Add connected pages: hierarchy, backlinks, and live embeds

Implement Notion-like upgrade phase 3: owner-scoped nested pages with cycle rejection and child promotion on delete; Markdown-derived internal links and backlinks; live collection embeds resolved in the content-save transaction. Add migration 008, API/docs, integration coverage, and page/sidebar UI.

internal/handlers/page_graph.go

@@ -0,0 +1,196 @@
+package handlers
+
+import (
+	"context"
+	"database/sql"
+	"net/url"
+	"regexp"
+	"strings"
+
+	"github.com/andrew/go-computer/internal/db"
+)
+
+// Connected pages (Notion-like upgrades phase 3).
+//
+// Page hierarchy, internal links and embedded collection views are all
+// expressed in the page's Markdown, which stays the single source of truth:
+//
+//   - [Any label](computer:page/<id>) links to another page. /pages?id=<id>
+//     and page/<id> are accepted as aliases so hand-written links work too.
+//   - {{collection:<id>}} embeds a live collection view; an optional
+//     {{collection:<id>|view=<viewId>}} pins a saved view's filters/sorts.
+//
+// On every content save the resolved PageLink and PageEmbed rows are rebuilt
+// inside the same transaction. Only targets owned by the author are stored:
+// backlinks therefore cannot enumerate or reveal titles from another account.
+// Embedding displays data through the normal collection endpoints, so the
+// underlying bookmark-visibility checks still apply and nothing is published.
+
+var (
+	markdownLinkRE = regexp.MustCompile(`\[[^\]]*\]\(\s*<?([^)\s>]+)(?:\s+"[^"]*")?\s*>?\)`)
+	pageEmbedRE    = regexp.MustCompile(`\{\{\s*collection:([A-Za-z0-9_-]{1,128})(?:\s*\|\s*view=([A-Za-z0-9_-]{1,128}))?\s*\}\}`)
+	pageRefRE      = regexp.MustCompile(`^[A-Za-z0-9_-]{1,128}$`)
+)
+
+// internalPageID normalizes a Markdown link target to a page id. It returns
+// false for ordinary web links and malformed ids.
+func internalPageID(href string) (string, bool) {
+	href = strings.TrimSpace(strings.TrimSuffix(strings.TrimPrefix(href, "<"), ">"))
+	raw := ""
+	switch {
+	case strings.HasPrefix(href, "computer://page/"):
+		raw = strings.TrimPrefix(href, "computer://page/")
+	case strings.HasPrefix(href, "computer:page/"):
+		raw = strings.TrimPrefix(href, "computer:page/")
+	case strings.HasPrefix(href, "page/"):
+		raw = strings.TrimPrefix(href, "page/")
+	case strings.HasPrefix(href, "/pages/"):
+		raw = strings.TrimPrefix(href, "/pages/")
+	case strings.HasPrefix(href, "/pages?"):
+		if u, err := url.Parse(href); err == nil {
+			raw = u.Query().Get("id")
+		}
+	case strings.HasPrefix(href, "computer:page?id="):
+		if u, err := url.Parse(href); err == nil {
+			raw = u.Query().Get("id")
+		}
+	case strings.HasPrefix(href, "computer://page?id="):
+		if u, err := url.Parse(href); err == nil {
+			raw = u.Query().Get("id")
+		}
+	}
+	raw = strings.TrimSuffix(strings.TrimSpace(raw), "/")
+	if !pageRefRE.MatchString(raw) {
+		return "", false
+	}
+	return raw, true
+}
+
+type pageEmbedRef struct {
+	collectionID string
+	viewID       string
+}
+
+// pageGraph extracts the distinct page links and embeds from Markdown.
+func pageGraph(content, selfID string) ([]string, []pageEmbedRef) {
+	links := make([]string, 0)
+	seenLinks := map[string]bool{}
+	for _, m := range markdownLinkRE.FindAllStringSubmatchIndex(content, -1) {
+		// Skip images: ![alt](target) is not an internal link.
+		if m[0] > 0 && content[m[0]-1] == '!' {
+			continue
+		}
+		id, ok := internalPageID(content[m[2]:m[3]])
+		if !ok || id == selfID || seenLinks[id] {
+			continue
+		}
+		seenLinks[id] = true
+		links = append(links, id)
+	}
+	embeds := make([]pageEmbedRef, 0)
+	seenEmbeds := map[string]bool{}
+	for _, m := range pageEmbedRE.FindAllStringSubmatch(content, -1) {
+		ref := pageEmbedRef{collectionID: m[1], viewID: m[2]}
+		if seenEmbeds[ref.collectionID] {
+			continue
+		}
+		seenEmbeds[ref.collectionID] = true
+		embeds = append(embeds, ref)
+	}
+	return links, embeds
+}
+
+// syncPageGraph replaces one page's resolved links and embeds. It runs inside
+// the content-save transaction, so the derived rows and the Markdown commit
+// together (a failed save never leaves stale backlinks).
