fix: Task 8 review round 1 - quote-aware tokenizer, context check, skipped global nodes (S45)

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01At3V6SjwbMe6eMzVGW3Hs7
2026-09-26 16:32:00 +02:00
parent bb85868192
commit 93d2bde286
6 changed files with 893 additions and 114 deletions
@@ -39,6 +39,8 @@ final class DiviReplaceTextAbility implements Ability

    private const SAMPLES = 5;

+    public const GLOBAL_REASON = 'globales Modul – in der Bibliothek bearbeiten';
+
    public function __construct(
        private ContentSchema $schema,
        private DiviPageLoader $loader,
@@ -63,7 +65,9 @@ final class DiviReplaceTextAbility implements Ability
            . 'dynamischen Inhalte) auf den Divi-Seiten eines Inhaltstyps. Standard ist dry_run: true – liefert die '
            . 'Treffer je Seite mit tree_rev. Zum Ersetzen dry_run: false, targets = [{post_id, tree_rev}] aus dem '
            . 'Probelauf und confirm: true nach Rückfrage beim Nutzer. Jede Seite wird einzeln gespeichert und erhält '
-            . 'eine eigene Revision; results meldet jede Seite.';
+            . 'eine eigene Revision; results meldet jede Seite. In reinen Textfeldern (Titel, Button-Texte) werden '
+            . 'Leerzeichen am Anfang und Ende der Ersetzung entfernt. Globale Module werden nicht geändert, sondern '
+            . 'in skipped_nodes genannt (in der Divi-Bibliothek bearbeiten).';
    }

    public function inputSchema(): array
@@ -104,6 +108,7 @@ final class DiviReplaceTextAbility implements Ability
            'dry_run' => ['type' => 'boolean'],
            'matches' => ['type' => 'array', 'items' => ['type' => 'object']],
            'skipped' => ['type' => 'array', 'items' => ['type' => 'object']],
+            'skipped_nodes' => ['type' => 'array', 'items' => ['type' => 'object']],
            'next_cursor' => ['type' => ['string', 'null']],
            'results' => ['type' => 'array', 'items' => ['type' => 'object']],
        ], ['dry_run']);
@@ -169,6 +174,7 @@ final class DiviReplaceTextAbility implements Ability
        $page = Cursor::page($candidates, $limit, $offset);
        $matches = [];
        $skipped = [];
+        $skippedNodes = [];
        $total = 0;
        foreach ($page['items'] as $postId) {
            $postId = (int) $postId;
@@ -190,7 +196,10 @@ final class DiviReplaceTextAbility implements Ability
                continue;
            }

-            if ($found['count'] > 0) {
+            foreach ($found['skipped_nodes'] as $node) {
+                $skippedNodes[] = ['post_id' => $postId] + $node;
+            }
+            if ($found['count'] > 0 || $found['skipped_nodes'] !== []) {
                $total += $found['count'];
                $matches[] = [
                    'post_id' => $postId,
@@ -199,19 +208,25 @@ final class DiviReplaceTextAbility implements Ability
                    'tree_rev' => $divi->treeRev(),
                    'count' => $found['count'],
                    'samples' => $found['samples'],
+                    'skipped_nodes' => count($found['skipped_nodes']),
                ];
            }
        }

        return ToolResult::ok(sprintf(
-            'Probelauf: %d Treffer auf %d Seiten – nichts geändert.%s',
+            'Probelauf: %d Treffer auf %d Seiten – nichts geändert.%s%s',
            $total,
            count($matches),
+            $skippedNodes === [] ? '' : sprintf(
+                ' %d Knoten (globale Module) werden nicht geändert, siehe skipped_nodes.',
+                count($skippedNodes)
+            ),
            $page['next_cursor'] === null ? '' : ' Weitere Seiten mit next_cursor prüfen.'
        ))
            ->with('dry_run', true)
            ->with('matches', $matches)
            ->with('skipped', $skipped)
+            ->with('skipped_nodes', $skippedNodes)
            ->with('next_cursor', $page['next_cursor'])
            ->toArray();
    }
@@ -263,7 +278,12 @@ final class DiviReplaceTextAbility implements Ability
            count($results),
            array_sum(array_column($replaced, 'count'))
        ));
+        $skippedNodes = [];
        foreach ($results as $entry) {
+            foreach ($entry['skipped_nodes'] ?? [] as $node) {
+                $skippedNodes[] = ['post_id' => $entry['post_id']] + $node;
+                $tool->warning(sprintf('Eintrag %d, Knoten %s: %s', $entry['post_id'], $node['id'], $node['reason']));
+            }
            if (! $entry['ok']) {
                $tool->warning(sprintf('Eintrag %d: %s', $entry['post_id'], $entry['error']));
                continue;
@@ -277,7 +297,10 @@ final class DiviReplaceTextAbility implements Ability
            }
        }

