| @@ -40,19 +40,44 @@ | ||
| 40 | 40 | |
| 41 | 41 | /** |
| 42 | 42 | * Post meta keys that hold builder data, in priority order. |
| 43 | 43 | * |
| 44 | - * Several generations of the same builder are listed on purpose: Oxygen 6 | |
| 45 | - * is Breakdance under the hood (`_breakdance_data`), while earlier Oxygen | |
| 46 | - * releases used `_oxygen_data` or the shortcode-based | |
| 47 | - * `ct_builder_shortcodes`. A site can only have one of them. | |
| 44 | + * Several generations of the same builder are listed on purpose. Oxygen 6 | |
| 45 | + * is Breakdance under the hood and writes the same tree, under its own | |
| 46 | + * prefix: Breakdance keeps it in `_breakdance_data`, Oxygen 6 in | |
| 47 | + * `_oxygen_data` (the key is `__bdox('_meta_prefix') . 'data'`, and the | |
| 48 | + * prefix is `_oxygen_` under Oxygen). Both store it inside a | |
| 49 | + * `tree_json_string` envelope, see unwrap_tree_envelope(). Earlier Oxygen | |
| 50 | + * releases used the shortcode-based `ct_builder_shortcodes` and its JSON | |
| 51 | + * sibling. A site can only have one of them. | |
| 48 | 52 | * |
| 49 | 53 | * @var string[] |
| 50 | 54 | */ |
| 51 | 55 | private const BUILDER_META_KEYS = [ |
| 52 | - '_breakdance_data', // Oxygen 6+ / Breakdance | |
| 53 | - '_oxygen_data', // Oxygen (earlier releases) | |
| 54 | - 'ct_builder_shortcodes', // Oxygen classic | |
| 56 | + '_breakdance_data', // Breakdance | |
| 57 | + '_oxygen_data', // Oxygen 6+ (Breakdance engine, Oxygen prefix) | |
| 58 | + // Oxygen classic. 4.x writes the tree as JSON to `ct_builder_json` | |
| 59 | + // while still keeping `ct_builder_shortcodes`. A post carrying only | |
| 60 | + // the JSON key used to match no key at all and fall through to an | |
| 61 | + // empty `post_content`, which reads as a one-word page (#776). | |
| 62 | + // | |
| 63 | + // Oxygen 4.8.3 then renamed every `ct_*` post meta key to `_ct_*` | |
| 64 | + // (`oxygen_vsb_update_4_8_3()` runs `oxy_prefix_meta_keys()` on | |
| 65 | + // upgrade, and `oxy_get_post_meta()` only reads the prefixed name | |
| 66 | + // from then on). A current Oxygen classic site therefore has only the | |
| 67 | + // underscored keys, which nothing here listed, so every one of its | |
| 68 | + // pages resolved as empty. The prefixed keys come first because they | |
| 69 | + // are what Oxygen itself reads; the bare ones cover a site that has | |
| 70 | + // not run the migration (or reverted it with `?unprefix_meta`). | |
| 71 | + // | |
| 72 | + // These four are not read in this loop: from_oxygen_classic() pairs | |
| 73 | + // each JSON key with its shortcode sibling so the two forms can be | |
| 74 | + // compared. They are listed here because this list is also what the | |
| 75 | + // word-count index watches and what FAQ detection scans. | |
| 76 | + '_ct_builder_json', // Oxygen classic 4.8.3+ (JSON tree) | |
| 77 | + 'ct_builder_json', // Oxygen classic 4.0-4.8.2 (JSON tree) | |
| 78 | + '_ct_builder_shortcodes', // Oxygen classic 4.8.3+ (shortcode tree) | |
| 79 | + 'ct_builder_shortcodes', // Oxygen classic < 4.8.3 (shortcode tree) | |
| 55 | 80 | '_elementor_data', // Elementor |
| 56 | 81 | // Beaver Builder. Published layout first: `_fl_builder_draft` holds |
| 57 | 82 | // unsaved changes and would score content the visitor cannot see. |
| 58 | 83 | // Both are arrays of stdClass nodes, which is why the walker below |
| @@ -97,8 +122,75 @@ | ||
| 97 | 122 | */ |
| 98 | 123 | private const BRICKS_COMPONENTS_OPTION = 'bricks_components'; |
| 99 | 124 | |
| 100 | 125 | /** |
| 126 | + * Bricks' element that renders the post's own `post_content`. | |
| 127 | + * | |
| 128 | + * A Bricks page normally discards `post_content` entirely, which is why | |
| 129 | + * anything left there is invisible. Dropping this element onto the canvas | |
| 130 | + * is the one way an author puts it back on the page, so its presence flips | |
| 131 | + * `post_content` from stale leftovers to content the visitor reads. | |
| 132 | + * | |
| 133 | + * @since 2.3.1 | |
| 134 | + * @var string | |
| 135 | + */ | |
| 136 | + private const BRICKS_POST_CONTENT_ELEMENT = 'post-content'; | |
| 137 | + | |
| 138 | + /** | |
| 139 | + * Bricks' Heading element, and the tag it renders when none is stored. | |
| 140 | + * | |
| 141 | + * Bricks leaves a setting out of storage while it equals its default, so a | |
| 142 | + * Heading left on its default tag is stored with no `tag` at all. Bricks | |
| 143 | + * 2.4.1 renders it as `h3` (`Element_Heading::$tag`, overridable by the | |
| 144 | + * active theme style's `tag`), and the walker, which only wraps text whose | |
| 145 | + * node names a tag, read it as body copy (#908). | |
| 146 | + * | |
| 147 | + * @since 2.15.0 | |
| 148 | + * @var string | |
| 149 | + */ | |
| 150 | + private const BRICKS_HEADING_ELEMENT = 'heading'; | |
| 151 | + | |
| 152 | + /** | |
| 153 | + * Tag a Bricks Heading renders when neither it nor a theme style sets one. | |
| 154 | + * | |
| 155 | + * @since 2.15.0 | |
| 156 | + * @var string | |
| 157 | + */ | |
| 158 | + private const BRICKS_HEADING_DEFAULT_TAG = 'h3'; | |
| 159 | + | |
| 160 | + /** | |
| 161 | + * Tag an Elementor Heading widget renders when `header_size` is not stored. | |
| 162 | + * | |
| 163 | + * Elementor saves `settings.toJSON({ remove: ['default'] })`, so a heading | |
| 164 | + * left on its default size has no `header_size` in `_elementor_data`, and | |
| 165 | + * that default is `h2` (#908). | |
| 166 | + * | |
| 167 | + * @since 2.15.0 | |
| 168 | + * @var string | |
| 169 | + */ | |
| 170 | + private const ELEMENTOR_HEADING_DEFAULT_TAG = 'h2'; | |
| 171 | + | |
| 172 | + /** | |
| 173 | + * Resolved Bricks trees for this request, keyed by post ID. | |
| 174 | + * | |
| 175 | + * Rendering one page asks for the tree about twenty times — every | |
| 176 | + * description, every schema node, the FAQ guard — and resolving it is not | |
| 177 | + * free. `bricks_content_source()` clears `Bricks\Database::$active_templates` | |
| 178 | + * before asking Bricks which content template applies, which defeats | |
| 179 | + * Bricks' own early-return and re-runs its whole template-condition engine; | |
| 180 | + * `expand_bricks_components()` then walks the tree again. Measured on a | |
| 181 | + * Bricks page with no content of its own, that was ten full runs of the | |
| 182 | + * rules engine per request. | |
| 183 | + * | |
| 184 | + * Per-request only, and only ever read back within one page render — a | |
| 185 | + * request that writes Bricks content does not also render it. | |
| 186 | + * | |
| 187 | + * @since 2.3.1 | |
| 188 | + * @var array<int,array<int,mixed>> | |
| 189 | + */ | |
| 190 | + private static array $bricks_trees = []; | |
| 191 | + | |
| 192 | + /** | |
| 101 | 193 | * JSON keys whose values are user-visible text. |
| 102 | 194 | * |
| 103 | 195 | * Builder trees mix content with configuration, so a blind string sweep |
| 104 | 196 | * would count CSS classes and option slugs as words. Matching on the key |
| @@ -109,11 +201,38 @@ | ||
| 109 | 201 | private const CONTENT_KEYS = [ |
| 110 | 202 | 'text', 'title', 'subtitle', 'heading', 'subheading', 'content', |
| 111 | 203 | 'description', 'caption', 'excerpt', 'label', 'value', 'html', |
| 112 | 204 | 'editor', 'quote', 'answer', 'question', 'body', 'button_text', |
| 205 | + // Oxygen classic keeps an element's copy in `options.ct_content` | |
| 206 | + // (headline, text block, rich text, link and button labels). It is the | |
| 207 | + // field Oxygen's own serializer moves between the tags when it writes | |
| 208 | + // shortcodes (`parse_components_tree()`), and the one Relevanssi and | |
| 209 | + // Oxygen's WPML integration read. Missing from this list, the walker | |
| 210 | + // kept only copy that happened to contain markup: a page of plain | |
| 211 | + // headings and paragraphs lost almost all of its words. | |
| 212 | + 'ct_content', | |
| 213 | + // Oxygen's composite elements keep their copy under `options.original` | |
| 214 | + // instead, one key per field. Taken from the list Oxygen itself treats | |
| 215 | + // as text when it serializes (`$options_to_encode`); the numeric price | |
| 216 | + // fields and the progress bar's right-hand percentage are left out. | |
| 217 | + 'testimonial_text', 'testimonial_author', 'testimonial_author_info', | |
| 218 | + 'icon_box_heading', 'icon_box_text', | |
| 219 | + 'pricing_box_package_title', 'pricing_box_package_subtitle', 'pricing_box_content', | |
| 220 | + 'progress_bar_left_text', | |
| 113 | 221 | ]; |
| 114 | 222 | |
| 115 | 223 | /** |
| 224 | + * Oxygen classic's storage generations, as JSON key => shortcode key. | |
| 225 | + * | |
| 226 | + * @since 2.10.0 | |
| 227 | + * @var array<string,string> | |
| 228 | + */ | |
| 229 | + private const OXYGEN_CLASSIC_KEYS = [ | |
| 230 | + '_ct_builder_json' => '_ct_builder_shortcodes', | |
| 231 | + 'ct_builder_json' => 'ct_builder_shortcodes', | |
| 232 | + ]; | |
| 233 | + | |
