fix: GlobalUsages finds every reference Divi renders, fail closed (F1, ruling P23)

Candidates come from one ID/status/type query: content with globalModule, global_module or a JSON-escaped letter. Their content is read in chunks of 50 and parsed by UsageMatcher: Divi 5 openers are json-decoded (escaped keys and values, leading zeros, any depth), and Divi 4 global_module attributes are read in any quoting. An undecodable escaped opener counts as a possible reference. Beyond 5000 candidates the rest count unread. The guide names the formats and the wp_block parity. Also fixes the grammar of the blocked-hidden count. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-09-27 04:42:46 +02:00
parent f1937ca6ff
commit fec5d9da7b
7 changed files with 299 additions and 34 deletions
@@ -12,8 +12,11 @@ Die Divi-Bibliothek speichert Layouts, Abschnitte, Zeilen und Module zur Wiederv
 - `lit/library-list` (über `mcp-adapter-execute-ability`): Einträge auflisten, Filter `layout_type`
  (`layout`, `section`, `row`, `module`), `scope` (`normal`, `global`), `search` (Titel), `category`
  (Name), weiterblättern mit `cursor`. Gelistet werden nur Einträge, die das Konto bearbeiten darf. Mit `id`:
-  ein Eintrag im Detail – Knoten-Übersicht (`outline`), `tree_rev` und die Seiten, die ihn als globales Modul
-  zeigen (`affected`, `affected_count`, `hidden_count`).
+  ein Eintrag im Detail – Knoten-Übersicht (`outline`), `tree_rev` und die Seiten, die auf ihn verweisen
+  (`affected`, `affected_count`, `hidden_count`). Gezählt wird jeder Verweis, den Divi darstellt: Divi-5-Instanzen
+  in jeder Schreibweise und Divi-4-Shortcodes mit `global_module`; ein Verweis, der sich nicht eindeutig lesen
+  lässt, zählt vorsichtshalber mit. Steckt der Eintrag in einem wiederverwendbaren Block (`wp_block`), erscheint
+  dieser Block in der Liste, nicht die Seite, die ihn einbettet – wie bei Divi selbst.
 - `lit-library-insert`: einen Eintrag in eine Divi-Seite einfügen. `mode: "global"` (Standard bei globalen
  Einträgen) fügt eine verknüpfte Instanz ein, `mode: "copy"` (Standard sonst) eine unabhängige Kopie.
  Position wie bei `lit-divi-insert` (`parent`, `before`, `after`, ohne Angabe ans Seitenende); `tree_rev`
@@ -224,9 +224,9 @@ final class GlobalEdits
        }
        if ($report->blockedHidden > 0) {
            $parts[] = sprintf(
-                '%d %s, die dieses Konto nicht lesen darf',
+                '%d %s dieses Konto nicht lesen darf',
                $report->blockedHidden,
-                $report->blockedHidden === 1 ? 'Eintrag' : 'Einträge'
+                $report->blockedHidden === 1 ? 'Eintrag, den' : 'Einträge, die'
            );
        }

@@ -10,9 +10,10 @@ defined('ABSPATH') || exit;

 /**
 * Posts (pages, Theme Builder layouts, other library entries) that show library entry $id: every
- * living post whose content has "globalModule":"<id>" – Divi's own way to find instances (string
- * scan, BlockParserStore, F10). One query per report (ID, status and type only – never the content,
- * audit L1); the rights of the caller are then checked in chunks with a bounded post cache:
+ * living post whose content references it in any form Divi renders (rows(), ruling P23 – Divi 5
+ * instances in any JSON spelling, Divi 4 global_module shortcodes; fail closed). Candidates come
+ * from one query (ID, status and type), their content is parsed in bounded chunks (audit L1); the
+ * rights of the caller are then checked in chunks with a bounded post cache:
 * - readable posts are listed (at most LIMIT, with title); unreadable ones only count (audit L4);
 * - live posts (ruling P8a) the caller may not edit_post block a confirmed edit – by ID when
 *   readable, else as a count; Theme Builder layouts also without Divi's theme_builder role option
@@ -33,6 +34,12 @@ class GlobalUsages
    /** Posts whose rights are checked with one primed cache. */
    public const CHUNK = 200;

