Escaping by position

Whatever you type below is placed into the page in four different spots. PHV knows, at compile time, where each <?= ?> lands in the HTML and treats the value accordingly.

1. Element content

Written as <p><?= $input ?></p>. Tags show up as text:

2. Quoted attribute

Written as title="<?= $input ?>". Quotes cannot end the attribute:

Hover this text to see the attribute value.

3. URL

The start of href only accepts a SafeUrl, made by safe_url($url), which lets through http(s), mailto, tel and relative URLs and turns anything else into #:

this link points to #

4. Inside <script>

A <script> only accepts a SafeJs, made by json_html($input). The script below reads your text as a JavaScript string and counts its characters:

(the script did not run)

That <script> runs because PHV added a CSP nonce to it. A script that got in through raw() would not get one, so the browser would block it.

What does not compile

Putting a plain string where escaping cannot protect it is an error, not a warning:

$ phv check bad.phv
error: a URL attribute cannot start with an unchecked value
 --> bad.phv:4:14
  |
4 | <a href="<?= $url ?>">this link</a>
  |              ^^^^
  = note: use `safe_url($url)`, which only allows http(s), mailto, tel and relative URLs (blocks `javascript:`); `escape()` does not make a URL safe

The same goes for unquoted attributes, event handlers such as onclick, style and plain strings inside <script>.

Source of this page

Read from disk by lib/source.phv as the page is served.

public/escape.phv

<?phv
import std.http.request;
import lib.source.sources;

$req = request();
$input = $req->query("input") ?? "<script>alert(\"hi\")</script> & 'quotes'";
$url = $req->query("url") ?? "javascript:alert(document.cookie)";
if ($input->len() > 200) {
    $input = $input->slice(0, 200);
}

$body = render("views/escape.phv.html", input: $input, url: $url);
echo render(
    "views/layout.phv.html",
    title: "Escaping",
    path: "/escape",
    body: $body,
    files: sources(["public/escape.phv", "views/escape.phv.html"]),
);

views/escape.phv.html

<?phv params(string $input, string $url) ?>
<h1>Escaping by position</h1>
<p class="lead">Whatever you type below is placed into the page in four different spots. PHV knows, at compile
time, where each <code>&lt;?= ?&gt;</code> lands in the HTML and treats the value accordingly.</p>

<form method="get" action="/escape" class="stack">
  <label for="input">Any text</label>
  <input id="input" name="input" value="<?= $input ?>" maxlength="200">
  <label for="url">A link target</label>
  <input id="url" name="url" value="<?= $url ?>">
  <button>Update the page</button>
</form>

<div class="demo">
  <h2>1. Element content</h2>
  <p>Written as <code>&lt;p&gt;&lt;?= $input ?&gt;&lt;/p&gt;</code>. Tags show up as text:</p>
  <p class="out"><?= $input ?></p>

  <h2>2. Quoted attribute</h2>
  <p>Written as <code>title="&lt;?= $input ?&gt;"</code>. Quotes cannot end the attribute:</p>
  <p class="out"><span title="<?= $input ?>">Hover this text to see the attribute value.</span></p>

  <h2>3. URL</h2>
  <p>The start of <code>href</code> only accepts a <code>SafeUrl</code>, made by <code>safe_url($url)</code>,
  which lets through <code>http(s)</code>, <code>mailto</code>, <code>tel</code> and relative URLs and turns
  anything else into <code>#</code>:</p>
  <p class="out"><a href="<?= safe_url($url) ?>">this link</a> points to <code><?= safe_url($url) ?></code></p>

  <h2>4. Inside &lt;script&gt;</h2>
  <p>A <code>&lt;script&gt;</code> only accepts a <code>SafeJs</code>, made by <code>json_html($input)</code>. The
  script below reads your text as a JavaScript string and counts its characters:</p>
  <p class="out" id="js-out">(the script did not run)</p>
  <script>
    const input = <?= json_html($input) ?>;
    document.getElementById("js-out").textContent =
      "JavaScript received a string of " + [...input].length + " characters.";
  </script>
  <p class="muted">That <code>&lt;script&gt;</code> runs because PHV added a CSP nonce to it. A script that got in
  through <code>raw()</code> would not get one, so the browser would block it.</p>
</div>

<h2>What does not compile</h2>
<p>Putting a plain <code>string</code> where escaping cannot protect it is an error, not a warning:</p>
<pre class="term"><code>$ phv check bad.phv
error: a URL attribute cannot start with an unchecked value
 --> bad.phv:4:14
  |
4 | &lt;a href="&lt;?= $url ?&gt;"&gt;this link&lt;/a&gt;
  |              ^^^^
  = note: use `safe_url($url)`, which only allows http(s), mailto, tel and relative URLs (blocks `javascript:`); `escape()` does not make a URL safe</code></pre>
<p>The same goes for unquoted attributes, event handlers such as <code>onclick</code>, <code>style</code> and
plain strings inside <code>&lt;script&gt;</code>.</p>