| 234 | + /** | |
| 116 | 235 | * JSON keys whose values hold a link destination. |
| 117 | 236 | * |
| 118 | 237 | * Builders store a link's destination in a structured field separate from |
| 119 | 238 | * its label, either as a bare URL string or as a `{ url: … }` object. |
| @@ -126,8 +245,71 @@ | ||
| 126 | 245 | 'link', 'url', 'href', 'link_url', 'button_link', 'permalink', 'link_to', |
| 127 | 246 | ]; |
| 128 | 247 | |
| 129 | 248 | /** |
| 249 | + * JSON keys whose values hold an embedded video's source. | |
| 250 | + * | |
| 251 | + * A builder's video widget keeps its destination in a provider-specific | |
| 252 | + * field — Elementor picks `youtube_url`, `vimeo_url`, `dailymotion_url` or | |
| 253 | + * `hosted_url` according to the chosen source type — none of which is a | |
| 254 | + * link field or a content field, so a video on a builder page reached the | |
| 255 | + * analyzers as nothing at all. | |
| 256 | + * | |
| 257 | + * These are deliberately kept out of URL_KEYS. A video is an embed, not an | |
| 258 | + * outbound link: rendering one as `<a href>` would add a spurious external | |
| 259 | + * link to every page carrying a video and skew the link counts. They are | |
| 260 | + * reconstructed as `<iframe>`/`<video>` instead, which the video detector | |
| 261 | + * recognises and the link and image counters ignore. | |
| 262 | + * | |
| 263 | + * @since 2.3.1 | |
| 264 | + * @var string[] | |
| 265 | + */ | |
| 266 | + private const VIDEO_KEYS = [ | |
| 267 | + 'youtube_url', 'vimeo_url', 'dailymotion_url', 'videopress_url', | |
| 268 | + 'hosted_url', 'video_url', 'video_src', 'video_link', | |
| 269 | + ]; | |
| 270 | + | |
| 271 | + /** | |
| 272 | + * File extensions that mean a video source is a file, not a provider page. | |
| 273 | + * | |
| 274 | + * @since 2.3.1 | |
| 275 | + * @var string[] | |
| 276 | + */ | |
| 277 | + private const VIDEO_FILE_EXTENSIONS = ['mp4', 'webm', 'ogv', 'mov', 'm4v']; | |
| 278 | + | |
| 279 | + /** | |
| 280 | + * Source keys to trust for a declared video source type. | |
| 281 | + * | |
| 282 | + * A widget keeps one field per provider and does not clear the others when | |
| 283 | + * the author switches source: an Elementor video moved from YouTube to Self | |
| 284 | + * Hosted still carries the earlier `youtube_url`. Reading whichever key | |
| 285 | + * turns up first then emits the video the author replaced. The widget says | |
| 286 | + * which one it is actually playing, so that is read first and the flat key | |
| 287 | + * sweep is only the fallback for a builder that declares nothing. | |
| 288 | + * | |
| 289 | + * @since 2.3.1 | |
| 290 | + * @var array<string,string[]> | |
| 291 | + */ | |
| 292 | + private const VIDEO_KEYS_BY_TYPE = [ | |
| 293 | + 'youtube' => ['youtube_url'], | |
| 294 | + 'vimeo' => ['vimeo_url'], | |
| 295 | + 'dailymotion' => ['dailymotion_url'], | |
| 296 | + 'videopress' => ['videopress_url'], | |
| 297 | + 'hosted' => ['hosted_url', 'video_url', 'video_src', 'video_link'], | |
| 298 | + 'media' => ['hosted_url', 'video_url', 'video_src', 'video_link'], | |
| 299 | + 'file' => ['hosted_url', 'video_url', 'video_src', 'video_link'], | |
| 300 | + 'self_hosted' => ['hosted_url', 'video_url', 'video_src', 'video_link'], | |
| 301 | + ]; | |
| 302 | + | |
| 303 | + /** | |
| 304 | + * Keys a builder uses to name which video source a widget is playing. | |
| 305 | + * | |
| 306 | + * @since 2.3.1 | |
| 307 | + * @var string[] | |
| 308 | + */ | |
| 309 | + private const VIDEO_TYPE_KEYS = ['video_type', 'videotype', 'video_source', 'source_type']; | |
| 310 | + | |
| 311 | + /** | |
| 130 | 312 | * JSON keys whose values hold an image, as a URL string or `{ url, alt }`. |
| 131 | 313 | * |
| 132 | 314 | * @var string[] |
| 133 | 315 | */ |
| @@ -144,9 +326,13 @@ | ||
| 144 | 326 | * |
| 145 | 327 | * @var string[] |
| 146 | 328 | */ |
| 147 | 329 | private const HEADING_TAG_KEYS = [ |
| 148 | - 'header_size', 'heading_tag', 'html_tag', 'title_tag', 'tag', 'level', 'size', | |
| 330 | + // Lower-cased on both sides of the comparison, so `headingtag` is the | |
| 331 | + // camelCase `headingTag` Bricks uses throughout its own controls and | |
| 332 | + // which ThinkRank's Bricks elements declare. Without it their section | |
| 333 | + // headings counted as body copy and never reached heading structure. | |
| 334 | + 'header_size', 'heading_tag', 'headingtag', 'html_tag', 'title_tag', 'tag', 'level', 'size', | |
| 149 | 335 | ]; |
| 150 | 336 | |
| 151 | 337 | /** |
| 152 | 338 | * Keys whose value is alternative text for a sibling image. |
| @@ -155,8 +341,22 @@ | ||
| 155 | 341 | */ |
| 156 | 342 | private const ALT_KEYS = ['alt', 'alt_text', 'image_alt', 'title']; |
| 157 | 343 | |
| 158 | 344 | /** |
| 345 | + * The global post and every global `setup_postdata()` writes. | |
| 346 | + * | |
| 347 | + * Rendering points them at the post being analyzed, then puts each one | |
| 348 | + * back exactly as it was, unset included (#860). | |
| 349 | + * | |
| 350 | + * @since 2.12.0 | |
| 351 | + * @var string[] | |
| 352 | + */ | |
| 353 | + private const POSTDATA_GLOBALS = [ | |
| 354 | + 'post', 'id', 'authordata', 'currentday', 'currentmonth', | |
| 355 | + 'page', 'pages', 'multipage', 'more', 'numpages', | |
| 356 | + ]; | |
| 357 | + | |
| 358 | + /** | |
| 159 | 359 | * Resolve the content worth analyzing for a post. |
| 160 | 360 | * |
| 161 | 361 | * @param \WP_Post $post Post being analyzed. |
| 162 | 362 | * @return string HTML/text to analyze. |
| @@ -161,12 +361,301 @@ | ||
| 161 | 361 | * @param \WP_Post $post Post being analyzed. |
| 162 | 362 | * @return string HTML/text to analyze. |
| 163 | 363 | */ |
| 164 | 364 | public static function resolve(\WP_Post $post): string { |
| 165 | - return self::resolve_markup((string) $post->post_content, $post); | |
| 365 | + $raw = (string) $post->post_content; | |
| 366 | + | |
| 367 | + // A page built in Gutenberg and then switched to Bricks keeps its old | |
| 368 | + // blocks in `post_content` forever — Bricks never clears them, and | |
| 369 | + // never renders them either. Resolving that first meant the stale draft | |
| 370 | + // beat the tree the visitor actually reads, and it did not stop at the | |
| 371 | + // score: the same string becomes the meta description, og:description, | |
| 372 | + // twitter:description and the schema description. Starting from nothing | |
| 373 | + // sends the resolution straight to Bricks' storage, which is where this | |
| 374 | + // page's words are (#651). | |
| 375 | + // | |
| 376 | + // Only for the stored path. `resolve_markup()` is also called with live | |
| 377 | + // editor content, and the Bricks panel's resolver reads the canvas — | |
| 378 | + // discarding that would replace what the author is typing with the last | |
| 379 | + // save. | |
| 380 | + if (self::bricks_supersedes_post_content((int) $post->ID)) { | |
| 381 | + $raw = ''; | |
| 382 | + } | |
| 383 | + | |
| 384 | + return self::resolve_markup($raw, $post); | |
| 166 | 385 | } |
| 167 | 386 | |
| 168 | 387 | /** |
| 388 | + * Whether Bricks renders this post and throws its `post_content` away. | |
| 389 | + * | |
| 390 | + * True means anything still stored in `post_content` is invisible: it is | |
| 391 | + * not on the page, so it must not be scored, described or published as | |
| 392 | + * structured data. False covers both a post Bricks does not own and a | |
| 393 | + * Bricks page that puts `post_content` back with a Post Content element. | |
| 394 | + * | |
| 395 | + * @since 2.3.1 | |
| 396 | + * | |
| 397 | + * @param int $post_id Post being resolved. | |
| 398 | + * @return bool | |
| 399 | + */ | |
| 400 | + public static function bricks_supersedes_post_content(int $post_id): bool { | |
| 401 | + $tree = self::bricks_tree($post_id); | |
| 402 | + | |
| 403 | + if (empty($tree)) { | |
| 404 | + return false; | |
| 405 | + } | |
| 406 | + | |
| 407 | + foreach ($tree as $element) { | |
| 408 | + if (is_array($element) | |
| 409 | + && self::BRICKS_POST_CONTENT_ELEMENT === ($element['name'] ?? null) | |
| 410 | + ) { | |
| 411 | + return false; | |
| 412 | + } | |
| 413 | + } | |
| 414 | + | |
| 415 | + return !self::bricks_tree_prints_post_content($tree); | |
| 416 | + } | |
| 417 | + | |
| 418 | + /** | |
| 419 | + * Whether a Bricks tree prints the body through a dynamic-data tag. | |
| 420 | + * | |
| 421 | + * The Post Content element is not the only way back onto the page: Bricks' | |
| 422 | + * `{post_content}` tag renders the same thing from inside an ordinary text | |
| 423 | + * element, and a single-post template written that way is a common shape. | |
| 424 | + * Missing it would mean the post's real body is discarded everywhere — | |
| 425 | + * scoring, the meta/og/twitter descriptions, the schema description — for a | |
| 426 | + * page that is displaying it. | |
| 427 | + * | |
| 428 | + * Matched over the encoded tree rather than per setting, because the tag can | |