+    /** Candidates whose content is read and parsed per call; the rest count as references (ruling P23). */
+    public const PARSE_CAP = 5000;
+
+    /** Candidate contents read per query (bounded memory, audit L1). */
+    public const CONTENT_CHUNK = 50;
+
    /** Posts that render now or will on their own: published, private, scheduled (ruling P8a). */
    public const LIVE = ['publish', 'private', 'future'];

@@ -118,22 +125,45 @@ class GlobalUsages
    }

    /**
-     * Every post (no revision, not trashed) that shows $libraryId, by ID – one query, no content. Like
-     * Divi's render, "globalModule":"0123" names entry 123 (hardening); the LIKE only narrows the
-     * rows the REGEXP looks at.
+     * Every post (no revision, not trashed) that shows $libraryId (ruling P23, fail closed). The
+     * candidates come from one query (ID, status and type only): content with "globalModule", a
+     * Divi 4 "global_module" or a JSON escape of a letter (an escaped key such as globalModule
+     * always has one). Their content is then read in chunks of CONTENT_CHUNK and parsed by
+     * UsageMatcher; beyond PARSE_CAP candidates the rest counts as references unread (audit L1).
     *
     * @return list<array{id: int, status: string, type: string}>
     */
    public function rows(int $libraryId): array
+    {
+        $candidates = $this->candidates();
+        $result = [];
+        foreach (array_chunk(array_slice($candidates, 0, self::PARSE_CAP), self::CONTENT_CHUNK) as $chunk) {
+            $contents = $this->contents(array_map(static fn (array $row): int => $row['id'], $chunk));
+            foreach ($chunk as $row) {
+                // A candidate whose content could not be read counts too (fail closed).
+                if (! isset($contents[$row['id']]) || UsageMatcher::shows($contents[$row['id']], $libraryId)) {
+                    $result[] = $row;
+                }
+            }
+        }
+
+        return array_merge($result, array_slice($candidates, self::PARSE_CAP));
+    }
+
+    /** @return list<array{id: int, status: string, type: string}> */
+    private function candidates(): array
    {
        $db = $this->db();
+        $escapedLetter = str_repeat(chr(92), 2) . 'u00[4-7][0-9a-fA-F]';
        // phpcs:disable WordPress.DB.PreparedSQL.InterpolatedNotPrepared, WordPress.DB.DirectDatabaseQuery
        $rows = $db->get_results(
            $db->prepare(
-                'SELECT ID, post_status, post_type FROM ' . $db->posts . ' WHERE post_content LIKE %s AND '
-                    . "post_content REGEXP %s AND post_type <> 'revision' AND " . self::LISTED . ' ORDER BY ID ASC',
-                '%' . $db->esc_like('"globalModule":"') . '%',
-                '"globalModule":"0*' . $libraryId . '"'
+                'SELECT ID, post_status, post_type FROM ' . $db->posts . ' WHERE (post_content LIKE %s OR '
+                    . "post_content LIKE %s OR post_content REGEXP %s) AND post_type <> 'revision' AND "
+                    . self::LISTED . ' ORDER BY ID ASC',
+                '%' . $db->esc_like('globalModule') . '%',
+                '%' . $db->esc_like('global_module') . '%',
+                $escapedLetter
            ),
            'ARRAY_A'
        );
@@ -151,6 +181,32 @@ class GlobalUsages
        return $result;
    }

