) — the simplified * pattern will happily match, but handle() produces an internallink * with a broken src; spec tests for this stay in skip.php. * - Balanced-parens inside URLs [link](foo(bar)) — matches truncate * at first `)`, producing odd output; also in skip.php. * - Title HTML attribute — DokuWiki link handler instructions have no * title-attribute slot, and plumbing one through every renderer just * for this is out of scope. The title parses cleanly but is discarded. * - Mixed text + image in the label ([prefix  suffix](url)) * — matches DW's policy: Internallink only converts the label to a * media descriptor when it matches `^{{…}}$` exactly. */ class GfmLink extends AbstractMode { // URL slot character set: any non-paren / non-newline char, OR a // backslash-escape sequence so an escaped `\)` doesn't terminate the // URL early (spec examples 504/506/508). Backslash-unescape is // applied post-extraction; the pattern only needs to keep escaped // close-parens from prematurely ending the match. private const URL_CHAR = '(?:\\\\.|[^)\n])'; // Label character set: forbids unescaped `[` / `]` so the outer // bracket pair stays balanced, but allows `\[` / `\]` so an escaped // bracket can appear inside the label (spec example 523). The same // backslash-escape trick the URL slot already uses. A bare `\n` is // permitted as long as it is not followed by a blank line — soft // line breaks inside link text are allowed by the spec, blank lines // are not (and they would also tie up `\n#`-anchored block modes). private const LABEL_CHAR = '(?:\\\\.|[^\[\]\n]|\n(?![ \t]*\n))'; // Image sub-pattern reused for both the label alternative in the main // pattern and the image-as-label detector in handle(). No capture // groups here — the lexer wraps user patterns in a capture and // additional captures would renumber unpredictably. private const IMAGE_SUB = '!\[' . self::LABEL_CHAR . '*\]\(' . self::URL_CHAR . '+\)'; /** @inheritdoc */ public function getSort() { return 300; } /** @inheritdoc */ public function connectTo($mode) { // Outer shape: `[text-or-image](url)`. Text class forbids // unescaped brackets and newlines but allows `\[` / `\]`; the // image alternative explicitly matches one inline image. URL // slot is permissive — handle() does URL / title splitting // post-entry, mirroring how DW Internallink parses inside `[[...]]`. $pattern = '\[(?!\[)(?:' . self::LABEL_CHAR . '+|' . self::IMAGE_SUB . ')\]\(' . self::URL_CHAR . '+\)'; $this->Lexer->addSpecialPattern($pattern, $mode, 'gfm_link'); } /** @inheritdoc */ public function handle($match, $state, $pos, Handler $handler) { // Detect image-as-label `[](target)`. Parallels // Internallink's `^{{…}}$` check — when the label is exactly an // inline image, parse it into a media descriptor; otherwise // treat the label as plain text. if (preg_match('/^\[(' . self::IMAGE_SUB . ')\]\((' . self::URL_CHAR . '+)\)$/', $match, $m)) { $label = $this->parseImageDescriptor($m[1]); $targetUrl = $this->extractUrl($m[2]); } else { // Plain text label can't contain `]`, so the first `](` is // the label/target separator. $sep = strpos($match, ']('); $label = Escape::unescapeBackslashes(substr($match, 1, $sep - 1)); $targetUrl = $this->extractUrl(substr($match, $sep + 2, -1)); } // Classify on the raw URL so windowssharelink detection sees the // literal `\\host\path` runs intact — GFM's `\\` → `\` collapse // would otherwise destroy the share prefix. [$call, $args] = Link::classify($targetUrl, $label); if ($call !== 'windowssharelink') { $args[0] = Escape::unescapeBackslashes($args[0]); } $handler->addCall($call, $args, $pos); return true; } /** * Extract the URL from a parenthesized payload: trim surrounding * whitespace, take the first whitespace-delimited token, then * apply GFM's URL-slot transformations (entity decoding; * backslash-unescape happens later, after Link::classify, because * windowssharelink detection needs the raw `\\` runs intact). * Any trailing title is discarded (no renderer slot for it). */ private function extractUrl(string $inside): string { $inside = trim($inside); $url = substr($inside, 0, strcspn($inside, " \t\n")); // remove optional title return HtmlEntity::decode($url); } /** * Parse an inline image sub-match `` into the media * descriptor shape Media::parseMedia() returns, so the link handler * can treat it as a media label identically to `[[page|{{img}}]]`. */ private function parseImageDescriptor(string $imageMatch): array { $sep = strpos($imageMatch, ']('); $alt = Escape::unescapeBackslashes(substr($imageMatch, 2, $sep - 2)); $imgUrl = Escape::unescapeBackslashes($this->extractUrl(substr($imageMatch, $sep + 2, -1))); $p = MediaHelper::parseParameters($imgUrl); $type = (media_isexternal($p['src']) || link_isinterwiki($p['src'])) ? 'externalmedia' : 'internalmedia'; return [ 'type' => $type, 'src' => $p['src'], 'title' => $alt !== '' ? $alt : null, 'align' => $p['align'], 'width' => $p['width'], 'height' => $p['height'], 'cache' => $p['cache'], 'linking' => $p['linking'], ]; } }