| 429 | + * sit in any string field of any element and Bricks allows modifiers after | |
| 430 | + * the name (`{post_content:...}`). | |
| 431 | + * | |
| 432 | + * @since 2.3.1 | |
| 433 | + * | |
| 434 | + * @param array $tree Bricks element tree. | |
| 435 | + * @return bool | |
| 436 | + */ | |
| 437 | + private static function bricks_tree_prints_post_content(array $tree): bool { | |
| 438 | + $encoded = wp_json_encode($tree); | |
| 439 | + | |
| 440 | + return is_string($encoded) && false !== stripos($encoded, '{post_content'); | |
| 441 | + } | |
| 442 | + | |
| 443 | + /** | |
| 444 | + * The post's content as the visitor actually receives it. | |
| 445 | + * | |
| 446 | + * `post_content` for everything except a Bricks page that discards it, and | |
| 447 | + * there the Bricks tree's text. Descriptions are derived from a post's body | |
| 448 | + * in half a dozen places; every one of them wants this rather than the raw | |
| 449 | + * column (#651). | |
| 450 | + * | |
| 451 | + * @since 2.3.1 | |
| 452 | + * | |
| 453 | + * @param \WP_Post $post Post being described. | |
| 454 | + * @return string | |
| 455 | + */ | |
| 456 | + public static function visible_content(\WP_Post $post): string { | |
| 457 | + $superseding = self::superseding_content($post); | |
| 458 | + | |
| 459 | + return '' !== $superseding ? $superseding : (string) $post->post_content; | |
| 460 | + } | |
| 461 | + | |
| 462 | + /** | |
| 463 | + * Replacement body text for a post whose `post_content` does not render. | |
| 464 | + * | |
| 465 | + * Empty for every ordinary post, which is what makes this safe to call from | |
| 466 | + * paths that already handle excerpts their own way: they keep that handling | |
| 467 | + * and only a Bricks page is diverted. | |
| 468 | + * | |
| 469 | + * @since 2.3.1 | |
| 470 | + * | |
| 471 | + * @param \WP_Post $post Post being described. | |
| 472 | + * @return string Visible body text, or '' when `post_content` is fine. | |
| 473 | + */ | |
| 474 | + public static function superseding_content(\WP_Post $post): string { | |
| 475 | + if (!self::bricks_supersedes_post_content((int) $post->ID)) { | |
| 476 | + return ''; | |
| 477 | + } | |
| 478 | + | |
| 479 | + $bricks = self::from_bricks((int) $post->ID); | |
| 480 | + | |
| 481 | + return self::is_blank($bricks) ? '' : $bricks; | |
| 482 | + } | |
| 483 | + | |
| 484 | + /** | |
| 485 | + * Body text to derive a description from, when the usual source is wrong. | |
| 486 | + * | |
| 487 | + * A hand-written excerpt is the author's own summary and is correct however | |
| 488 | + * the page is built, so it yields '' here and the caller's normal | |
| 489 | + * `get_the_excerpt()` path keeps it. Only a Bricks page with no excerpt — | |
| 490 | + * where core would derive one from discarded `post_content` — gets diverted. | |
| 491 | + * | |
| 492 | + * @since 2.3.1 | |
| 493 | + * | |
| 494 | + * @param \WP_Post $post Post being described. | |
| 495 | + * @return string Text to summarize, or '' to leave the caller's path alone. | |
| 496 | + */ | |
| 497 | + public static function superseding_excerpt_source(\WP_Post $post): string { | |
| 498 | + if ('' !== trim((string) $post->post_excerpt)) { | |
| 499 | + return ''; | |
| 500 | + } | |
| 501 | + | |
| 502 | + return self::superseding_content($post); | |
| 503 | + } | |
| 504 | + | |
| 505 | + /** | |
| 506 | + * The Bricks element tree that renders for a post. | |
| 507 | + * | |
| 508 | + * Public because what Bricks puts on the page is not only a scoring | |
| 509 | + * question: the schema graph has to know whether a Bricks element already | |
| 510 | + * publishes the page's FAQ before adding one of its own (#649, #650). | |
| 511 | + * | |
| 512 | + * Flat, in Bricks' own storage shape — `expand_bricks_components()` | |
| 513 | + * appends component definitions to the same list rather than nesting them, | |
| 514 | + * so one `foreach` reaches every element. | |
| 515 | + * | |
| 516 | + * @since 2.3.1 | |
| 517 | + * | |
| 518 | + * @param int $post_id Post being resolved. | |
| 519 | + * @return array<int,mixed> Elements, or [] when Bricks renders nothing here. | |
| 520 | + */ | |
| 521 | + /** | |
| 522 | + * The builder meta keys, for callers that need to inspect the raw storage | |
| 523 | + * rather than the text extracted from it. | |
| 524 | + * | |
| 525 | + * The SEO Analyzer reads these to answer "is there a ThinkRank FAQ element | |
| 526 | + * on this post?", which is a question about the stored tree, not about the | |
| 527 | + * words in it (#686). | |
| 528 | + * | |
| 529 | + * @since 2.7.0 | |
| 530 | + * @return string[] | |
| 531 | + */ | |
| 532 | + public static function builder_meta_keys(): array { | |
| 533 | + return self::BUILDER_META_KEYS; | |
| 534 | + } | |
| 535 | + | |
| 536 | + public static function bricks_tree(int $post_id): array { | |
| 537 | + if (array_key_exists($post_id, self::$bricks_trees)) { | |
| 538 | + return self::$bricks_trees[$post_id]; | |
| 539 | + } | |
| 540 | + | |
| 541 | + self::$bricks_trees[$post_id] = self::resolve_bricks_tree($post_id); | |
| 542 | + | |
| 543 | + return self::$bricks_trees[$post_id]; | |
| 544 | + } | |
| 545 | + | |
| 546 | + /** | |
| 547 | + * Discard the resolved-tree memo. Test seam. | |
| 548 | + * | |
| 549 | + * @since 2.3.1 | |
| 550 | + * @return void | |
| 551 | + */ | |
| 552 | + public static function flush_bricks_cache(): void { | |
| 553 | + self::$bricks_trees = []; | |
| 554 | + } | |
| 555 | + | |
| 556 | + /** | |
| 557 | + * Read and resolve a post's Bricks tree, ignoring the memo. | |
| 558 | + * | |
| 559 | + * @since 2.3.1 | |
| 560 | + * | |
| 561 | + * @param int $post_id Post being resolved. | |
| 562 | + * @return array<int,mixed> | |
| 563 | + */ | |
| 564 | + private static function resolve_bricks_tree(int $post_id): array { | |
| 565 | + if (!self::bricks_owns_post($post_id)) { | |
| 566 | + return []; | |
| 567 | + } | |
| 568 | + | |
| 569 | + $source = self::bricks_content_source($post_id); | |
| 570 | + if (!$source) { | |
| 571 | + return []; | |
| 572 | + } | |
| 573 | + | |
| 574 | + $stored = get_post_meta($source, self::bricks_meta_key(), true); | |
| 575 | + | |
| 576 | + if (is_string($stored)) { | |
| 577 | + $stored = '' === trim($stored) ? null : json_decode($stored, true); | |
| 578 | + } | |
| 579 | + | |
| 580 | + if (!is_array($stored) || empty($stored)) { | |
| 581 | + return []; | |
| 582 | + } | |
| 583 | + | |
| 584 | + return self::expand_bricks_components(self::bricks_render_order($stored)); | |
| 585 | + } | |
| 586 | + | |
| 587 | + /** | |
| 588 | + * A flat Bricks element list, in the order Bricks renders it. | |
| 589 | + * | |
| 590 | + * Bricks stores one flat list and links it with `parent` and `children` | |
| 591 | + * ids. `Frontend::render_data()` renders the root elements in list order | |
| 592 | + * and each element's children in the order of its `children` array, so a | |
| 593 | + * child's position in the list says nothing about where it appears on the | |
| 594 | + * page. Walking the list as stored put a section's contents wherever they | |
| 595 | + * happened to be saved (#907). | |
| 596 | + * | |
| 597 | + * Anything the walk does not reach (an orphan, a cycle) keeps its stored | |
| 598 | + * position after the rest, so no copy is dropped. | |
| 599 | + * | |
| 600 | + * @since 2.15.0 | |
| 601 | + * | |
| 602 | + * @param array $elements Flat Bricks element list. | |
| 603 | + * @return array The same elements, in render order. | |
| 604 | + */ | |
| 605 | + private static function bricks_render_order(array $elements): array { | |
| 606 | + $by_id = []; | |
| 607 | + foreach ($elements as $index => $element) { | |
| 608 | + $id = is_array($element) ? ($element['id'] ?? null) : null; | |
| 609 | + if (is_scalar($id) && '' !== (string) $id && !isset($by_id[(string) $id])) { | |
| 610 | + $by_id[(string) $id] = $index; | |
| 611 | + } | |
| 612 | + } | |
| 613 | + | |
| 614 | + if (empty($by_id)) { | |
| 615 | + return $elements; | |
| 616 | + } | |
| 617 | + | |
| 618 | + $ordered = []; | |
| 619 | + $placed = []; | |
| 620 | + | |
| 621 | + $place = static function ($index) use (&$place, &$ordered, &$placed, $elements, $by_id): void { | |
| 622 | + if (isset($placed[$index])) { | |
| 623 | + return; | |
| 624 | + } | |
| 625 | + | |
| 626 | + $placed[$index] = true; | |
| 627 | + $ordered[] = $elements[$index]; | |
| 628 | + | |
| 629 | + $children = is_array($elements[$index]) ? ($elements[$index]['children'] ?? []) : []; | |
| 630 | + if (!is_array($children)) { | |
| 631 | + return; | |
| 632 | + } | |
| 633 | + | |
| 634 | + foreach ($children as $child_id) { | |
| 635 | + if (is_scalar($child_id) && isset($by_id[(string) $child_id])) { | |
| 636 | + $place($by_id[(string) $child_id]); | |
| 637 | + } | |
| 638 | + } | |
| 639 | + }; | |
| 640 | + | |
| 641 | + foreach ($elements as $index => $element) { | |
| 642 | + $parent = is_array($element) ? ($element['parent'] ?? null) : null; | |
