block. // Missing a formatter reduces presentation; it never hides the content. // // OPERATIONAL SAFEGUARDS // Paths are contained to SCRIPTS_DIR. // $rel is resolved with realpath() and required to remain under SCRIPTS_DIR, be a // regular file, and carry a .md extension. This is deliberate defence for a parameter // that will arrive from a request the moment this is wired up — without it, a // traversal sequence reaches any file the web user can read. // // Markdown is rendered in safe mode. // Parsedown runs with setSafeMode(true), and the
fallback escapes everything.
// These files are trusted today, but they are also synced between hosts.
//
// Substituted conf values are escaped.
// htmlspecialchars() is applied to both the value and the variable name, so a conf
// value containing markup cannot inject into the rendered page.
//
// Read-only. Discovers and renders; never writes a doc.
//
// EXPORTS
// vv_docs_tree() every *.md under SCRIPTS_DIR, relative paths, sorted
// vv_docs_render() one file to HTML with conf substitution applied
//
// CONFIGURATION
// SCRIPTS_DIR the containment root and the discovery root
// PARSEDOWN_PATH /usr/local/emhttp/plugins/varaverk/lib/Parsedown.php — optional
// ═══════════════════════════════════════════════════════════════════════════════════════════════
require_once __DIR__ . '/config.php';
define('PARSEDOWN_PATH', '/usr/local/emhttp/plugins/varaverk/lib/Parsedown.php');
function vv_docs_tree(): array {
$base = SCRIPTS_DIR;
$tree = [];
$files = new RecursiveIteratorIterator(
new RecursiveDirectoryIterator($base, FilesystemIterator::SKIP_DOTS),
RecursiveIteratorIterator::SELF_FIRST
);
foreach ($files as $f) {
if ($f->isFile() && strtolower($f->getExtension()) === 'md') {
$rel = ltrim(str_replace($base, '', $f->getPathname()), '/');
$tree[] = $rel;
}
}
sort($tree);
return $tree;
}
function vv_docs_render(string $rel, array $vars): string {
// Containment check — $rel is expected to come from a request parameter once this is
// wired to a page. Resolve it and require the result to stay inside SCRIPTS_DIR and to
// still be a .md file, so a traversal sequence cannot reach arbitrary files.
$base = realpath(SCRIPTS_DIR);
$path = realpath(SCRIPTS_DIR . '/' . $rel);
if ($base === false || $path === false) return 'File not found.
';
if (!str_starts_with($path, $base . '/')) return 'File not found.
';
if (strtolower(pathinfo($path, PATHINFO_EXTENSION)) !== 'md') return 'File not found.
';
if (!is_file($path)) return 'File not found.
';
return vv_docs_markdown(file_get_contents($path), $vars);
}
// Markdown → HTML for the subset these docs actually use: headings, paragraphs, bullet and
// ordered lists, tables, blockquotes, fenced code, horizontal rules, and inline bold / italic /
// code / links.
//
// Hand-rolled rather than vendored. Parsedown was the original plan and PARSEDOWN_PATH is still
// honoured below if the file ever appears, but it has never been present on this system, so the
// only path this function ever took was a dump of raw markdown — which is not a document,
// it is the source of one. A renderer for a subset we control is a few dozen lines and adds
// nothing to a repo that gets pushed.
//
// Escape first, then format. Every line is passed through htmlspecialchars() before any tag is
// introduced, so the only HTML in the output is HTML this function put there. That is what makes
// safe mode unnecessary rather than merely configured — and it is why the $VAR substitution runs
// here, after escaping. Injecting into the markdown before rendering, as this file used to
// do, meant both Parsedown's safe mode and the fallback escaped the tags and printed them
// as literal text. The feature never worked; nothing called it, so nothing reported it.
function vv_docs_markdown(string $md, array $vars): string {
if (file_exists(PARSEDOWN_PATH)) {
require_once PARSEDOWN_PATH;
$pd = new Parsedown();
$pd->setSafeMode(true);
return $pd->text($md);
}
$inline = function (string $s) use ($vars): string {
$s = htmlspecialchars($s, ENT_QUOTES, 'UTF-8');
// `$VAR` → the live conf value. Unresolved names render in their own class rather than
// silently reading as prose, so a stale reference in a doc is visible as a defect.
$s = preg_replace_callback('/`\$([A-Z0-9_]+)`/', function ($m) use ($vars) {
return isset($vars[$m[1]])
? '' . htmlspecialchars($vars[$m[1]]) . ''
: '$' . $m[1] . '';
}, $s);
$s = preg_replace('/`([^`]+)`/', '$1', $s);
$s = preg_replace('/\*\*([^*]+)\*\*/', '$1', $s);
$s = preg_replace('/(?$1', $s);
// Links are restricted to http/https and relative paths — a doc is trusted, but these
// files sync between hosts, so javascript: must not be reachable through one.