+func syncPageGraph(ctx context.Context, tx *sql.Tx, pageID, uid, content string) error {
+	links, embeds := pageGraph(content, pageID)
+
+	if _, err := tx.ExecContext(ctx, `DELETE FROM "PageLink" WHERE "sourcePageId"=$1`, pageID); err != nil {
+		return err
+	}
+	for _, target := range links {
+		var visible bool
+		if err := tx.QueryRowContext(ctx, `SELECT EXISTS(SELECT 1 FROM "Page" WHERE id=$1 AND "ownerId"=$2)`, target, uid).Scan(&visible); err != nil {
+			return err
+		}
+		// A foreign/missing target is ignored rather than stored. The link
+		// text remains in the author's Markdown, but it can never become a
+		// cross-account backlink.
+		if !visible {
+			continue
+		}
+		if _, err := tx.ExecContext(ctx, `INSERT INTO "PageLink"("sourcePageId","targetPageId") VALUES($1,$2) ON CONFLICT DO NOTHING`, pageID, target); err != nil {
+			return err
+		}
+	}
+
+	if _, err := tx.ExecContext(ctx, `DELETE FROM "PageEmbed" WHERE "pageId"=$1`, pageID); err != nil {
+		return err
+	}
+	for i, ref := range embeds {
+		var visible bool
+		if err := tx.QueryRowContext(ctx, `SELECT EXISTS(SELECT 1 FROM "Collection" WHERE id=$1 AND "ownerId"=$2)`, ref.collectionID, uid).Scan(&visible); err != nil {
+			return err
+		}
+		if !visible {
+			continue
+		}
+		viewID := any(nil)
+		if ref.viewID != "" {
+			var valid bool
+			if err := tx.QueryRowContext(ctx, `SELECT EXISTS(SELECT 1 FROM "CollectionView" WHERE id=$1 AND "collectionId"=$2)`, ref.viewID, ref.collectionID).Scan(&valid); err != nil {
+				return err
+			}
+			// An unknown view falls back to the collection's default view
+			// instead of dropping the embed entirely.
+			if valid {
+				viewID = ref.viewID
+			}
+		}
+		if _, err := tx.ExecContext(ctx, `INSERT INTO "PageEmbed"("pageId","collectionId","viewId",position) VALUES($1,$2,$3,$4)`, pageID, ref.collectionID, viewID, i); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+// pageParent validates a requested parent and rejects cycles. The walk stays
+// within the owner's pages and is bounded by the seen set, so a corrupt
+// pre-existing cycle cannot loop forever.
+func pageParent(ctx context.Context, q db.Queryer, pageID, uid string, raw any) (any, error) {
+	if raw == nil {
+		return nil, nil
+	}
+	parent, ok := raw.(string)
+	if !ok {
+		return nil, bad("parentId must be a page id or null")
+	}
+	parent = strings.TrimSpace(parent)
+	if parent == "" {
+		return nil, nil
+	}
+	if !pageRefRE.MatchString(parent) {
+		return nil, bad("Invalid parentId")
+	}
+	if parent == pageID {
+		return nil, bad("A page cannot be its own parent")
+	}
+	seen := map[string]bool{pageID: true}
+	for current := parent; current != ""; {
+		if seen[current] {
+			return nil, bad("Page hierarchy cannot contain a cycle")
+		}
+		seen[current] = true
+		var next sql.NullString
+		err := q.QueryRowContext(ctx, `SELECT "parentId" FROM "Page" WHERE id=$1 AND "ownerId"=$2`, current, uid).Scan(&next)
+		if err == sql.ErrNoRows {
+			return nil, bad("Parent page not found")
+		}
+		if err != nil {
+			return nil, err
+		}
+		current = next.String
+	}
+	return parent, nil
+}

internal/handlers/pages.go

@@ -18,7 +18,23 @@ import (
 func pageView(ctx context.Context, q db.Queryer, id, uid string) (object, error) {
 	// Linked bookmarks use the viewer projection (review PRIV-02): a page can
 	// link bookmarks shared with its owner, and those arrive narrowed.