| 643 | + if (empty($parent) || !is_scalar($parent) || !isset($by_id[(string) $parent])) { | |
| 644 | + $place($index); | |
| 645 | + } | |
| 646 | + } | |
| 647 | + | |
| 648 | + foreach ($elements as $index => $element) { | |
| 649 | + if (!isset($placed[$index])) { | |
| 650 | + $ordered[] = $element; | |
| 651 | + } | |
| 652 | + } | |
| 653 | + | |
| 654 | + return $ordered; | |
| 655 | + } | |
| 656 | + | |
| 657 | + /** | |
| 169 | 658 | * Resolve an arbitrary chunk of editor markup for the given post. |
| 170 | 659 | * |
| 171 | 660 | * The editor sends its live content to the scorer so an author sees their |
| 172 | 661 | * unsaved edits reflected. On a builder page that live string is the raw |
| @@ -187,9 +676,9 @@ | ||
| 187 | 676 | * @param \WP_Post $post Post the markup belongs to. |
| 188 | 677 | * @return string Content to analyze. |
| 189 | 678 | */ |
| 190 | 679 | public static function resolve_markup(string $raw, \WP_Post $post): string { |
| 191 | - $content = self::render_post_content($raw); | |
| 680 | + $content = self::render_post_content($raw, $post); | |
| 192 | 681 | |
| 193 | 682 | // Block markup that renders to nothing usually means the builder that |
| 194 | 683 | // owns those blocks did not register them in this context — Divi 5 |
| 195 | 684 | // loads its module library lazily per-request, so in CLI, REST, admin |
| @@ -237,19 +726,45 @@ | ||
| 237 | 726 | * |
| 238 | 727 | * Best-effort: a third-party block that fatals must not take the whole |
| 239 | 728 | * score down with it. |
| 240 | 729 | * |
| 241 | - * @param string $raw Raw post content. | |
| 730 | + * Runs as the post's own context, as it would on the front end. Admin and | |
| 731 | + * REST requests have no current post, so a shortcode reading | |
| 732 | + * `get_the_ID()` got nothing, and one looping a related-posts query left | |
| 733 | + * the global post on the last of them: its `wp_reset_postdata()` goes back | |
| 734 | + * to the main query's post, and there is none. On the Classic Editor this | |
| 735 | + * runs after the form prints its hidden `post_ID` and before the title and | |
| 736 | + * editor, which then showed the related post, and Update saved it over the | |
| 737 | + * original (#860). | |
| 738 | + * | |
| 739 | + * @param string $raw Raw post content. | |
| 740 | + * @param \WP_Post $post Post the content belongs to. | |
| 242 | 741 | * @return string Rendered content. |
| 243 | 742 | */ |
| 244 | - private static function render_post_content(string $raw): string { | |
| 743 | + private static function render_post_content(string $raw, \WP_Post $post): string { | |
| 245 | 744 | if ('' === trim($raw)) { |
| 246 | 745 | return ''; |
| 247 | 746 | } |
| 248 | 747 | |
| 249 | - $content = $raw; | |
| 748 | + $content = $raw; | |
| 749 | + $previous = self::snapshot_post_globals(); | |
| 250 | 750 | |
| 251 | 751 | try { |
| 752 | + // phpcs:ignore WordPress.WP.GlobalVariablesOverride.Prohibited -- Made current for the render, restored in finally. | |
| 753 | + $GLOBALS['post'] = $post; | |
| 754 | + | |
| 755 | + // Fires `the_post`, which these paths never fired before: admin, | |
| 756 | + // REST and cron analysis had no current post at all. That is the | |
| 757 | + // same signal the front-end loop sends and it is what makes | |
| 758 | + // `get_the_ID()` work inside a shortcode, but it is a new call on a | |
| 759 | + // path that runs in bulk — the word-count index resolves every post | |
| 760 | + // it visits — so a theme that counts views on `the_post` will count | |
| 761 | + // them during indexing. Accepted deliberately: without it a | |
| 762 | + // shortcode cannot resolve its own post, which is the bug (#860). | |
| 763 | + if (function_exists('setup_postdata')) { | |
| 764 | + setup_postdata($post); | |
| 765 | + } | |
| 766 | + | |
| 252 | 767 | if (function_exists('has_blocks') && function_exists('do_blocks') && has_blocks($raw)) { |
| 253 | 768 | $content = do_blocks($raw); |
| 254 | 769 | } |
| 255 | 770 | |
| @@ -258,8 +773,10 @@ | ||
| 258 | 773 | $content = do_shortcode($content); |
| 259 | 774 | } |
| 260 | 775 | } catch (\Throwable $e) { |
| 261 | 776 | return $raw; |
| 777 | + } finally { | |
| 778 | + self::restore_post_globals($previous); | |
| 262 | 779 | } |
| 263 | 780 | |
| 264 | 781 | return self::is_blank($content) ? $raw : $content; |
| 265 | 782 | } |
| @@ -264,8 +781,49 @@ | ||
| 264 | 781 | return self::is_blank($content) ? $raw : $content; |
| 265 | 782 | } |
| 266 | 783 | |
| 267 | 784 | /** |
| 785 | + * The post globals as they are now; a global that is unset has no key. | |
| 786 | + * | |
| 787 | + * @since 2.12.0 | |
| 788 | + * | |
| 789 | + * @return array<string,mixed> | |
| 790 | + */ | |
| 791 | + private static function snapshot_post_globals(): array { | |
| 792 | + $snapshot = []; | |
| 793 | + | |
| 794 | + foreach (self::POSTDATA_GLOBALS as $name) { | |
| 795 | + if (array_key_exists($name, $GLOBALS)) { | |
| 796 | + $snapshot[$name] = $GLOBALS[$name]; | |
| 797 | + } | |
| 798 | + } | |
| 799 | + | |
| 800 | + return $snapshot; | |
| 801 | + } | |
| 802 | + | |
| 803 | + /** | |
| 804 | + * Put the post globals back as snapshot_post_globals() found them. | |
| 805 | + * | |
| 806 | + * Assigned directly rather than through `setup_postdata()`: there may have | |
| 807 | + * been no post to set up, and re-running it would fire `the_post` again. | |
| 808 | + * | |
| 809 | + * @since 2.12.0 | |
| 810 | + * | |
| 811 | + * @param array<string,mixed> $snapshot From snapshot_post_globals(). | |
| 812 | + * @return void | |
| 813 | + */ | |
| 814 | + private static function restore_post_globals(array $snapshot): void { | |
| 815 | + foreach (self::POSTDATA_GLOBALS as $name) { | |
| 816 | + if (array_key_exists($name, $snapshot)) { | |
| 817 | + // phpcs:ignore WordPress.NamingConventions.PrefixAllGlobals.NonPrefixedVariableFound -- Core's own globals, put back as they were. | |
| 818 | + $GLOBALS[$name] = $snapshot[$name]; | |
| 819 | + } else { | |
| 820 | + unset($GLOBALS[$name]); | |
| 821 | + } | |
| 822 | + } | |
| 823 | + } | |
| 824 | + | |
| 825 | + /** | |
| 268 | 826 | * Extract text from the attributes of parsed blocks. |
| 269 | 827 | * |
| 270 | 828 | * @param string $raw Raw post content containing block markup. |
| 271 | 829 | * @return string Collected text, or '' when nothing was found. |
| @@ -319,31 +877,16 @@ | ||
| 319 | 877 | * @param int $post_id Post being resolved. |
| 320 | 878 | * @return string Extracted text, or '' when Bricks has nothing for it. |
| 321 | 879 | */ |
| 322 | 880 | private static function from_bricks(int $post_id): string { |
| 323 | - if (!self::bricks_owns_post($post_id)) { | |
| 324 | - return ''; | |
| 325 | - } | |
| 881 | + $tree = self::bricks_tree($post_id); | |
| 326 | 882 | |
| 327 | - $source = self::bricks_content_source($post_id); | |
| 328 | - if (!$source) { | |
| 883 | + if (empty($tree)) { | |
| 329 | 884 | return ''; |
| 330 | 885 | } |
| 331 | 886 | |
| 332 | - $stored = get_post_meta($source, self::bricks_meta_key(), true); | |
| 333 | - | |
| 334 | - if (is_string($stored)) { | |
| 335 | - $stored = '' === trim($stored) ? null : json_decode($stored, true); | |
| 336 | - } | |
| 337 | - | |
| 338 | - if (!is_array($stored) || empty($stored)) { | |
| 339 | - return ''; | |
| 340 | - } | |
| 341 | - | |
| 342 | 887 | return self::strip_bricks_dynamic_tags( |
| 343 | - self::text_from_tree( | |
| 344 | - self::without_bricks_element_labels(self::expand_bricks_components($stored)) | |
| 345 | - ) | |
| 888 | + self::text_from_tree(self::with_bricks_heading_tags(self::without_bricks_element_labels($tree))) | |
| 346 | 889 | ); |
| 347 | 890 | } |
| 348 | 891 | |
| 349 | 892 | /** |
| @@ -541,9 +1084,9 @@ | ||
| 541 | 1084 | continue; |
| 542 | 1085 | } |
| 543 | 1086 | |
| 544 | 1087 | if (isset($component['id']) && $component['id'] === $cid && !empty($component['elements'])) { |
| 545 | - return is_array($component['elements']) ? $component['elements'] : []; | |
| 1088 | + return is_array($component['elements']) ? self::bricks_render_order($component['elements']) : []; | |
| 546 | 1089 | } |
| 547 | 1090 | } |
| 548 | 1091 | |
| 549 | 1092 | return []; |
| @@ -576,8 +1119,116 @@ | ||
| 576 | 1119 | return $tree; |
| 577 | 1120 | } |
| 578 | 1121 | |
| 579 | 1122 | /** |
| 1123 | + * Give each Bricks Heading the tag it renders with when none is stored. | |
| 1124 | + * | |
| 1125 | + * Applied on the Bricks path only. Most other builders' text nodes carry no | |
| 1126 | + * tag because they are not headings, so a generic "text without a tag is a | |
| 1127 | + * heading" rule in heading_tag_from() would turn every paragraph into one. | |
| 1128 | + * | |
| 1129 | + * A `tag` of `custom` is left alone: the element then renders its | |
| 1130 | + * `customTag`, which is not necessarily a heading. | |
| 1131 | + * | |
| 1132 | + * @since 2.15.0 | |
| 1133 | + * | |
| 1134 | + * @param array $tree Bricks content area. | |