+    /**
+     * @param list<int> $ids
+     * @return array<int, string> Content by ID.
+     */
+    private function contents(array $ids): array
+    {
+        $db = $this->db();
+        // phpcs:disable WordPress.DB.PreparedSQL.InterpolatedNotPrepared, WordPress.DB.DirectDatabaseQuery
+        $rows = $db->get_results(
+            $db->prepare(
+                'SELECT ID, post_content FROM ' . $db->posts . ' WHERE ID IN ('
+                    . implode(', ', array_fill(0, count($ids), '%d')) . ')',
+                $ids
+            ),
+            'ARRAY_A'
+        );
+        // phpcs:enable
+
+        $contents = [];
+        foreach (is_array($rows) ? $rows : [] as $row) {
+            $contents[(int) $row['ID']] = (string) $row['post_content'];
+        }
+
+        return $contents;
+    }
+
    /**
     * Ruling P8a: whether the caller may edit_post the live post $row; a Theme Builder layout or
     * template additionally needs Divi's role option theme_builder (audit L6 – as Divi's own
@@ -0,0 +1,87 @@
+<?php
+
+declare(strict_types=1);
+
+namespace LIT\MCP\Integrations\Divi\Library;
+
+defined('ABSPATH') || exit;
+
+/**
+ * Whether post content shows library entry $id (ruling P23, re-review F1), fail closed:
+ * - Divi 5: the textual "globalModule":"<digits>" Divi's render takes (leading zeros = intval), and
+ *   every block opener whose JSON – in any spelling, escaped keys or values included – decodes to a
+ *   globalModule with that ID at any depth;
+ * - Divi 4: a shortcode attribute global_module=<digits> in any quoting (intval);
+ * - an opener with JSON escapes (or a globalModule) that cannot be decoded, or content the regex
+ *   engine cannot scan, counts as a possible reference.
+ * Pure PHP.
+ */
+final class UsageMatcher
+{
+    /** A block opener and its attribute JSON (the Parser's delimiter grammar). */
+    private const OPENER = '/<!--\s+wp:(?:[a-z][a-z0-9_-]*\/)?[a-z][a-z0-9_-]*\s+'
+        . '({(?:(?:[^}]+|}+(?=})|(?!}\s+\/?-->).)*+)?})\s+\/?-->/s';
+
+    /** Divi 4: global_module=, any quote characters (straight, typographic, entities) before the digits. */
+    private const SHORTCODE = '/\bglobal_module\s*=\s*[^\d\s\]]{0,12}(\d+)/i';
+
+    public static function shows(string $content, int $id): bool
+    {
+        $textual = preg_match('/"globalModule":"0*' . $id . '"/', $content);
+        if ($textual !== 0) {
+            return true; // A match, or a regex failure (fail closed).
+        }
+
+        if (stripos($content, 'global_module') !== false) {
+            $found = preg_match_all(self::SHORTCODE, $content, $matches);
+            if ($found === false) {
+                return true;
+            }
+            foreach ($matches[1] as $value) {
+                if ((int) $value === $id) {
+                    return true;
+                }
+            }
+        }
+
+        $escape = chr(92) . 'u';
+        if (! str_contains($content, 'globalModule') && ! str_contains($content, $escape)) {
+            return false;
+        }
+
+        $found = preg_match_all(self::OPENER, $content, $matches);
+        if ($found === false) {
+            return true;
+        }
+        foreach ($matches[1] as $json) {
+            if (! str_contains($json, $escape) && ! str_contains($json, 'globalModule')) {
+                continue;
+            }
+            $attrs = json_decode($json, true);
+            if (! is_array($attrs) || self::names($attrs, $id)) {
+                return true;
+            }
+        }
+
+        return false;
+    }
+
+    /** @param array<array-key, mixed> $attrs Whether a globalModule key at any depth names $id. */
+    private static function names(array $attrs, int $id): bool
+    {
+        foreach ($attrs as $key => $value) {
+            if (
+                $key === 'globalModule'
+                && (is_int($value) || (is_string($value) && preg_match('/^\d+$/D', $value) === 1))
+                && (int) $value === $id
+            ) {
+                return true;
+            }
+            if (is_array($value) && self::names($value, $id)) {
+                return true;
+            }
+        }
+
+        return false;
+    }
+}
@@ -342,7 +342,7 @@ final class GlobalEditsTest extends TestCase
        self::assertStringNotContainsString('1081', (string) wp_json_encode($data));
        self::assertToolError($refused, 'forbidden');
        self::assertStringNotContainsString('1081', $refused->get_error_message());
-        self::assertStringContainsString('nicht lesen darf', $refused->get_error_message());
+        self::assertStringContainsString('1 Eintrag, den dieses Konto nicht lesen darf', $refused->get_error_message());
        self::assertSame([], $this->saved);
    }

@@ -13,37 +13,78 @@ use WP_Post;