-	return db.One(ctx, q, `SELECT to_jsonb(p)||jsonb_build_object('bookmarks',COALESCE((SELECT jsonb_agg(to_jsonb(pb)||jsonb_build_object('bookmark',`+db.BookmarkViewerView+`) ORDER BY pb.position) FROM "PageBookmark" pb JOIN "Bookmark" b ON b.id=pb."bookmarkId" WHERE pb."pageId"=p.id AND `+db.BookmarkVisible+`),'[]'::jsonb)) FROM "Page" p WHERE p.id=$1 AND p."ownerId"=$2`, id, uid)
+	//
+	// Connected pages (phase 3): backlinks and embeds are owner-scoped in the
+	// query itself, not trusted from the caller. `links` is the resolved
+	// outgoing edge set (useful for navigation); the Markdown remains the
+	// source of truth and is rebuilt by syncPageGraph on every content save.
+	return db.One(ctx, q, `SELECT to_jsonb(p)||jsonb_build_object(
+	'bookmarks',COALESCE((SELECT jsonb_agg(to_jsonb(pb)||jsonb_build_object('bookmark',`+db.BookmarkViewerView+`) ORDER BY pb.position) FROM "PageBookmark" pb JOIN "Bookmark" b ON b.id=pb."bookmarkId" WHERE pb."pageId"=p.id AND `+db.BookmarkVisible+`),'[]'::jsonb),
+	'backlinks',COALESCE((SELECT jsonb_agg(jsonb_build_object('id',s.id,'title',s.title,'updatedAt',s."updatedAt") ORDER BY s."updatedAt" DESC,s.id) FROM "PageLink" pl JOIN "Page" s ON s.id=pl."sourcePageId" WHERE pl."targetPageId"=p.id AND s."ownerId"=p."ownerId"),'[]'::jsonb),
+	'links',COALESCE((SELECT jsonb_agg(jsonb_build_object('id',t.id,'title',t.title) ORDER BY t.title,t.id) FROM "PageLink" pl JOIN "Page" t ON t.id=pl."targetPageId" WHERE pl."sourcePageId"=p.id AND t."ownerId"=p."ownerId"),'[]'::jsonb),
+	'embeds',COALESCE((SELECT jsonb_agg(jsonb_build_object(
+		'collectionId',e."collectionId",'viewId',e."viewId",'position',e.position,
+		'collection',to_jsonb(c)||jsonb_build_object(
+			'properties',COALESCE((SELECT jsonb_agg(to_jsonb(pr) ORDER BY pr.position,pr.id) FROM "CollectionProperty" pr WHERE pr."collectionId"=c.id),'[]'::jsonb),
+			'views',COALESCE((SELECT jsonb_agg(to_jsonb(v) ORDER BY v.position,v.id) FROM "CollectionView" v WHERE v."collectionId"=c.id),'[]'::jsonb)),
+		'view',(SELECT to_jsonb(v) FROM "CollectionView" v WHERE v.id=e."viewId"))
+		ORDER BY e.position,e."collectionId") FROM "PageEmbed" e JOIN "Collection" c ON c.id=e."collectionId" WHERE e."pageId"=p.id AND c."ownerId"=p."ownerId"),'[]'::jsonb)
+	) FROM "Page" p WHERE p.id=$1 AND p."ownerId"=$2`, id, uid)
 }
 func page(r *http.Request, id string) (object, error) {
 	return pageView(r.Context(), database(r), id, user(r))
@@ -31,12 +47,15 @@ func HandlePagesList(w http.ResponseWriter, r *http.Request) {
 	// literal characters, not patterns. The pg_trgm indexes on title/content
 	// let Postgres use the same ILIKE predicate without scanning every page.
 	search := "%" + likeEscaper.Replace(r.URL.Query().Get("search")) + "%"
-	v, e := db.Many(r.Context(), database(r), `SELECT (to_jsonb(p)-'content')||jsonb_build_object('_count',jsonb_build_object('bookmarks',(SELECT count(*) FROM "PageBookmark" pb WHERE pb."pageId"=p.id))) FROM "Page" p WHERE "ownerId"=$1 AND (title ILIKE $2 ESCAPE '\' OR content ILIKE $2 ESCAPE '\') ORDER BY pinned DESC,position,"createdAt"`, user(r), search)
+	v, e := db.Many(r.Context(), database(r), `SELECT (to_jsonb(p)-'content')||jsonb_build_object('_count',jsonb_build_object(
+		'bookmarks',(SELECT count(*) FROM "PageBookmark" pb WHERE pb."pageId"=p.id),
+		'children',(SELECT count(*) FROM "Page" ch WHERE ch."parentId"=p.id AND ch."ownerId"=p."ownerId"))) FROM "Page" p WHERE "ownerId"=$1 AND (title ILIKE $2 ESCAPE '\' OR content ILIKE $2 ESCAPE '\') ORDER BY pinned DESC,position,"createdAt"`, user(r), search)
 	respond(w, 200, v, e)
 }
 func pageData(b object) (object, error) {
-	d := fields(b, "title content icon pinned")
-	if e := stringFields(d, "icon", "title content"); e != nil {
+	d := fields(b, "title content icon pinned parentId")
+	// parentId is nullable: an explicit null promotes the page to the root.
