Lightpanda: the headless browser designed for AI and automation
selector: implement :link, :lang, :empty spec semantics; parser fixes
Francis Bouvierfd174986cf9c
diff --git a/src/browser/webapi/selector/List.zig b/src/browser/webapi/selector/List.zigindex e5cbfd5e6..8bea494f5 100644--- a/src/browser/webapi/selector/List.zig+++ b/src/browser/webapi/selector/List.zig@@ -612,13 +612,15 @@ fn matchesPseudoClass(el: *Node.Element, pseudo: Selector.PseudoClass, scope: *N},.focus_visible => return false,- // Link states- .link => return false,- .visited => return false,- .any_link => {- if (el.getTag() != .anchor) return false;+ // Link states. In a headless browser no link is ever visited, so+ // :link matches every hyperlink (a or area with an href attribute)+ // and :visited matches nothing.+ .link, .any_link => {+ const tag = el.getTag();+ if (tag != .anchor and tag != .area) return false;return el.getAttributeSafe(comptime .wrap("href")) != null;},+ .visited => return false,.target => {const element_id = el.getAttributeSafe(comptime .wrap("id")) orelse return false;const doc = node.ownerDocument(frame) orelse return false;@@ -638,7 +640,19 @@ fn matchesPseudoClass(el: *Node.Element, pseudo: Selector.PseudoClass, scope: *Nreturn node == scope;},.empty => {- return node.firstChild() == null;+ // Only element and content (non-empty text/cdata) children affect+ // emptiness; comments and processing instructions are ignored.+ var child = node.firstChild();+ while (child) |c| : (child = c.nextSibling()) {+ switch (c._type) {+ .cdata => |cdata| switch (cdata._type) {+ .comment, .processing_instruction => {},+ else => if (cdata.getLength() > 0) return false,+ },+ else => return false,+ }+ }+ return true;},.first_child => return isFirstChild(el),.last_child => return isLastChild(el),@@ -660,7 +674,36 @@ fn matchesPseudoClass(el: *Node.Element, pseudo: Selector.PseudoClass, scope: *N},// Functional- .lang => return false,+ .lang => |expected| {+ if (expected.len == 0) return false;+ // The element's language is the nearest ancestor-or-self lang+ // attribute. Elements in a document with no declared language+ // fall back to the UA default (en); detached subtrees have no+ // language at all.+ var lang: ?[]const u8 = null;+ var current: ?*Node = node;+ while (current) |cur| : (current = cur.parentNode()) {+ switch (cur._type) {+ .element => |ancestor| {+ if (ancestor.getAttributeSafe(comptime .wrap("lang"))) |value| {+ lang = value;+ break;+ }+ },+ .document => {+ lang = "en";+ break;+ },+ else => {},+ }+ }+ const value = lang orelse return false;+ // Match the exact language or a `-` separated sub-tag prefix+ // (:lang(en) matches lang="en-AU"), ASCII case-insensitively.+ if (value.len < expected.len) return false;+ if (!std.ascii.eqlIgnoreCase(value[0..expected.len], expected)) return false;+ return value.len == expected.len or value[expected.len] == '-';+ },.not => |selectors| {for (selectors) |selector| {if (matches(node, selector, scope, frame)) {diff --git a/src/browser/webapi/selector/Parser.zig b/src/browser/webapi/selector/Parser.zigindex caf6f1d7f..29adb4b3b 100644--- a/src/browser/webapi/selector/Parser.zig+++ b/src/browser/webapi/selector/Parser.zig@@ -81,7 +81,11 @@ pub fn parseList(arena: Allocator, input: []const u8) ParseError![]const Selectovar remaining = preprocessed;while (true) {const trimmed = std.mem.trimLeft(u8, remaining, &std.ascii.whitespace);- if (trimmed.len == 0) break;+ if (trimmed.len == 0) {+ // Either an empty selector or a dangling comma ("div,"): both are+ // parse errors per the selector-list grammar.+ return error.InvalidSelector;+ }var comma_pos: usize = trimmed.len;var depth: usize = 0;@@ -137,10 +141,12 @@ pub fn parseList(arena: Allocator, input: []const u8) ParseError![]const Selectoconst selector_input = std.mem.trimRight(u8, trimmed[0..comma_pos], &std.ascii.whitespace);- if (selector_input.len > 0) {- const selector = try parse(arena, selector_input);- try selectors.append(arena, selector);+ if (selector_input.len == 0) {+ // An empty segment between commas (",div") is a parse error.+ return error.InvalidSelector;}+ const selector = try parse(arena, selector_input);+ try selectors.append(arena, selector);if (comma_pos >= trimmed.len) break;remaining = trimmed[comma_pos + 1 ..];@@ -946,6 +952,15 @@ fn attribute(self: *Parser, arena: Allocator) !Selector.Attribute {self.input = self.input[1..];_ = self.skipSpaces();+ // Optional namespace component: `*|name` (any namespace) or `|name`+ // (no namespace). Attributes are stored by qualified name and almost+ // never namespaced, so both forms match by name.+ if (self.peek() == '*' and self.input.len > 1 and self.input[1] == '|') {+ self.input = self.input[2..];+ } else if (self.peek() == '|' and self.input.len > 1 and self.input[1] != '=') {+ self.input = self.input[1..];+ }+const attr_name = try self.attributeName(arena);// Normalize the name to lowercase for fast matching (consistent with Attribute.normalizeNameForLookup)@@ -953,8 +968,12 @@ fn attribute(self: *Parser, arena: Allocator) !Selector.Attribute {var case_insensitive = false;_ = self.skipSpaces();- if (self.peek() == ']') {- self.input = self.input[1..];+ // Unexpected EOF closes all open constructs per CSS Syntax, so a+ // dangling `[foo` is a valid presence selector.+ if (self.peek() == ']' or self.peek() == 0) {+ if (self.peek() == ']') {+ self.input = self.input[1..];+ }return .{ .name = name, .matcher = .presence, .case_insensitive = case_insensitive };}@@ -976,10 +995,12 @@ fn attribute(self: *Parser, arena: Allocator) !Selector.Attribute {_ = self.skipSpaces();}- if (self.peek() != ']') {+ // As above, EOF stands in for the closing bracket ([align="center").+ if (self.peek() == ']') {+ self.input = self.input[1..];+ } else if (self.peek() != 0) {return error.InvalidAttributeSelector;}- self.input = self.input[1..];const matcher: Selector.AttributeMatcher = switch (matcher_type) {.exact => .{ .exact = value },