Lightpanda: the headless browser designed for AI and automation
fix: broken attribute lookup cross frame
Karl Seguin24ddb4ac396b
diff --git a/src/browser/tests/frames/cross_realm_attributes.html b/src/browser/tests/frames/cross_realm_attributes.htmlnew file mode 100644index 000000000..ac8722d0c--- /dev/null+++ b/src/browser/tests/frames/cross_realm_attributes.html@@ -0,0 +1,52 @@+<!DOCTYPE html>+<head></head>+<body>+<script src="../testing.js"></script>++<!--+Attribute lookups must work across realms. Names not in the static intern+table (data-*, custom attributes) are canonicalized per-frame, so a parent+reading an iframe element's attributes (or vice versa) can't rely on+pointer identity — it must fall back to comparing bytes. Regression: parent+reads of iframe data-* returned null (WPT encoding/*).+-->+<iframe id="idf" src="support/cross_realm_attributes.html"></iframe>++<script id="cross_realm_reads">+ testing.onload(() => {+ const el = document.getElementById('idf').contentDocument.getElementById('s1');++ testing.expectEqual('5F', el.getAttribute('data-cp'));+ testing.expectEqual('5F', el.dataset.cp);+ testing.expectEqual('A1B2', el.dataset.bytes);+ testing.expectEqual('ns1', el.getAttribute('nonstandardattr'));+ testing.expectTrue(el.hasAttribute('nonstandardattr'));+ testing.expectFalse(el.hasAttribute('data-nope'));+ testing.expectEqual(null, el.getAttribute('data-nope'));+ });+</script>++<script id="cross_realm_writes">+ testing.onload(() => {+ const el = document.getElementById('idf').contentDocument.getElementById('s1');+ const count = el.attributes.length;++ // Updating an existing attribute from the parent realm must find the+ // entry (not append a duplicate).+ el.setAttribute('data-cp', '60');+ testing.expectEqual('60', el.getAttribute('data-cp'));+ testing.expectEqual(count, el.attributes.length);++ // A new name canonicalized by the parent realm must be readable from+ // the iframe's own realm (and back).+ el.setAttribute('data-parentset', 'pv');+ const iwin = document.getElementById('idf').contentWindow;+ testing.expectEqual('pv', iwin.readAttr('data-parentset'));+ testing.expectEqual('60', iwin.readAttr('data-cp'));++ el.removeAttribute('data-parentset');+ testing.expectFalse(el.hasAttribute('data-parentset'));+ testing.expectEqual(null, iwin.readAttr('data-parentset'));+ });+</script>+</body>diff --git a/src/browser/tests/frames/support/cross_realm_attributes.html b/src/browser/tests/frames/support/cross_realm_attributes.htmlnew file mode 100644index 000000000..7197b1cf2--- /dev/null+++ b/src/browser/tests/frames/support/cross_realm_attributes.html@@ -0,0 +1,10 @@+<!DOCTYPE html>+<body>+ <span id="s1" data-cp="5F" data-bytes="A1B2" nonstandardattr="ns1">x</span>+ <script>+ // Called by the parent test: reads execute in this frame's realm.+ function readAttr(name) {+ return document.getElementById('s1').getAttribute(name);+ }+ </script>+</body>diff --git a/src/browser/webapi/element/Attribute.zig b/src/browser/webapi/element/Attribute.zigindex d60f39910..111bb924c 100644--- a/src/browser/webapi/element/Attribute.zig+++ b/src/browser/webapi/element/Attribute.zig@@ -385,29 +385,12 @@ pub const List = struct {const normalized =if (self.normalize) try normalizeNameForLookup(name, frame) else name;- // A name that was never canonicalized can't be in any list.- const canonical = lookupCanonicalName(normalized.str(), frame) orelse {- return .{ .normalized = normalized, .entry = null };- };return .{.normalized = normalized,- .entry = self.getEntryWithCanonicalName(canonical.ptr),+ .entry = self.getEntryWithNormalizedName(normalized),};}- // This is one of the wins of the canonical names...we can compare strings- // as a single pointer comparison. By applying the same canonicalization to- // the lists attributes name and to an input, we get the same pointer for- // a given string.- fn getEntryWithCanonicalName(self: *const List, name_ptr: [*]const u8) ?*Entry {- for (self._entries[0..self._len]) |*e| {- if (e._name_ptr == name_ptr) {- return e;- }- }- return null;- }-fn getEntryWithNormalizedName(self: *const List, name: String) ?*Entry {const name_str = name.str();for (self._entries[0..self._len]) |*e| {@@ -514,10 +497,10 @@ pub fn validateAttributeName(name: String) !void {}// Every stored entry name either comes from the static String.intern or from-// the frame._attribute_names. This doesn't just avoid extra dupes/allocations,-// it gives us comparable pointer.-// For the lifetime of a frame:-// canonicalizeName("attrx").ptr == canonicalizeName("attrx").ptr+// the frame._attribute_names. Beyond avoiding extra dupes/allocations, this+// gives a stable pointer for the frame's lifetime, which List.LookupKey+// relies on for identity. The pointer is NOT comparable across frames (each+// frame has its own pool), which is why lookups byte-compare.fn canonicalizeName(name: []const u8, frame: *Frame) ![]const u8 {if (String.intern(name)) |static| {return static;@@ -529,19 +512,6 @@ fn canonicalizeName(name: []const u8, frame: *Frame) ![]const u8 {return gop.key_ptr.*;}-// Let's say you want to find an attribute name "attrx". We could iterate an-// attribute list to do the string comparison. OR, we could run "attrx" through-// the same canonicalization as we applied to the attribute's names. If we don't-// find a canonical value, then it's then this attribute was never canonicalized-// before (and thus, we can shortcircuit a search). If we DO canonicalize it-// the we get do a pointer comparison rather than a full string comparison.-fn lookupCanonicalName(name: []const u8, frame: *const Frame) ?[]const u8 {- if (String.intern(name)) |static| {- return static;- }- return frame._attribute_names.getKey(name);-}-fn normalizeNameForLookup(name: String, frame: *Frame) !String {if (!needsLowerCasing(name.str())) {return name;