| 1135 | + * @return array Tree with each untagged Heading's default tag filled in. | |
| 1136 | + */ | |
| 1137 | + private static function with_bricks_heading_tags(array $tree): array { | |
| 1138 | + $default = null; | |
| 1139 | + | |
| 1140 | + foreach ($tree as $index => $element) { | |
| 1141 | + if (!is_array($element) || self::BRICKS_HEADING_ELEMENT !== ($element['name'] ?? null)) { | |
| 1142 | + continue; | |
| 1143 | + } | |
| 1144 | + | |
| 1145 | + $settings = $element['settings'] ?? []; | |
| 1146 | + if (!is_array($settings)) { | |
| 1147 | + continue; | |
| 1148 | + } | |
| 1149 | + | |
| 1150 | + $tag = $settings['tag'] ?? ''; | |
| 1151 | + if (is_string($tag) && '' !== trim($tag)) { | |
| 1152 | + continue; | |
| 1153 | + } | |
| 1154 | + | |
| 1155 | + if (null === $default) { | |
| 1156 | + $default = self::bricks_default_heading_tag(); | |
| 1157 | + } | |
| 1158 | + | |
| 1159 | + $settings['tag'] = $default; | |
| 1160 | + $tree[$index]['settings'] = $settings; | |
| 1161 | + } | |
| 1162 | + | |
| 1163 | + return $tree; | |
| 1164 | + } | |
| 1165 | + | |
| 1166 | + /** | |
| 1167 | + * The tag Bricks gives a Heading that does not set one. | |
| 1168 | + * | |
| 1169 | + * The active theme style can change it. Bricks only loads theme styles for | |
| 1170 | + * a front-end render, so in admin, REST and CLI requests this is the | |
| 1171 | + * element's own default. | |
| 1172 | + * | |
| 1173 | + * @since 2.15.0 | |
| 1174 | + * | |
| 1175 | + * @return string Heading tag, h1 to h6. | |
| 1176 | + */ | |
| 1177 | + private static function bricks_default_heading_tag(): string { | |
| 1178 | + if (class_exists('\\Bricks\\Theme_Styles') | |
| 1179 | + && method_exists('\\Bricks\\Theme_Styles', 'get_setting_by_key') | |
| 1180 | + ) { | |
| 1181 | + try { | |
| 1182 | + $styled = \Bricks\Theme_Styles::get_setting_by_key(self::BRICKS_HEADING_ELEMENT, 'tag'); | |
| 1183 | + } catch (\Throwable $e) { | |
| 1184 | + $styled = null; | |
| 1185 | + } | |
| 1186 | + | |
| 1187 | + if (is_string($styled) && preg_match('/^h[1-6]$/i', trim($styled))) { | |
| 1188 | + return strtolower(trim($styled)); | |
| 1189 | + } | |
| 1190 | + } | |
| 1191 | + | |
| 1192 | + return self::BRICKS_HEADING_DEFAULT_TAG; | |
| 1193 | + } | |
| 1194 | + | |
| 1195 | + /** | |
| 1196 | + * Give each Elementor Heading widget its default `header_size` if unstored. | |
| 1197 | + * | |
| 1198 | + * @since 2.15.0 | |
| 1199 | + * | |
| 1200 | + * @param array $elements Decoded `_elementor_data`. | |
| 1201 | + * @return array The same tree, with untagged Heading widgets tagged. | |
| 1202 | + */ | |
| 1203 | + private static function with_elementor_heading_tags(array $elements): array { | |
| 1204 | + foreach ($elements as $index => $element) { | |
| 1205 | + if (!is_array($element)) { | |
| 1206 | + continue; | |
| 1207 | + } | |
| 1208 | + | |
| 1209 | + if ('heading' === ($element['widgetType'] ?? null)) { | |
| 1210 | + $settings = $element['settings'] ?? []; | |
| 1211 | + if (is_array($settings)) { | |
| 1212 | + $size = $settings['header_size'] ?? ''; | |
| 1213 | + if (!is_string($size) || '' === trim($size)) { | |
| 1214 | + $settings['header_size'] = self::ELEMENTOR_HEADING_DEFAULT_TAG; | |
| 1215 | + $element['settings'] = $settings; | |
| 1216 | + } | |
| 1217 | + } | |
| 1218 | + } | |
| 1219 | + | |
| 1220 | + if (!empty($element['elements']) && is_array($element['elements'])) { | |
| 1221 | + $element['elements'] = self::with_elementor_heading_tags($element['elements']); | |
| 1222 | + } | |
| 1223 | + | |
| 1224 | + $elements[$index] = $element; | |
| 1225 | + } | |
| 1226 | + | |
| 1227 | + return $elements; | |
| 1228 | + } | |
| 1229 | + | |
| 1230 | + /** | |
| 580 | 1231 | * Remove Bricks dynamic-data tags from extracted text. |
| 581 | 1232 | * |
| 582 | 1233 | * Bricks stores `{post_title}`, `{post_meta:price}`, `{echo:my_fn}` and the |
| 583 | 1234 | * like verbatim and resolves them when it renders. Extraction reads the |
| @@ -668,16 +1319,32 @@ | ||
| 668 | 1319 | return $bricks; |
| 669 | 1320 | } |
| 670 | 1321 | |
| 671 | 1322 | foreach (self::BUILDER_META_KEYS as $key) { |
| 1323 | + if (self::is_oxygen_classic_key($key)) { | |
| 1324 | + // Resolved as a pair, once, at the first of its keys. | |
| 1325 | + if ('_ct_builder_json' !== $key) { | |
| 1326 | + continue; | |
| 1327 | + } | |
| 1328 | + | |
| 1329 | + $oxygen = self::from_oxygen_classic($post_id); | |
| 1330 | + if (!self::is_blank($oxygen)) { | |
| 1331 | + return $oxygen; | |
| 1332 | + } | |
| 1333 | + continue; | |
| 1334 | + } | |
| 1335 | + | |
| 672 | 1336 | $stored = get_post_meta($post_id, $key, true); |
| 673 | 1337 | |
| 674 | 1338 | if (is_string($stored) && '' !== trim($stored)) { |
| 675 | 1339 | $decoded = json_decode($stored, true); |
| 1340 | + if ('_elementor_data' === $key && is_array($decoded)) { | |
| 1341 | + $decoded = self::with_elementor_heading_tags($decoded); | |
| 1342 | + } | |
| 676 | 1343 | |
| 677 | 1344 | // JSON node tree (Breakdance/Oxygen 6, Elementor). |
| 678 | 1345 | if (is_array($decoded)) { |
| 679 | - $text = self::text_from_tree($decoded); | |
| 1346 | + $text = self::text_from_tree(self::unwrap_tree_envelope($decoded)); | |
| 680 | 1347 | if (!self::is_blank($text)) { |
| 681 | 1348 | return $text; |
| 682 | 1349 | } |
| 683 | 1350 | continue; |
| @@ -682,20 +1349,8 @@ | ||
| 682 | 1349 | } |
| 683 | 1350 | continue; |
| 684 | 1351 | } |
| 685 | 1352 | |
| 686 | - // Shortcode tree (Oxygen classic). | |
| 687 | - if (strpos($stored, '[') !== false && function_exists('do_shortcode')) { | |
| 688 | - try { | |
| 689 | - $rendered = do_shortcode($stored); | |
| 690 | - } catch (\Throwable $e) { | |
| 691 | - $rendered = $stored; | |
| 692 | - } | |
| 693 | - if (!self::is_blank($rendered)) { | |
| 694 | - return $rendered; | |
| 695 | - } | |
| 696 | - } | |
| 697 | - | |
| 698 | 1353 | continue; |
| 699 | 1354 | } |
| 700 | 1355 | |
| 701 | 1356 | // Some builders store an already-decoded tree — an array for most, |
| @@ -701,8 +1356,9 @@ | ||
| 701 | 1356 | // Some builders store an already-decoded tree — an array for most, |
| 702 | 1357 | // an array of objects for Beaver Builder (#449). |
| 703 | 1358 | $tree = self::as_children($stored); |
| 704 | 1359 | if (null !== $tree) { |
| 1360 | + $tree = self::unwrap_tree_envelope($tree); | |
| 705 | 1361 | $text = self::text_from_tree($tree); |
| 706 | 1362 | if (!self::is_blank($text)) { |
| 707 | 1363 | return $text; |
| 708 | 1364 | } |
| @@ -712,8 +1368,298 @@ | ||
| 712 | 1368 | return ''; |
| 713 | 1369 | } |
| 714 | 1370 | |
| 715 | 1371 | /** |
| 1372 | + * The node tree inside a Breakdance / Oxygen 6 storage envelope. | |
| 1373 | + * | |
| 1374 | + * Neither builder stores its tree directly. The meta value is | |
| 1375 | + * `{"tree_json_string": "<the tree, JSON-encoded again>"}`, so one | |
| 1376 | + * json_decode() yields the envelope, not the tree. Walked as a tree, the | |
| 1377 | + * envelope is a single string leaf: kept whole as "content" when any | |
| 1378 | + * element held rich text (the encoded JSON then reached scoring, the | |
| 1379 | + * get-post-content ability and Markdown for AI), dropped when none did, | |
| 1380 | + * leaving the page empty (#905). | |
| 1381 | + * | |
| 1382 | + * An envelope whose inner string does not decode returns an empty tree, | |
| 1383 | + * never the string: handing the raw JSON back to the walker would bring | |
| 1384 | + * the JSON-as-content failure back on corrupt data. Anything that is not | |
| 1385 | + * an envelope is returned unchanged, so a bare tree still resolves. | |
| 1386 | + * | |
| 1387 | + * @since 2.15.0 | |
| 1388 | + * | |
| 1389 | + * @param array $decoded Decoded meta value. | |
| 1390 | + * @return array The node tree. | |
| 1391 | + */ | |
| 1392 | + private static function unwrap_tree_envelope(array $decoded): array { | |
| 1393 | + if (!array_key_exists('tree_json_string', $decoded)) { | |
| 1394 | + return $decoded; | |
| 1395 | + } | |
| 1396 | + | |
| 1397 | + $inner = is_string($decoded['tree_json_string']) | |
| 1398 | + ? json_decode($decoded['tree_json_string'], true) | |
| 1399 | + : $decoded['tree_json_string']; | |
| 1400 | + | |
| 1401 | + if (is_array($inner)) { | |
| 1402 | + return $inner; | |
| 1403 | + } | |
| 1404 | + | |
| 1405 | + // Re-serialised envelopes can carry the tree as an object. | |
| 1406 | + $inner = self::as_children($inner); | |
| 1407 | + | |
| 1408 | + return null !== $inner ? $inner : []; | |
| 1409 | + } | |
| 1410 | + | |
| 1411 | + /** | |
| 1412 | + * Whether a meta key is one of Oxygen classic's storage keys. | |
| 1413 | + * | |
| 1414 | + * @since 2.10.0 | |
| 1415 | + * | |
| 1416 | + * @param string $key Meta key. | |
| 1417 | + * @return bool | |
| 1418 | + */ | |
| 1419 | + private static function is_oxygen_classic_key(string $key): bool { | |
| 1420 | + return isset(self::OXYGEN_CLASSIC_KEYS[$key]) || in_array($key, self::OXYGEN_CLASSIC_KEYS, true); | |
| 1421 | + } | |
| 1422 | + | |
| 1423 | + /** | |