-        return $tool->with('dry_run', false)->with('results', $results)->toArray();
+        return $tool->with('dry_run', false)
+            ->with('results', $results)
+            ->with('skipped_nodes', $skippedNodes)
+            ->toArray();
    }

    /**
@@ -330,6 +353,7 @@ final class DiviReplaceTextAbility implements Ability
            return $base + ['ok' => false, 'error' => $e->getMessage() . ' Diese Seite wurde nicht geändert.'];
        }

+        $base['skipped_nodes'] = $found['skipped_nodes'];
        if ($found['count'] === 0) {
            return $base + ['ok' => true, 'count' => 0, 'tree_rev' => $page->treeRev()];
        }
@@ -357,7 +381,7 @@ final class DiviReplaceTextAbility implements Ability
     * the entry's AttrWriter (field kind; markers rejected); URL and code fields are skipped.
     *
     * @return array{count: int, samples: list<array<string, string>>,
-     *     blocks: list<array{Block, array<string, mixed>}>}
+     *     blocks: list<array{Block, array<string, mixed>}>, skipped_nodes: list<array{id: string, reason: string}>}
     * @throws InvalidArgumentException When a place would be changed unsafely – the page fails.
     */
    private function replace(DiviPage $page, TextReplacer $replacer): array
@@ -366,9 +390,16 @@ final class DiviReplaceTextAbility implements Ability
        $count = 0;
        $samples = [];
        $blocks = [];
+        $skipped = [];
        foreach ($page->index->entries() as $entry) {
            $block = $entry['block'];
-            if (! $block->isDivi() || $block->isOpaque()) {
+            if ($block->isDivi() && $block->isOpaque()) {
+                if ($replacer->mentions(Serializer::block($block))) {
+                    $skipped[] = ['id' => (string) $entry['id'], 'reason' => self::opaqueReason($block)];
+                }
+                continue;
+            }
+            if (! $block->isDivi()) {
                continue;
            }

@@ -397,30 +428,22 @@ final class DiviReplaceTextAbility implements Ability
            }
        }

-        return ['count' => $count, 'samples' => $samples, 'blocks' => $blocks];
+        return ['count' => $count, 'samples' => $samples, 'blocks' => $blocks, 'skipped_nodes' => $skipped];
    }

    /**
-     * IDs of Divi entries whose content contains the search text as stored: raw, as
-     * serialize_block_attributes() escapes it (<, >, &, " and --), and additionally with \uXXXX
-     * for non-ASCII (WordPress' PHP serializer). Case-insensitive adds the lower and upper case
-     * forms (LIKE is case-insensitive for letters, not for the hex digits of \u escapes). The
-     * exact match happens in the tree; this is only a prefilter.
+     * IDs of Divi entries whose content contains the search text in one of its stored forms
+     * (TextReplacer::forms()). The exact match happens in the tree; this is only a prefilter.
     *
     * @return list<int>
     */
    private function candidates(string $postType, string $search, bool $caseSensitive, int $limit, int $offset): array
    {
        $db = $this->db();
-        $variants = $caseSensitive ? [$search] : [$search, mb_strtolower($search), mb_strtoupper($search)];
-        $forms = [];
-        foreach ($variants as $variant) {
-            $escaped = substr(Serializer::attributes(['s' => $variant]), 6, -2);
-            array_push($forms, $variant, $escaped, self::unicodeEscaped($escaped));
-        }
+        $forms = TextReplacer::forms($search, $caseSensitive);
        $likes = array_map(
            static fn (string $form): string => '%' . $db->esc_like($form) . '%',
-            array_values(array_unique($forms))
+            $forms
        );

        $sql = "SELECT ID FROM {$db->posts} WHERE post_type = %s"
@@ -443,14 +466,12 @@ final class DiviReplaceTextAbility implements Ability
        return array_map('intval', is_array($ids) ? $ids : []);
    }

-    /** Non-ASCII characters as JSON \uXXXX escapes (UTF-16, lower-case hex like json_encode()). */
-    private static function unicodeEscaped(string $value): string
+    /** Why an opaque node is not changed (global modules are edited in the Divi library, Etappe 3). */
+    private static function opaqueReason(Block $block): string
    {
-        return (string) preg_replace_callback('/[^\x00-\x7F]/u', static function (array $match): string {
-            $units = unpack('n*', (string) mb_convert_encoding($match[0], 'UTF-16BE', 'UTF-8'));
-
-            return implode('', array_map(static fn (int $unit): string => sprintf('\\u%04x', $unit), (array) $units));
-        }, $value);
+        return $block->isGlobal()
+            ? self::GLOBAL_REASON
+            : 'Knoten nicht lesbar (Attribute kein JSON-Objekt) – im Divi-Builder bearbeiten';
    }

    /** The client message of an error (its JSON "message"), without the JSON around it. */