinks: number
+  /** Where each `[[title]]` in the body leads, as an /n/ path; a title that leads nowhere is missing. */
+  wikilinks: Record<string, string>
+}
+
+/** A document that links to the one asked about, and the line it does it on. */
+export interface Backlink {
+  id: number
+  name: string
+  slug: string
+  path: string
+  context: string
 }
 
 /** A space as the overview shows it: itself, and what was last written in it. */
@@ -429,6 +442,9 @@ export const api = {
 
   revision: (id: number, rev: number) => request<Revision>(`/nodes/${id}/revisions/${rev}`),
 
+  /** The documents that link to this one, the last written first. */
+  backlinks: (id: number) => request<Backlink[]>(`/nodes/${id}/backlinks`),
+
   /** Into the trash; `purged` when it was in the trash already and is now gone. */
   deleteNode: (id: number) =>
     request<{ ok: true; removed: number; purged: boolean }>(`/nodes/${id}`, { method: 'DELETE' }),

web/src/lib/commands.ts

@@ -2,7 +2,7 @@ import { tick } from 'svelte'
 import { api, ApiError, isContainer, type BackupLogin, type BackupStatus } from './api'
 import { workspace } from './store.svelte'
 import { emptyTrash } from './Trash.svelte'
-import { formatDay, parseDay, parseDue, relative, sealedNote, type Heading, type ScrollAmount, type SidewaysAmount } from './util'
+import { formatDay, parseDay, parseDue, relative, sealedNote, titleHref, type Heading, type ScrollAmount, type SidewaysAmount } from './util'
 
 export type { ScrollAmount, SidewaysAmount }
 
@@ -76,13 +76,24 @@ const requireOpen = () => {
   return workspace.open
 }
 
+const TITLE_HREF = titleHref('')
+
 /**
  * Where a link in a document goes. `/n/` links are the workspace's own — the
- * importer writes them between documents — and are opened in place; anything
+ * importer writes them between documents — and are opened in place, as is a
+ * `[[title]]`, or the offer to write it when no document has it; anything
  * else would take the workspace's tab with it, so it gets a tab of its own,
  * as does a workspace link asked for with `gx`.
  */
 export function followLink(href: string, tab = false) {
+  if (href.startsWith(TITLE_HREF)) {
+    const title = decodeURIComponent(href.slice(TITLE_HREF.length))
+    const path = workspace.open?.wikilinks[title]
+    if (tab) {
+      if (!path) return workspace.say(`no document called "${title}" yet`, 'error')
+      href = `/n/${path}`
+    } else return void workspace.guard(() => workspace.followTitle(title))
+  }
   if (href.startsWith('/n/') && !tab) {
     const path = href.slice(3).split('#')[0]
     return void workspace.guard(() => workspace.openPath(decodeURIComponent(path)))
@@ -789,6 +800,19 @@ export const commands: CommandSpec[] = [
       })
     },
   },
+  {
+    name: 'backlinks',
+    aliases: ['bl'],
+    group: 'documents',
+    help: 'the documents that link to the open one — j/k, Enter opens one (gb)',
+    run: async () => {
+      if (!requireOpen()) return
+      await workspace.guard(async () => {
+        await workspace.loadBacklinks()
+        workspace.overlay = 'backlinks'
+      })
+    },
+  },
   {
     name: 'trash',
     args: '[purge]',

web/src/lib/markdown.ts

@@ -1,6 +1,7 @@
 import DOMPurify from 'dompurify'
 import { marked, type TokenizerAndRendererExtension } from 'marked'
 import { mathEngine } from './math.svelte'
+import { titleHref } from './util'
 
 marked.setOptions({ gfm: true, breaks: true })
 
@@ -37,8 +38,38 @@ const mathInline: TokenizerAndRendererExtension = {
   renderer: (token) => tex(token.text, token.display),
 }
 