| 1424 | + * Text of an Oxygen classic page, from whichever stored form holds more. | |
| 1425 | + * | |
| 1426 | + * Oxygen 4.x keeps the same tree twice: as JSON, and as the shortcodes it | |
| 1427 | + * used before 4.0. The JSON is preferred because it carries copy the | |
| 1428 | + * shortcode form hides (a composite element's text is base64-encoded | |
| 1429 | + * inside `ct_options`, which is configuration and stripped). It is not | |
| 1430 | + * trusted blindly, though. Reading `ct_builder_json` first once meant a | |
| 1431 | + * key missing from CONTENT_KEYS silently threw the page away while the | |
| 1432 | + * shortcode copy sat unread next to it, because a non-empty JSON result | |
| 1433 | + * stopped the search. Comparing the two means the next such gap costs | |
| 1434 | + * nothing: the richer form wins. | |
| 1435 | + * | |
| 1436 | + * A generation is only read as a pair. The prefixed keys are what Oxygen | |
| 1437 | + * 4.8.3+ reads, so an unprefixed leftover next to them is stale. | |
| 1438 | + * | |
| 1439 | + * @since 2.10.0 | |
| 1440 | + * | |
| 1441 | + * @param int $post_id Post ID. | |
| 1442 | + * @return string Extracted text, or '' when Oxygen classic stored nothing. | |
| 1443 | + */ | |
| 1444 | + private static function from_oxygen_classic(int $post_id): string { | |
| 1445 | + foreach (self::OXYGEN_CLASSIC_KEYS as $json_key => $shortcode_key) { | |
| 1446 | + $json = get_post_meta($post_id, $json_key, true); | |
| 1447 | + $shortcodes = get_post_meta($post_id, $shortcode_key, true); | |
| 1448 | + | |
| 1449 | + $from_json = ''; | |
| 1450 | + if (is_string($json) && '' !== trim($json)) { | |
| 1451 | + $decoded = json_decode($json, true); | |
| 1452 | + if (is_array($decoded)) { | |
| 1453 | + // `[oxygen data="..."]` is a dynamic-data placeholder | |
| 1454 | + // Oxygen fills at render time. The shortcode path drops it | |
| 1455 | + // with every other tag, so it goes here too or the two | |
| 1456 | + // forms would disagree on the same page. | |
| 1457 | + $from_json = (string) preg_replace( | |
| 1458 | + '/\[oxygen\b[^\]]*\]/i', | |
| 1459 | + ' ', | |
| 1460 | + self::text_from_tree($decoded) | |
| 1461 | + ); | |
| 1462 | + } | |
| 1463 | + } | |
| 1464 | + | |
| 1465 | + $from_shortcodes = ''; | |
| 1466 | + if (is_string($shortcodes) && strpos($shortcodes, '[') !== false) { | |
| 1467 | + $from_shortcodes = self::text_from_shortcodes($shortcodes); | |
| 1468 | + } | |
| 1469 | + | |
| 1470 | + if (self::is_blank($from_json) && self::is_blank($from_shortcodes)) { | |
| 1471 | + continue; | |
| 1472 | + } | |
| 1473 | + | |
| 1474 | + return self::visible_word_count($from_json) >= self::visible_word_count($from_shortcodes) | |
| 1475 | + ? $from_json | |
| 1476 | + : $from_shortcodes; | |
| 1477 | + } | |
| 1478 | + | |
| 1479 | + return ''; | |
| 1480 | + } | |
| 1481 | + | |
| 1482 | + /** | |
| 1483 | + * Rough count of the words a visitor would read in extracted text. | |
| 1484 | + * | |
| 1485 | + * Only used to compare two extractions of the same page, so it needs to | |
| 1486 | + * be consistent rather than locale-exact. | |
| 1487 | + * | |
| 1488 | + * @since 2.10.0 | |
| 1489 | + * | |
| 1490 | + * @param string $text Extracted text or markup. | |
| 1491 | + * @return int | |
| 1492 | + */ | |
| 1493 | + private static function visible_word_count(string $text): int { | |
| 1494 | + $plain = trim((string) preg_replace('/\s+/u', ' ', wp_strip_all_tags($text))); | |
| 1495 | + | |
| 1496 | + return '' === $plain ? 0 : count(explode(' ', $plain)); | |
| 1497 | + } | |
| 1498 | + | |
| 1499 | + /** | |
| 1500 | + * Shortcode attributes that carry copy a visitor reads. | |
| 1501 | + * | |
| 1502 | + * An allow-list, not a deny-list. Oxygen Classic tags carry far more | |
| 1503 | + * attributes than they do copy — `id`, `class`, `selector`, `url`, | |
| 1504 | + * `ct_options` and friends — and a deny-list silently admits every | |
| 1505 | + * attribute a future builder release invents, which is how markup ends up | |
| 1506 | + * being counted as prose. | |
| 1507 | + * | |
| 1508 | + * @var string[] | |
| 1509 | + */ | |
| 1510 | + private const SHORTCODE_TEXT_ATTRIBUTES = [ | |
| 1511 | + 'text', | |
| 1512 | + 'content', | |
| 1513 | + 'heading', | |
| 1514 | + 'title', | |
| 1515 | + 'subtitle', | |
| 1516 | + 'label', | |
| 1517 | + 'caption', | |
| 1518 | + 'description', | |
| 1519 | + 'alt', | |
| 1520 | + 'button_text', | |
| 1521 | + 'link_text', | |
| 1522 | + ]; | |
| 1523 | + | |
| 1524 | + /** | |
| 1525 | + * Extract readable text from a shortcode tree, without rendering it. | |
| 1526 | + * | |
| 1527 | + * Oxygen Classic is the only builder whose storage is shortcodes rather | |
| 1528 | + * than JSON, and the previous implementation handed the string to | |
| 1529 | + * `do_shortcode()`. That silently depends on Oxygen having registered its | |
| 1530 | + * `ct_*` handlers in the current request — which it has on a front-end | |
| 1531 | + * view, and has not during bulk analysis, the post-list column, cron or | |
| 1532 | + * REST/MCP. With no handlers registered `do_shortcode()` returns its input | |
| 1533 | + * unchanged, so the raw shortcode source was scored as if it were the | |
| 1534 | + * page's prose: `[ct_section`, `id="section-1"` and the rest counted toward | |
| 1535 | + * the word count, while the actual copy sitting in `text="..."` attributes | |
| 1536 | + * was never counted at all (#776). | |
| 1537 | + * | |
| 1538 | + * `strip_shortcodes()` is no help either — it also only knows registered | |
| 1539 | + * shortcodes, so it leaves the same text untouched. | |
| 1540 | + * | |
| 1541 | + * Reading the stored tree directly is what every other builder here already | |
| 1542 | + * does, and it matches the class's stated design: no render engine, no | |
| 1543 | + * dependency on load order, safe during a bulk run. | |
| 1544 | + * | |
| 1545 | + * Parsing unconditionally, rather than rendering when Oxygen happens to be | |
| 1546 | + * loaded and parsing otherwise, is deliberate. It makes the extracted text | |
| 1547 | + * the same in every context, so the score in the editor matches the score | |
| 1548 | + * from a bulk run or from MCP. The old code produced whichever of the two | |
| 1549 | + * the request happened to allow, which is why the same post could report | |
| 1550 | + * two different word counts depending on how it was asked. | |
| 1551 | + * | |
| 1552 | + * The trade-off is that rendered output (resolved images, links, anything | |
| 1553 | + * Oxygen pulls in from a reusable part) is no longer reflected here. For | |
| 1554 | + * what this text feeds — word count, content scoring, meta-description | |
| 1555 | + * fallbacks and schema text — that markup was never the point, and counting | |
| 1556 | + * it only when the builder happened to be booted was the bug. | |
| 1557 | + * | |
| 1558 | + * @since 2.10.0 | |
| 1559 | + * | |
| 1560 | + * @param string $stored Raw shortcode source. | |
| 1561 | + * @return string Extracted text. | |
| 1562 | + */ | |
| 1563 | + private static function text_from_shortcodes(string $stored): string { | |
| 1564 | + // Oxygen stores each element's settings as a JSON blob in `ct_options`. | |
| 1565 | + // It is configuration, never copy, and it contains braces and brackets | |
| 1566 | + // that would otherwise confuse the tag scan below, so it goes first. | |
| 1567 | + // | |
| 1568 | + // The blob is matched as a balanced JSON object, not as "up to the | |
| 1569 | + // next quote". Oxygen wraps it in single quotes but does not escape | |
| 1570 | + // an apostrophe inside it (`"nicename":"Bob's Plumbing"`), so the | |
| 1571 | + // quote-to-quote match stopped mid-value and the rest of the blob, | |
| 1572 | + // `s Plumbing"}'` and all, was left in the tag and leaked into the | |
| 1573 | + // text. Strings inside the object are skipped whole, so neither a quote | |
| 1574 | + // nor a brace inside a value can end the match early. | |
| 1575 | + $source = (string) preg_replace( | |
| 1576 | + '/\sct_options\s*=\s*\'(?<obj>\{(?:[^{}"]++|"(?:[^"\\\\]|\\\\.)*+"|(?&obj))*+\})\'/s', | |
| 1577 | + '', | |
| 1578 | + $stored | |
| 1579 | + ); | |
| 1580 | + | |
| 1581 | + // Anything not shaped like Oxygen's JSON blob keeps the old, | |
| 1582 | + // quote-delimited strip. | |
| 1583 | + $source = (string) preg_replace( | |
| 1584 | + '/\sct_options\s*=\s*(["\']).*?\1/s', | |
| 1585 | + '', | |
| 1586 | + $source | |
| 1587 | + ); | |
| 1588 | + | |
| 1589 | + $attributes = implode('|', array_map( | |
| 1590 | + static fn(string $name): string => preg_quote($name, '/'), | |
| 1591 | + self::SHORTCODE_TEXT_ATTRIBUTES | |
| 1592 | + )); | |
| 1593 | + | |
| 1594 | + // Replace each shortcode tag with whatever readable copy its attributes | |
| 1595 | + // carry. Text between tags is left exactly where it is, so the result | |
| 1596 | + // keeps the page's reading order rather than hoisting all the headings | |
| 1597 | + // to the front. | |