 final class GlobalUsagesTest extends TestCase
 {
-    /** Audit L1: one query, ID, status and type only – never the content, no COUNT, no second scan. */
-    public function testFindsInstancesLikeDiviWithOneQueryWithoutContent(): void
+    /**
+     * Ruling P23 (re-review F1): candidates by a wide, fail-closed prefilter (ID, status and type
+     * only), then their content in chunks, parsed in PHP (UsageMatcher) – every format Divi renders.
+     */
+    public function testFindsEveryFormatDiviRendersFromCandidates(): void
    {
+        $bs = chr(92);
+        $contents = [
+            1053 => '<!-- wp:divi/global-layout {"globalModule":"1058","blockName":"divi/text"} /-->',
+            1054 => '[et_pb_section global_module="01058"][/et_pb_section]',
+            1055 => '<!-- wp:divi/global-layout {"' . $bs . 'u0067lobalModule":"1058"} /-->',
+            1056 => '<!-- wp:divi/global-layout {"globalModule":"' . $bs . 'u00310' . '58"} /-->',
+            1057 => '<!-- wp:divi/global-layout {"globalModule":"1059"} /-->',
+            1061 => '<!-- wp:divi/global-layout {"' . $bs . 'u0067lobalModule":"1058",} /-->',
+        ];
        $db = Mockery::mock(\wpdb::class)->makePartial();
        $queries = [];
-        $db->allows('get_results')->andReturnUsing(static function (string $sql) use (&$queries): array {
+        $db->allows('get_results')->andReturnUsing(static function (string $sql) use (&$queries, $contents): array {
            $queries[] = $sql;
+            if (str_contains($sql, 'post_content FROM')) {
+                preg_match('/IN \(([0-9, ]+)\)/', $sql, $in);
+                $ids = array_map('intval', explode(',', $in[1]));

-            return [
-                ['ID' => '1053', 'post_status' => 'draft', 'post_type' => 'page'],
-                ['ID' => '1061', 'post_status' => 'private', 'post_type' => 'et_header_layout'],
-            ];
+                return array_map(static fn (int $id): array => ['ID' => (string) $id, 'post_content' => $contents[$id]], $ids);
+            }
+
+            return array_map(
+                static fn (int $id): array => ['ID' => (string) $id, 'post_status' => 'publish', 'post_type' => 'page'],
+                array_keys($contents)
+            );
        });

        $rows = (new GlobalUsages($db))->rows(1058);

-        self::assertSame(
-            [
-                ['id' => 1053, 'status' => 'draft', 'type' => 'page'],
-                ['id' => 1061, 'status' => 'private', 'type' => 'et_header_layout'],
-            ],
-            $rows
-        );
-        self::assertCount(1, $queries);
+        self::assertSame([1053, 1054, 1055, 1056, 1061], array_column($rows, 'id'), '1061: undecodable, fail closed');
        self::assertStringStartsWith('SELECT ID, post_status, post_type FROM', $queries[0]);
-        self::assertStringContainsString("post_content LIKE '%\"globalModule\":\"%'", $queries[0]);
-        self::assertStringContainsString("post_content REGEXP '\"globalModule\":\"0*1058\"'", $queries[0]);
+        self::assertStringNotContainsString('post_content,', $queries[0]);
+        self::assertStringContainsString("post_content LIKE '%globalModule%'", $queries[0]);
+        self::assertStringContainsString("post_content LIKE '%global" . $bs . "_module%'", $queries[0]);
+        self::assertStringContainsString("post_content REGEXP '" . $bs . $bs . 'u00[4-7][0-9a-fA-F]' . "'", $queries[0]);
        self::assertStringContainsString("post_status NOT IN ('trash', 'auto-draft', 'inherit')", $queries[0]);
        self::assertStringContainsString("post_type <> 'revision'", $queries[0]);
-        self::assertStringNotContainsString('LIMIT', $queries[0]);
-        self::assertStringNotContainsString('post_content,', $queries[0]);
+        self::assertCount(2, $queries, 'one candidate query, one content chunk');
+    }
+
+    /** Audit L1: contents are read in chunks; beyond the parse cap candidates count as references. */
+    public function testReadsContentsInChunksAndFailsClosedBeyondTheCap(): void
+    {
+        $total = GlobalUsages::PARSE_CAP + 3;
+        $db = Mockery::mock(\wpdb::class)->makePartial();
+        $chunks = 0;
+        $db->allows('get_results')->andReturnUsing(static function (string $sql) use (&$chunks, $total): array {
+            if (str_contains($sql, 'post_content FROM')) {
+                ++$chunks;
+                preg_match('/IN \(([0-9, ]+)\)/', $sql, $in);
+
+                return array_map(
+                    static fn (string $id): array => ['ID' => trim($id), 'post_content' => 'nichts'],
+                    explode(',', $in[1])
+                );
+            }
+
+            return array_map(
+                static fn (int $id): array => ['ID' => (string) $id, 'post_status' => 'draft', 'post_type' => 'page'],
+                range(1, $total)
+            );
+        });
+
+        $rows = (new GlobalUsages($db))->rows(1058);
+
+        self::assertSame((int) ceil(GlobalUsages::PARSE_CAP / GlobalUsages::CONTENT_CHUNK), $chunks);
+        self::assertSame(range(GlobalUsages::PARSE_CAP + 1, $total), array_column($rows, 'id'));
    }