+/*
+ * `[[Plan]]` and `[[Plan|the plan]]`: a link to a document by its title. The
+ * server says where each title leads, since it is the one that knows the
+ * titles; one that leads nowhere is drawn dimmed, and following it offers to
+ * write the document.
+ */
+const WIKILINK = /^\[\[([^[\]|\n]+)(?:\|([^[\]\n]*))?\]\]/
+
+/** What the page being rendered says about its `[[titles]]`, for the renderer. */
+let wikiContext: { leads: Record<string, string>; sharing: boolean } = { leads: {}, sharing: false }
+
+const wikilink: TokenizerAndRendererExtension = {
+  name: 'wikilink',
+  level: 'inline',
+  start: (src) => src.indexOf('[['),
+  tokenizer(src) {
+    const match = WIKILINK.exec(src)
+    const target = match?.[1].trim()
+    if (match && target) return { type: 'wikilink', raw: match[0], target, label: match[2]?.trim() || target }
+  },
+  renderer: (token) => {
+    const label = escapeHtml(token.label)
+    // The page it leads to is behind the login, as any /n/ link on a share is.
+    if (wikiContext.sharing) return `<span class="private-link" title="a page of the private workspace">${label}</span>`
+    const path = wikiContext.leads[token.target]
+    if (path) return `<a href="/n/${encodeURI(path)}" class="wikilink">${label}</a>`
+    return `<a href="${titleHref(token.target)}" class="wikilink missing" title="not written yet — follow it to write it">${label}</a>`
+  },
+}
+
 marked.use({
-  extensions: [mathBlock, mathInline],
+  extensions: [mathBlock, mathInline, wikilink],
   renderer: {
     code(token) {
       return token.lang === 'math' ? tex(token.text, true) : false
@@ -84,13 +115,18 @@ function escapeHtml(text: string): string {
  * the cookie reading the document with its key handed out, so the key itself
  * never sits in a URL a proxy could log. A link to another document, an
  * address inside the private workspace that would lead the reader to its
- * login page, is kept as its text.
+ * login page, is kept as its text. `wikilinks` says where the `[[titles]]`
+ * lead, as the server answered for the document.
  */
-export function renderMarkdown(source: string, share?: { token: string }): string {
+export function renderMarkdown(
+  source: string,
+  { share, wikilinks = {} }: { share?: { token: string }; wikilinks?: Record<string, string> } = {},
+): string {
   const md = share
     ? source.replaceAll('/api/media/', `/api/share/${encodeURIComponent(share.token)}/media/`)
     : source
 
+  wikiContext = { leads: wikilinks, sharing: !!share }
   const html = marked.parse(md, { async: false }) as string
   if (!share) return DOMPurify.sanitize(html, SANITIZE)
 

web/src/lib/store.svelte.ts

@@ -5,6 +5,7 @@ import {
   unreachable,
   type Agenda,
   type AgendaCard,
+  type Backlink,
   type Board,
   type Card,
   type Crumb,
@@ -281,7 +282,7 @@ class Workspace {
   pane = $state<Pane>('list')
   flash = $state<Flash>(null)
 
-  overlay = $state<'none' | 'help' | 'media' | 'overview' | 'agenda' | 'actions' | 'verify' | 'outline' | 'trash' | 'history'>('none')
+  overlay = $state<'none' | 'help' | 'media' | 'overview' | 'agenda' | 'actions' | 'verify' | 'outline' | 'trash' | 'history' | 'backlinks'>('none')
   /** `gO`: the headings of what is on screen, to jump between from the keyboard. */
   outline = $state.raw<Heading[]>([])
   outlineCursor = $state(0)
@@ -317,6 +318,9 @@ class Workspace {
   history = $state<RevisionSummary[]>([])
   historyCursor = $state(0)
   revision = $state.raw<Revision | null>(null)
+  /** `:backlinks`: the documents that link to the open one. */
+  backlinks = $state<Backlink[]>([])
+  backlinksCursor = $state(0)
   overview = $state<SpaceOverview[]>([])
   overviewCursor = $state(0)
 
@@ -1096,6 +1100,29 @@ class Workspace {
     this.say(`restored the version from ${formatStamp(revision.saved_at)} — :history has the one it replaced`)
   }
 
+  /**
+   * Follow a `[[title]]`: to the document it names, or — when there is none
+   * — to a new one called that, here, after asking. Which document a title
+   * names the server says when the text is written, so it is written first.
+   */
+  async followTitle(title: string) {
+    await this.flushTile()
+    const path = this.open?.wikilinks[title]
+    if (path) return this.openPath(path)
+    if (!(await this.ask(`no document called "${title}" yet — write it in ${this.here}?`))) return
+    await this.createDocument(title)
+  }
+
+  /** `:backlinks`: what links to the open document, the last written first. */
+  async loadBacklinks() {
+    const open = this.open
+    if (!open) return
+    // A link typed and not yet written is not one the server knows about.
+    await this.flush()
+    this.backlinks = await api.backlinks(open.id)
+    this.backlinksCursor = 0
+  }
+
   async loadTrash() {
     this.trash = await api.trash()
     this.trashCursor = Math.min(this.trashCursor, Math.max(this.trash.length - 1, 0))
@@ -1135,10 +1162,28 @@ class Workspace {
     }
     if (this.path.some((crumb) => crumb.id === id)) this.path = saved.path
     if (saved.kind === 'space') this.spaces = await api.listSpaces()
+    await this.reread()
     await this.refresh()
     this.say(`renamed to ${saved.slug}`)
   }
 