$s = preg_replace('/\[([^\]]+)\]\((https?:\/\/[^\s)]+|[^\s):]+)\)/', '$1', $s);
return $s;
};
$out = '';
$list = null; // 'ul' | 'ol' | null
$inTable = false;
$inCode = false;
$para = [];
$quote = [];
$li = [];
$flushPara = function () use (&$para, &$out, $inline) {
if (!$para) return;
$out .= '' . $inline(implode(' ', $para)) . "
\n";
$para = [];
};
$flushQuote = function () use (&$quote, &$out, $inline) {
if (!$quote) return;
$out .= '' . $inline(implode(' ', $quote)) . "
\n";
$quote = [];
};
// A list item is buffered rather than emitted on sight, because markdown wraps: a bullet
// whose text runs onto the next line is one item, not an item followed by a paragraph.
$flushLi = function () use (&$li, &$out, $inline) {
if (!$li) return;
$out .= '' . $inline(implode(' ', $li)) . " \n";
$li = [];
};
$closeList = function () use (&$list, &$out, &$flushLi) {
$flushLi();
if ($list) { $out .= "$list>\n"; $list = null; }
};
$closeTable = function () use (&$inTable, &$out) {
if ($inTable) { $out .= "\n"; $inTable = false; }
};
foreach (explode("\n", str_replace("\r\n", "\n", $md)) as $line) {
if (preg_match('/^```/', $line)) {
$flushPara(); $closeList(); $closeTable();
$out .= $inCode ? "
\n" : '';
$inCode = !$inCode;
continue;
}
if ($inCode) { $out .= htmlspecialchars($line, ENT_QUOTES, 'UTF-8') . "\n"; continue; }
$t = trim($line);
// One guard rather than a flush in every branch: the quote ends the moment a line is
// not a quote line, whatever that next line turns out to be.
if ($quote && !str_starts_with($t, '>')) $flushQuote();
if ($t === '') { $flushPara(); $closeList(); $closeTable(); continue; }
if (preg_match('/^(---+|\*\*\*+)$/', $t)) { $flushPara(); $closeList(); $closeTable(); $out .= "
\n"; continue; }
if (preg_match('/^(#{1,6})\s+(.*)$/', $t, $m)) {
$flushPara(); $closeList(); $closeTable();
$n = strlen($m[1]);
$out .= "" . $inline($m[2]) . " \n";
continue;
}
// Consecutive > lines are one quote, not one per line — same wrapping convention as
// paragraphs, which is how they are written.
if (preg_match('/^>\s?(.*)$/', $t, $m)) {
$closeList(); $closeTable();
if (!$quote) $flushPara();
$quote[] = $m[1];
continue;
}
// Tables: a header row, a separator of dashes, then body rows. The separator is what
// identifies the block — a lone pipe in prose is not a table.
if (strpos($t, '|') !== false && preg_match('/^\|?[\s:-]*-[\s|:-]*\|/', $t)) {
continue; // separator consumed by the header below
}
if (strpos($t, '|') !== false && substr_count($t, '|') >= 2) {
$cells = array_map('trim', explode('|', trim($t, '| ')));
if (!$inTable) {
$flushPara(); $closeList();
$out .= '';
foreach ($cells as $c) $out .= '' . $inline($c) . ' ';
$out .= " \n";
$inTable = true;
} else {
$out .= '';
foreach ($cells as $c) $out .= '' . $inline($c) . ' ';
$out .= " \n";
}
continue;
}
$closeTable();
if (preg_match('/^[-*]\s+(.*)$/', $t, $m)) {
$flushPara();
if ($list !== 'ul') { $closeList(); $out .= "\n"; $list = 'ul'; }
else { $flushLi(); }
$li[] = $m[1];
continue;
}
if (preg_match('/^\d+\.\s+(.*)$/', $t, $m)) {
$flushPara();
if ($list !== 'ol') { $closeList(); $out .= "\n"; $list = 'ol'; }
else { $flushLi(); }
$li[] = $m[1];
continue;
}
// Lazy continuation: plain text directly under an open list item belongs to that item.
// Without this a wrapped bullet renders as a bullet plus an orphan paragraph, which is
// how the first version of this shipped.
if ($list && $li) { $li[] = $t; continue; }
$para[] = $t;
}
$flushQuote(); $flushPara(); $closeList(); $closeTable();
if ($inCode) $out .= "\n";
return $out;
}