+	if e := stringFields(d, "icon parentId", "title content"); e != nil {
 		return nil, e
 	}
 	if e := boolFields(d, "pinned"); e != nil {
@@ -80,6 +99,13 @@ func HandlePagesCreate(w http.ResponseWriter, r *http.Request) {
 		if e := lockUser(r, tx); e != nil {
 			return e
 		}
+		if raw, ok := d["parentId"]; ok {
+			parent, e := pageParent(r.Context(), tx, d["id"].(string), user(r), raw)
+			if e != nil {
+				return e
+			}
+			d["parentId"] = parent
+		}
 		var n int
 		if e := tx.QueryRowContext(r.Context(), `SELECT COALESCE(max(position),-1)+1 FROM "Page" WHERE "ownerId"=$1`, user(r)).Scan(&n); e != nil {
 			return e
@@ -87,6 +113,17 @@ func HandlePagesCreate(w http.ResponseWriter, r *http.Request) {
 		d["position"] = n
 		var e error
 		v, e = db.Insert(r.Context(), tx, "Page", d)
+		if e != nil {
+			return e
+		}
+		if content, ok := d["content"].(string); ok {
+			if e := syncPageGraph(r.Context(), tx, d["id"].(string), user(r), content); e != nil {
+				return e
+			}
+		}
+		// Acknowledge the same snapshot the write produced (the update path's
+		// CR-18 discipline), so the 201 already carries links/embeds.
+		v, e = pageView(r.Context(), tx, d["id"].(string), user(r))
 		return e
 	})
 	respond(w, 201, v, e)
@@ -162,13 +199,25 @@ func HandlePagesUpdate(w http.ResponseWriter, r *http.Request) {
 				return conflict("Page was modified elsewhere; reload to see the latest version")
 			}
 		}
+		if raw, ok := d["parentId"]; ok {
+			parent, e := pageParent(r.Context(), tx, id, user(r), raw)
+			if e != nil {
+				return e
+			}
+			d["parentId"] = parent
+		}
 		if _, e := db.Update(r.Context(), tx, "Page", id, "ownerId", user(r), d); e != nil {
 			return e
 		}
-		if _, ok := d["content"]; ok {
+		if content, ok := d["content"].(string); ok {
 			if _, e := tx.ExecContext(r.Context(), `UPDATE "Page" SET version=version+1 WHERE id=$1`, id); e != nil {
 				return e
 			}
+			// Rebuild resolved links/embeds in the same transaction as the
+			// Markdown they were derived from.
+			if e := syncPageGraph(r.Context(), tx, id, user(r), content); e != nil {
+				return e
+			}
 		}
 		if bid := str(b, "addBookmarkId"); bid != "" {
 			if _, e := db.One(r.Context(), tx, `SELECT to_jsonb(b) FROM "Bookmark" b WHERE b.id=$1 AND `+db.BookmarkVisible, bid, user(r)); e != nil {
@@ -190,7 +239,22 @@ func HandlePagesUpdate(w http.ResponseWriter, r *http.Request) {
 	})
 	respond(w, 200, v, e)
 }
+
+// HandlePagesBacklinks returns the pages that link to this one. It is a
+// separate read surface for navigation; the same owner-scoped list is
+// embedded in pageView for the editor.
+func HandlePagesBacklinks(w http.ResponseWriter, r *http.Request) {
+	id := r.PathValue("id")
+	if _, e := owned(r, database(r), "Page", id, "ownerId"); e != nil {
+		respond(w, 0, nil, e)
+		return
+	}
+	v, e := db.Many(r.Context(), database(r), `SELECT jsonb_build_object('id',s.id,'title',s.title,'updatedAt',s."updatedAt") FROM "PageLink" pl JOIN "Page" s ON s.id=pl."sourcePageId" WHERE pl."targetPageId"=$1 AND s."ownerId"=$2 ORDER BY s."updatedAt" DESC,s.id`, id, user(r))
+	respond(w, 200, v, e)
+}
 func HandlePagesDelete(w http.ResponseWriter, r *http.Request) {
+	// Children are promoted to the top level by the self-FK's SET NULL; a
+	// delete never silently takes nested pages with it.
 	respond(w, 200, OK{true}, remove(r, "Page", r.PathValue("id"), "ownerId"))
 }
 func HandlePagesMarkdown(w http.ResponseWriter, r *http.Request) {