    /** Audit L4: readable posts with title, unreadable ones only counted. */
@@ -0,0 +1,78 @@
+<?php
+
+declare(strict_types=1);
+
+namespace LIT\MCP\Tests\Unit\Integrations\Divi\Library;
+
+use LIT\MCP\Integrations\Divi\Library\UsageMatcher;
+use PHPUnit\Framework\TestCase;
+
+/**
+ * Ruling P23 (re-review F1): every reference Divi renders counts – Divi 5 openers in any JSON
+ * spelling, Divi 4 shortcodes in any quoting; an escaped opener that cannot be decoded counts as a
+ * possible reference (fail closed). Escape bytes are built with chr(92), never written inline.
+ */
+final class UsageMatcherTest extends TestCase
+{
+    private static function u(string $hex): string
+    {
+        return chr(92) . 'u' . $hex;
+    }
+
+    /** @return array<string, array{string, bool}> Content => whether it shows entry 123. */
+    public static function contents(): array
+    {
+        $escapedKey = '"' . self::u('0067') . 'lobalModule"';
+
+        return [
+            'divi 5 canonical' => ['<!-- wp:divi/global-layout {"globalModule":"123","blockName":"divi/text"} /-->', true],
+            'divi 5 leading zeros' => ['<!-- wp:divi/global-layout {"globalModule":"0123","blockName":"divi/text"} /-->', true],
+            'divi 5 escaped key' => ['<!-- wp:divi/global-layout {' . $escapedKey . ':"123","blockName":"divi/text"} /-->', true],
+            'divi 5 escaped value' => [
+                '<!-- wp:divi/global-layout {"globalModule":"' . self::u('0031') . '23","blockName":"divi/text"} /-->',
+                true,
+            ],
+            'divi 5 escaped key with leading zeros' => [
+                '<!-- wp:divi/global-layout {' . $escapedKey . ':"00123"} /-->',
+                true,
+            ],
+            'divi 5 another entry' => ['<!-- wp:divi/global-layout {"globalModule":"1234","blockName":"divi/text"} /-->', false],
+            'divi 5 escaped key, another entry' => ['<!-- wp:divi/global-layout {' . $escapedKey . ':"124"} /-->', false],
+            'divi 5 id in local attributes' => [
+                '<!-- wp:divi/global-layout {"globalModule":"5","localAttrs":{"x":{"globalModule":"123"}}} /-->',
+                true,
+            ],
+            'divi 4 double quotes' => ['[et_pb_section global_module="123"][/et_pb_section]', true],
+            'divi 4 single quotes and zeros' => ["[et_pb_text global_module='0123']x[/et_pb_text]", true],
+            'divi 4 unquoted' => ['[et_pb_row global_module=123 admin_label="x"][/et_pb_row]', true],
+            'divi 4 typographic quotes' => ['[et_pb_row global_module=' . "\u{201D}" . '123' . "\u{201D}" . '][/et_pb_row]', true],
+            'divi 4 another entry' => ['[et_pb_section global_module="1230"][/et_pb_section]', false],
+            'undecodable escaped opener' => [
+                '<!-- wp:divi/global-layout {' . $escapedKey . ':"123",} /-->',
+                true,
+            ],
+            'undecodable escaped opener naming nothing readable' => [
+                '<!-- wp:divi/text {' . self::u('0067') . 'kaputt} /-->',
+                true,
+            ],
+            'escaped markup only' => [
+                '<!-- wp:divi/text {"content":{"innerContent":{"desktop":{"value":"' . self::u('003c') . 'p'
+                    . self::u('003e') . 'x"}}}} /-->',
+                false,
+            ],
+            'plain page' => ['<!-- wp:divi/section --><!-- /wp:divi/section -->', false],
+        ];
+    }
+
+    /** @dataProvider contents */
+    public function testFindsEveryReferenceDiviRenders(string $content, bool $shows): void
+    {
+        self::assertSame($shows, UsageMatcher::shows($content, 123));
+    }
+
+    public function testTheEscapeBytesAreReal(): void
+    {
+        self::assertSame(6, strlen(self::u('0067')));
+        self::assertSame(92, ord(self::u('0067')[0]));
+    }
+}