| 1598 | + // The attribute blob is matched quote-aware rather than as "anything up | |
| 1599 | + // to the first `]`". Oxygen copy contains brackets often enough to | |
| 1600 | + // matter — "Best tools [2026]", "[Updated] our policy" — and a naive | |
| 1601 | + // scan ends the tag inside the `text` attribute, dropping the copy | |
| 1602 | + // before the bracket and leaking the stray `"]` after it into the | |
| 1603 | + // prose. Which is this bug's own failure mode: the wrong text scored. | |
| 1604 | + // | |
| 1605 | + // A tag name must start with a letter or underscore. `[2026]` is not a | |
| 1606 | + // shortcode anyone can register, and scanning it as one dropped the | |
| 1607 | + // year out of "Best tools [2026]". | |
| 1608 | + $text = (string) preg_replace_callback( | |
| 1609 | + '/\[\/?[a-zA-Z_][a-zA-Z0-9_-]*((?:[^\]"\']|"[^"]*"|\'[^\']*\')*)\]/', | |
| 1610 | + static function (array $matches) use ($attributes): string { | |
| 1611 | + if ('' === trim($matches[1])) { | |
| 1612 | + return ' '; | |
| 1613 | + } | |
| 1614 | + | |
| 1615 | + if (!preg_match_all( | |
| 1616 | + '/\b(' . $attributes . ')\s*=\s*(["\'])(.*?)\2/s', | |
| 1617 | + $matches[1], | |
| 1618 | + $found, | |
| 1619 | + PREG_SET_ORDER | |
| 1620 | + )) { | |
| 1621 | + return ' '; | |
| 1622 | + } | |
| 1623 | + | |
| 1624 | + $parts = []; | |
| 1625 | + foreach ($found as $attribute) { | |
| 1626 | + $value = trim($attribute[3]); | |
| 1627 | + | |
| 1628 | + // An attribute holding markup or a JSON fragment is | |
| 1629 | + // configuration that happens to share a name with a copy | |
| 1630 | + // field, not something a visitor reads. | |
| 1631 | + if ('' === $value || preg_match('/^[\[{<]/', $value)) { | |
| 1632 | + continue; | |
| 1633 | + } | |
| 1634 | + | |
| 1635 | + $parts[] = $value; | |
| 1636 | + } | |
| 1637 | + | |
| 1638 | + return empty($parts) ? ' ' : ' ' . implode(' ', $parts) . ' '; | |
| 1639 | + }, | |
| 1640 | + $source | |
| 1641 | + ); | |
| 1642 | + | |
| 1643 | + // Oxygen escapes square brackets in an element's copy before writing | |
| 1644 | + // it between the tags, so that "Best tools [2026]" cannot be mistaken | |
| 1645 | + // for a shortcode (`oxygen_vsb_filter_shortcode_content_encode()`). | |
| 1646 | + // Decoded only now, after the tag scan, for the same reason; left | |
| 1647 | + // encoded, the placeholders were scored as words of their own. | |
| 1648 | + $text = str_replace( | |
| 1649 | + ['_OXY_OPENING_BRACKET_', '_OXY_CLOSING_BRACKET_'], | |
| 1650 | + ['[', ']'], | |
| 1651 | + $text | |
| 1652 | + ); | |
| 1653 | + | |
| 1654 | + // Entities are stored encoded in attributes (&, ’), and would | |
| 1655 | + // otherwise be counted as words. | |
| 1656 | + $text = html_entity_decode($text, ENT_QUOTES | ENT_HTML5, 'UTF-8'); | |
| 1657 | + | |
| 1658 | + return trim((string) preg_replace('/\s+/u', ' ', $text)); | |
| 1659 | + } | |
| 1660 | + | |
| 1661 | + /** | |
| 716 | 1662 | * A node's children, whether it stores them as an array or an object. |
| 717 | 1663 | * |
| 718 | 1664 | * The walker used to return immediately on `!is_array($node)`, so an |
| 719 | 1665 | * object node was dropped along with its entire subtree — silently, as |
| @@ -752,89 +1698,117 @@ | ||
| 752 | 1698 | * |
| 753 | 1699 | * Values are joined with block-level markup so downstream heading, link and |
| 754 | 1700 | * image detection keeps working on the result. |
| 755 | 1701 | * |
| 1702 | + * One depth-first walk, so the output follows the tree's own order, which | |
| 1703 | + * is the order the builders read here render in. This used to be two | |
| 1704 | + * passes over the whole tree, one for the reconstructed headings, links and | |
| 1705 | + * images and one for the remaining text, and the output followed pass | |
| 1706 | + * order: every heading and button on the page first, every paragraph after | |
| 1707 | + * them. That order became the meta description, og:description, the schema | |
| 1708 | + * description and Pro's Markdown for AI document (#907). | |
| 1709 | + * | |
| 756 | 1710 | * @param array $tree Decoded builder tree. |
| 757 | 1711 | * @return string Collected HTML. |
| 758 | 1712 | */ |
| 759 | 1713 | private static function text_from_tree(array $tree): string { |
| 760 | - $collected = []; | |
| 1714 | + // Each entry is [value, is_markup], in tree order. | |
| 1715 | + $entries = []; | |
| 761 | 1716 | |
| 762 | 1717 | // Strings already represented inside reconstructed markup, so the plain |
| 763 | - // sweep below doesn't emit a link label or heading a second time and | |
| 764 | - // double it in the word count. | |
| 1718 | + // text doesn't emit a link label or heading a second time and double it | |
| 1719 | + // in the word count. Applied after the walk, against the whole tree: | |
| 1720 | + // a string folded into markup anywhere is dropped everywhere, exactly | |
| 1721 | + // as it was when the markup pass ran over the whole tree first. Checking | |
| 1722 | + // it during the walk instead would let a bare copy that appears before | |
| 1723 | + // its heading through. | |
| 765 | 1724 | $consumed = []; |
| 766 | 1725 | |
| 767 | - // Pass 1 — rebuild <a>, <img> and <hN> from node *shape*. This has to | |
| 768 | - // happen per node rather than per leaf: a link's label and its | |
| 769 | - // destination are separate sibling fields, so once the tree is | |
| 770 | - // flattened to leaves the pairing is gone. | |
| 771 | - $reconstruct = static function ($node) use (&$reconstruct, &$collected, &$consumed): void { | |
| 772 | - $node = self::as_children($node); | |
| 773 | - if (null === $node) { | |
| 1726 | + // Markup is content wherever it appears; bare strings only count when | |
| 1727 | + // their key says they are content, so slugs and class names stay out of | |
| 1728 | + // the word count. | |
| 1729 | + $leaf = static function ($value, $key) use (&$entries): void { | |
| 1730 | + if (!is_string($value) || '' === trim($value)) { | |
| 774 | 1731 | return; |
| 775 | 1732 | } |
| 776 | 1733 | |
| 777 | - $markup = self::markup_for_node($node, $consumed); | |
| 778 | - if ('' !== $markup) { | |
| 779 | - $collected[] = $markup; | |
| 1734 | + $is_content_key = is_string($key) | |
| 1735 | + && in_array(strtolower($key), self::CONTENT_KEYS, true); | |
| 1736 | + | |
| 1737 | + if ($is_content_key || strpos($value, '<') !== false) { | |
| 1738 | + $entries[] = [$value, false]; | |
| 780 | 1739 | } |
| 1740 | + }; | |
| 781 | 1741 | |
| 782 | - foreach ($node as $child_key => $child) { | |
| 783 | - // A `link` / `image` sub-object is a destination descriptor the | |
| 784 | - // parent has already folded into its markup. Descending into it | |
| 785 | - // would emit the same URL a second time as a bare link, and | |
| 786 | - // would turn an image's own `url` field into a spurious <a>. | |
| 787 | - if (is_string($child_key) | |
| 788 | - && (in_array(strtolower($child_key), self::URL_KEYS, true) | |
| 789 | - || in_array(strtolower($child_key), self::IMAGE_KEYS, true)) | |
| 790 | - ) { | |
| 791 | - continue; | |
| 792 | - } | |
| 1742 | + // Text only, no reconstruction. Used for a `link` / `image` / video | |
| 1743 | + // sub-object: a destination descriptor the parent has already folded | |
| 1744 | + // into its markup. Rebuilding inside it would emit the same URL a second | |
| 1745 | + // time as a bare link and turn an image's own `url` field into a | |
| 1746 | + // spurious <a>, but any copy it carries still counts. | |
| 1747 | + $sweep = static function ($node, $key) use (&$sweep, $leaf): void { | |
| 1748 | + $children = self::as_children($node); | |
| 1749 | + if (null === $children) { | |
| 1750 | + $leaf($node, $key); | |
| 1751 | + return; | |
| 1752 | + } | |
| 793 | 1753 | |
| 794 | - $reconstruct($child); | |
| 1754 | + foreach ($children as $child_key => $child) { | |
| 1755 | + $sweep($child, is_string($child_key) ? $child_key : $key); | |
| 795 | 1756 | } |
| 796 | 1757 | }; |
| 797 | - $reconstruct($tree); | |
| 798 | 1758 | |
| 799 | - // Pass 2 — remaining visible text. | |
| 800 | - $walk = static function ($node, $key = null) use (&$walk, &$collected, &$consumed): void { | |
| 1759 | + // Rebuild <a>, <img> and <hN> from node *shape*, then carry on through | |
| 1760 | + // the node's own fields in order. This has to happen per node rather | |
| 1761 | + // than per leaf: a link's label and its destination are separate | |
| 1762 | + // sibling fields, so once the tree is flattened to leaves the pairing | |
| 1763 | + // is gone. | |
| 1764 | + $walk = static function ($node, $key = null) use (&$walk, $sweep, $leaf, &$entries, &$consumed): void { | |
| 801 | 1765 | $children = self::as_children($node); |
| 802 | - if (null !== $children) { | |
| 803 | - foreach ($children as $child_key => $child) { | |
| 804 | - $walk($child, is_string($child_key) ? $child_key : $key); | |
| 805 | - } | |
| 1766 | + if (null === $children) { | |
| 1767 | + $leaf($node, $key); | |