+  /**
+   * Read the open documents again after a rename or a move: their paths may
+   * have changed with it, and their text too, where the server rewrote a
+   * link to what moved. One being written is left as it is — saving it is
+   * then refused, and `:e!` takes the rewritten text.
+   */
+  async reread() {
+    for (const tile of this.tiles) {
+      if (!tile.open || tile.editing || tile.dirty) continue
+      const node = await api.getNode(tile.open.id).catch(() => null)
+      if (!node) continue
+      tile.open = node
+      tile.draft = { name: node.name, body: node.body, created_at: node.created_at }
+    }
+    if (this.open) this.#syncUrl(this.open.path)
+  }
+
   /**
    * Delete a space and everything in it. The app has to land somewhere
    * afterwards, so it falls into the next space — or, if that was the last
@@ -1168,6 +1213,7 @@ class Workspace {
   /** Reparent the node under the cursor, for `:mv`. */
   async moveNode(id: number, parentId: number) {
     await api.moveNode(id, parentId)
+    await this.reread()
     await this.refresh()
     this.say(`moved #${id}`)
   }
@@ -1194,6 +1240,7 @@ class Workspace {
     )
     const gone = new Set(nodes.map((n) => n.id))
     this.marked = this.marked.filter((n) => !gone.has(n.id))
+    if (moved > 0) await this.reread()
     await this.refresh(nodes.length === 1 ? nodes[0].id : undefined)
     const what = nodes.length === 1 ? nodes[0].name.trim() || nodes[0].slug : `${moved}`
     const already = unchanged > 0 ? ` — ${unchanged} already ${unchanged === 1 ? 'was' : 'were'}` : ''

web/src/lib/util.ts

@@ -234,16 +234,19 @@ export function sourceHeadings(source: string): { text: string; level: number; l
 
 /**
  * The link the cursor stands on in a line of markdown — `[text](href)`, an
- * embed's `![alt](src)`, `<https://…>` or a bare address — or else the first
+ * embed's `![alt](src)`, `[[title]]`, `<https://…>` or a bare address — or else the first
  * one after it on the line, as vim's `gf` looks ahead to the next name.
  */
 export function linkAt(line: string, column: number): string | null {
   const pattern =
-    /!?\[[^\]]*\]\(\s*<?([^)\s>]+)>?(?:\s+"[^"]*")?\s*\)|<((?:https?:\/\/|\/n\/)[^>\s]+)>|((?:https?:\/\/|\/n\/)[^\s<>()]+)/g
+    /\[\[([^[\]|\n]+)(?:\|[^[\]\n]*)?\]\]|!?\[[^\]]*\]\(\s*<?([^)\s>]+)>?(?:\s+"[^"]*")?\s*\)|<((?:https?:\/\/|\/n\/)[^>\s]+)>|((?:https?:\/\/|\/n\/)[^\s<>()]+)/g
   let ahead: string | null = null
   for (const match of line.matchAll(pattern)) {
-    // A sentence that ends on an address ends with a full stop, not the address.
+    // A sentence that ends on a bare address ends with a full stop, not the address.
-    const href = match[1] ?? match[2] ?? match[3].replace(/[.,;:!?]+$/, '')
+    const href = match[1]?.trim()
+      ? titleHref(match[1].trim())
+      : (match[2] ?? match[3] ?? match[4]?.replace(/[.,;:!?]+$/, ''))
+    if (!href) continue
     const start = match.index
     if (column >= start && column < start + match[0].length) return href
     if (start >= column) ahead ??= href
@@ -251,6 +254,12 @@ export function linkAt(line: string, column: number): string | null {
   return ahead
 }
 
+/**
+ * Where a `[[title]]` points when the page cannot say where it leads: the
+ * title itself, for following it to look it up, or to offer to write it.
+ */
+export const titleHref = (title: string) => `/n/?title=${encodeURIComponent(title)}`
+
 /** What a share link's reader will miss: a share never serves a sealed file. */
 export const sealedNote = (sealed: number) =>
   sealed > 0 ? ` — ${sealed} sealed file${sealed === 1 ? '' : 's'} in it stay hidden from readers` : ''