← All changes | includes/seo/class-word-count-index.php +101 -11 2.14.1 → 2.14.2 View file →

		@@ -62,10 +62,19 @@
62	62	public const META_KEY = '_thinkrank_word_count';
63	63
64	64	/**
65	65	* Counting rules version. Bump to retire every stored entry.
	66	+ *
	67	+ * Version 3 stopped counting shortcode syntax as words and stopped merging
	68	+ * two words that only a tag separated (#893), so every version 2 entry is
	69	+ * recounted once.
	70	+ *
	71	+ * Version 4 stopped counting tokens with no letter or digit in them. The
	72	+ * tag-to-space change in version 3 left the punctuation after an inline
	73	+ * tag ("<a>link</a>.") standing alone, where it was counted as a word, so
	74	+ * every version 3 entry is recounted once.
66	75	*/
67		- public const VERSION = 2;
	76	+ public const VERSION = 4;
68	77
69	78	/**
70	79	* Short codes for the counting unit, as stored in an entry.
71	80	*
		@@ -534,17 +543,9 @@
534	543	* @param string $content HTML or text.
535	544	* @return int
536	545	*/
537	546	public static function count_text(string $content): int {
538		- // Scripts and styles carry no reading matter, and a page builder's
539		- // output can hold a great deal of both. Counting them would make an
540		- // empty page look substantial, which is the failure that matters here.
541		- $text = preg_replace('#<(script|style)\b[^>]*>.*?</\1>#is', ' ', $content);
542		- $text = wp_strip_all_tags((string) $text);
543		- $text = html_entity_decode($text, ENT_QUOTES | ENT_HTML5, 'UTF-8');
544		- // Non-breaking spaces are spaces to a reader.
545		- $text = str_replace(["\xc2\xa0", "\xe2\x80\x8b"], ' ', $text);
546		- $text = trim((string) preg_replace('/\s+/u', ' ', $text));
	547	+ $text = self::reading_text($content);
547	548
548	549	if ('' === $text) {
549	550	return 0;
550	551	}
		@@ -556,10 +557,99 @@
556	557	case 'characters_excluding_spaces':
557	558	return mb_strlen(str_replace(' ', '', $text));
558	559
559	560	default:
560		- return count(preg_split('/\s+/u', $text, -1, PREG_SPLIT_NO_EMPTY) ?: []);
	561	+ return self::count_words($text);
561	562	}
	563	+ }
	564	+
	565	+ /**
	566	+ * Count the words in a line of reading text.
	567	+ *
	568	+ * A word is a whitespace-separated token holding at least one letter or
	569	+ * digit, in any script. A token of punctuation or symbols alone is not
	570	+ * one: reading_text() turns every tag into a space, so the full stop in
	571	+ * "<a>link</a>." and the "৳" in WooCommerce's
	572	+ * "<span>৳</span>100" stand on their own, and a dash set between spaces
	573	+ * ("one — two") does the same in plain prose. Core's JS word counter drops
	574	+ * punctuation too. Letters and digits are matched by Unicode property, so
	575	+ * Bengali, Arabic or Cyrillic words still count; a combining mark is part
	576	+ * of the letter before it, so a Bengali word keeps its vowel signs.
	577	+ *
	578	+ * Only the words unit uses this. The character units count every
	579	+ * character, as core does, which is how CJK locales are counted.
	580	+ *
	581	+ * @since 2.14.2
	582	+ *
	583	+ * @param string $text Text as returned by {@see self::reading_text()}.
	584	+ * @return int
	585	+ */
	586	+ public static function count_words(string $text): int {
	587	+ $tokens = preg_split('/\s+/u', $text, -1, PREG_SPLIT_NO_EMPTY);
	588	+
	589	+ if (!is_array($tokens)) {
	590	+ return 0;
	591	+ }
	592	+
	593	+ return count(preg_grep('/[\p{L}\p{N}]/u', $tokens) ?: []);
	594	+ }
	595	+
	596	+ /**
	597	+ * The text a reader reads in a piece of stored content, as one line.
	598	+ *
	599	+ * Everything that is markup rather than reading matter is removed:
	600	+ *
	601	+ * - **Scripts and styles.** A page builder's output can hold a great deal
	602	+ * of both. Counting them would make an empty page look substantial,
	603	+ * which is the failure that matters here.
	604	+ * - **Shortcode syntax** (#893). `[vc_column width="1/2"]` is not three
	605	+ * words, and a WPBakery or Divi classic page is mostly made of it, so
	606	+ * counting it reported a 216-word page as 325 and called it not thin.
	607	+ * The shortcodes are stripped, not rendered: `do_shortcode()` would
	608	+ * count what they output more accurately, but it means executing every
	609	+ * shortcode on the site inside a batch count, with whatever side effects
	610	+ * each one has (#860, #864). A shortcode that renders real prose is
	611	+ * undercounted, which is the safe direction for a thin content report.
	612	+ * Anything shortcode-shaped is stripped, registered or not, because a
	613	+ * builder's shortcodes are often not registered when the count runs.
	614	+ * Bracketed prose that looks like one ("see [note 4]") goes with it;
	615	+ * "[1]" and "[...]" do not, since a shortcode name starts with a letter.
	616	+ * - **Tags**, replaced with a space rather than deleted, the way core's
	617	+ * own word counter does, so `<p>five</p><p>six</p>` stays two words.
	618	+ *
	619	+ * @since 2.14.2
	620	+ *
	621	+ * @param string $content HTML or text.
	622	+ * @return string Plain text with whitespace collapsed to single spaces.
	623	+ */
	624	+ public static function reading_text(string $content): string {
	625	+ // A `/u` pattern answers null on bytes that are not valid UTF-8, and
	626	+ // the string casts below turned that into "": one Latin-1 byte from an
	627	+ // old import made the whole page read as empty, a word count of 0 in
	628	+ // both this report and the SEO score. Replace the bad bytes instead.
	629	+ if ('' !== $content && 1 !== preg_match('//u', $content) && function_exists('mb_scrub')) {
	630	+ $content = mb_scrub($content, 'UTF-8');
	631	+ }
	632	+
	633	+ $text = (string) preg_replace('#<(script|style)\b[^>]*>.*?</\1>#is', ' ', $content);
	634	+
	635	+ // Before tags: an attribute value may hold a ">", which would end a
	636	+ // tag match early. WordPress does not allow "[" or "]" inside a
	637	+ // shortcode's attributes, so neither is crossed; that also keeps a
	638	+ // stray "[" in prose from swallowing the text after it. The optional
	639	+ // outer brackets take the escaped form "[[name]]" whole, rather than
	640	+ // leaving two stray brackets to be counted as words.
	641	+ $text = (string) preg_replace('/\[?\[\/?[A-Za-z][\w-]*[^\[\]]*\]\]?/u', ' ', $text);
	642	+
	643	+ $text = (string) preg_replace('/<!--.*?-->/s', ' ', $text);
	644	+ $text = (string) preg_replace('#</?[A-Za-z][^>]*>#', ' ', $text);
	645	+ // Backstop for anything malformed the patterns above did not take.
	646	+ $text = wp_strip_all_tags($text);
	647	+ $text = html_entity_decode($text, ENT_QUOTES | ENT_HTML5, 'UTF-8');
	648	+ // Non-breaking spaces are spaces to a reader.
	649	+ $text = str_replace(["\xc2\xa0", "\xe2\x80\x8b"], ' ', $text);
	650	+
	651	+ return trim((string) preg_replace('/\s+/u', ' ', $text));
562	652	}
563	653
564	654	/**
565	655	* The unit the site counts in.