| 806 | 1768 | return; |
| 807 | 1769 | } |
| 808 | 1770 | |
| 809 | - if (!is_string($node) || '' === trim($node)) { | |
| 810 | - return; | |
| 1771 | + $markup = self::markup_for_node($children, $consumed); | |
| 1772 | + if ('' !== $markup) { | |
| 1773 | + $entries[] = [$markup, true]; | |
| 811 | 1774 | } |
| 812 | 1775 | |
| 813 | - // Already inside a reconstructed tag. | |
| 814 | - if (in_array($node, $consumed, true)) { | |
| 815 | - return; | |
| 816 | - } | |
| 1776 | + foreach ($children as $child_key => $child) { | |
| 1777 | + $next_key = is_string($child_key) ? $child_key : $key; | |
| 817 | 1778 | |
| 818 | - $is_content_key = is_string($key) | |
| 819 | - && in_array(strtolower($key), self::CONTENT_KEYS, true); | |
| 1779 | + if (is_string($child_key) | |
| 1780 | + && (in_array(strtolower($child_key), self::URL_KEYS, true) | |
| 1781 | + || in_array(strtolower($child_key), self::IMAGE_KEYS, true) | |
| 1782 | + || in_array(strtolower($child_key), self::VIDEO_KEYS, true)) | |
| 1783 | + ) { | |
| 1784 | + $sweep($child, $next_key); | |
| 1785 | + continue; | |
| 1786 | + } | |
| 820 | 1787 | |
| 821 | - // Markup is content wherever it appears; bare strings only count | |
| 822 | - // when their key says they are content, so slugs and class names | |
| 823 | - // stay out of the word count. | |
| 824 | - if ($is_content_key || strpos($node, '<') !== false) { | |
| 825 | - $collected[] = $node; | |
| 1788 | + $walk($child, $next_key); | |
| 826 | 1789 | } |
| 827 | 1790 | }; |
| 828 | 1791 | |
| 829 | 1792 | $walk($tree); |
| 830 | 1793 | |
| 1794 | + $collected = []; | |
| 1795 | + foreach ($entries as [$value, $is_markup]) { | |
| 1796 | + // Already inside a reconstructed tag. | |
| 1797 | + if (!$is_markup && in_array($value, $consumed, true)) { | |
| 1798 | + continue; | |
| 1799 | + } | |
| 1800 | + | |
| 1801 | + $collected[] = $value; | |
| 1802 | + } | |
| 1803 | + | |
| 831 | 1804 | if (empty($collected)) { |
| 832 | 1805 | return ''; |
| 833 | 1806 | } |
| 834 | 1807 | |
| 835 | 1808 | // De-duplicate: builder trees often repeat a value across responsive |
| 836 | - // breakpoints, which would otherwise multiply the word count. | |
| 1809 | + // breakpoints, which would otherwise multiply the word count. Keeps the | |
| 1810 | + // first occurrence, so a repeat never moves a value later in the page. | |
| 837 | 1811 | $collected = array_unique($collected); |
| 838 | 1812 | |
| 839 | 1813 | return implode("\n", $collected); |
| 840 | 1814 | } |
| @@ -839,8 +1813,79 @@ | ||
| 839 | 1813 | return implode("\n", $collected); |
| 840 | 1814 | } |
| 841 | 1815 | |
| 842 | 1816 | /** |
| 1817 | + * The video source a node is actually playing, if any. | |
| 1818 | + * | |
| 1819 | + * @since 2.3.1 | |
| 1820 | + * | |
| 1821 | + * @param array $node Builder node. | |
| 1822 | + * @return string Video source, or '' when the node carries none. | |
| 1823 | + */ | |
| 1824 | + private static function video_from(array $node): string { | |
| 1825 | + foreach ($node as $key => $value) { | |
| 1826 | + if (!is_string($key) || !is_string($value)) { | |
| 1827 | + continue; | |
| 1828 | + } | |
| 1829 | + | |
| 1830 | + if (!in_array(strtolower($key), self::VIDEO_TYPE_KEYS, true)) { | |
| 1831 | + continue; | |
| 1832 | + } | |
| 1833 | + | |
| 1834 | + $keys = self::VIDEO_KEYS_BY_TYPE[strtolower(trim($value))] ?? null; | |
| 1835 | + if (null === $keys) { | |
| 1836 | + continue; | |
| 1837 | + } | |
| 1838 | + | |
| 1839 | + // A recognised video_type settles it, including when that | |
| 1840 | + // provider's own field is empty. Falling through to the flat sweep | |
| 1841 | + // there handed back whichever sibling key happened to come first in | |
| 1842 | + // node order — the stale youtube_url left behind after switching | |
| 1843 | + // the widget to a hosted file, which is exactly what keying on the | |
| 1844 | + // declared type is meant to prevent. | |
| 1845 | + $declared = self::url_from($node, $keys); | |
| 1846 | + | |
| 1847 | + return self::is_video_source($declared) ? $declared : ''; | |
| 1848 | + } | |
| 1849 | + | |
| 1850 | + $url = self::url_from($node, self::VIDEO_KEYS); | |
| 1851 | + | |
| 1852 | + return self::is_video_source($url) ? $url : ''; | |
| 1853 | + } | |
| 1854 | + | |
| 1855 | + /** | |
| 1856 | + * Whether a value can be a video source. | |
| 1857 | + * | |
| 1858 | + * `looks_like_url()` also accepts `#anchor`, `mailto:` and `tel:`, which a | |
| 1859 | + * link node may legitimately hold but a video cannot: `<iframe src="#top">` | |
| 1860 | + * is not a video and would reach a video sitemap as one. | |
| 1861 | + * | |
| 1862 | + * @since 2.3.1 | |
| 1863 | + * | |
| 1864 | + * @param string $url Candidate source. | |
| 1865 | + * @return bool | |
| 1866 | + */ | |
| 1867 | + private static function is_video_source(string $url): bool { | |
| 1868 | + return '' !== $url | |
| 1869 | + && (1 === preg_match('#^(https?:)?//#i', $url) || str_starts_with($url, '/')); | |
| 1870 | + } | |
| 1871 | + | |
| 1872 | + /** | |
| 1873 | + * Whether a video source points at a file rather than a provider page. | |
| 1874 | + * | |
| 1875 | + * @since 2.3.1 | |
| 1876 | + * | |
| 1877 | + * @param string $url Video source. | |
| 1878 | + * @return bool | |
| 1879 | + */ | |
| 1880 | + private static function is_video_file(string $url): bool { | |
| 1881 | + $path = (string) wp_parse_url($url, PHP_URL_PATH); | |
| 1882 | + $ext = strtolower((string) pathinfo($path, PATHINFO_EXTENSION)); | |
| 1883 | + | |
| 1884 | + return in_array($ext, self::VIDEO_FILE_EXTENSIONS, true); | |
| 1885 | + } | |
| 1886 | + | |
| 1887 | + /** | |
| 843 | 1888 | * Rebuild the HTML a single builder node represents, if any. |
| 844 | 1889 | * |
| 845 | 1890 | * Looks only at the node's own fields (plus one level of nesting, because |
| 846 | 1891 | * builders commonly wrap a destination as `{ url: … }`). Returns an empty |
| @@ -856,12 +1901,23 @@ | ||
| 856 | 1901 | private static function markup_for_node(array $node, array &$consumed): string { |
| 857 | 1902 | $text = self::first_value($node, self::CONTENT_KEYS); |
| 858 | 1903 | $url = self::url_from($node, self::URL_KEYS); |
| 859 | 1904 | $image = self::image_from($node); |
| 1905 | + $video = self::video_from($node); | |
| 860 | 1906 | $tag = self::heading_tag_from($node); |
| 861 | 1907 | |
| 862 | 1908 | $parts = []; |
| 863 | 1909 | |
| 1910 | + // Video: an embed shape rather than a link, so the video detector can | |
| 1911 | + // see it while the link counters do not mistake it for an outbound | |
| 1912 | + // link. A file source becomes <video src>, anything else an <iframe>, | |
| 1913 | + // matching how the builder itself renders the two cases. | |
| 1914 | + if ('' !== $video) { | |
| 1915 | + $parts[] = self::is_video_file($video) | |
| 1916 | + ? sprintf('<video src="%s"></video>', esc_url_raw($video)) | |
| 1917 | + : sprintf('<iframe src="%s"></iframe>', esc_url_raw($video)); | |
| 1918 | + } | |
| 1919 | + | |
| 864 | 1920 | // Image: alt text matters as much as the tag, since alt checks run over |
| 865 | 1921 | // whatever this returns. |
| 866 | 1922 | if ('' !== $image['url']) { |
| 867 | 1923 | $alt = '' !== $image['alt'] ? $image['alt'] : (string) self::first_value($node, self::ALT_KEYS); |
| @@ -1050,13 +2106,30 @@ | ||
| 1050 | 2106 | return (bool) preg_match('#^(mailto:|tel:|\#)#i', $value); |
| 1051 | 2107 | } |
| 1052 | 2108 | |
| 1053 | 2109 | /** |
| 1054 | - * Whether a value carries no readable text. | |
| 2110 | + * Whether a value carries nothing worth analyzing. | |
| 1055 | 2111 | * |
| 2112 | + * Readable text is the usual signal, but not the only one: a page can be | |
| 2113 | + * made entirely of media. A builder section holding just a gallery | |
| 2114 | + * reconstructs to `<img>` tags and one holding just a video widget to a | |
| 2115 | + * single `<iframe>` — both strip to an empty string, so a text-only test | |
| 2116 | + * discarded them here and the page fell through to the next builder key, | |
| 2117 | + * then to the raw markup, and finally reported as having no content at all. | |
| 2118 | + * | |
| 2119 | + * Comments are dropped before the tag test: the raw markup this class falls | |
| 2120 | + * back to on a builder page is unrendered block comments, which must stay | |
| 2121 | + * blank rather than be mistaken for reconstructed media. | |
| 2122 | + * | |
| 1056 | 2123 | * @param string $value Candidate content. |
| 1057 | 2124 | * @return bool |
| 1058 | 2125 | */ |
| 1059 | 2126 | private static function is_blank(string $value): bool { |
| 1060 | - return '' === trim(wp_strip_all_tags($value)); | |
| 2127 | + if ('' !== trim(wp_strip_all_tags($value))) { | |
| 2128 | + return false; | |
| 2129 | + } | |
| 2130 | + | |
| 2131 | + $without_comments = (string) preg_replace('~<!--.*?-->~s', '', $value); | |
| 2132 | + | |
| 2133 | + return 1 !== preg_match('~<(?:a|img|iframe|video|source)\b~i', $without_comments); | |
| 1061 | 2134 | } |
| 1062 | 2135 | } |
