fix(client): replace lone UTF-16 surrogates instead of dropping the message

CHANGELOG.md

@@ -5,6 +5,10 @@
 ### Changed
 
 ### Fixed
+- A paused request with a binary body, e.g. a `Blob` upload with `blacklist=`/`whitelist=` set, hung forever: Chrome
+  sends the body with lone UTF-16 surrogates that the `json` gem rejects, and the fallback that unescaped them broke
+  the message further, so `Fetch.requestPaused` was dropped and the request was never continued. Lone surrogates are
+  now replaced with U+FFFD, so broken text from the page reads as `�` instead of `???` [rubycdp/cuprite#308]
 - Passing a node resolved on the previous page to `evaluate`, e.g. when the page navigates while Cuprite searches
   within an element, raised `Ferrum::BrowserError: Argument should belong to the same JavaScript world as target
   object`. It's `Ferrum::NodeNotFoundError` now, so callers can treat the node as stale and find it again

lib/ferrum/client/web_socket.rb

@@ -16,6 +16,7 @@ class WebSocket
       WEBSOCKET_BUG_SLEEP = 0.05
       DEFAULT_PORTS = { "ws" => 80, "wss" => 443 }.freeze
       SKIP_LOGGING_SCREENSHOTS = !ENV["FERRUM_LOGGING_SCREENSHOTS"]
+      ESCAPE_OR_LONE_SURROGATE = /\\u[dD][89abAB]\h{2}\\u[dD][c-fC-F]\h{2}|\\u([dD][89a-fA-F]\h{2})|\\./
 
       attr_reader :url, :messages
 
@@ -223,14 +224,12 @@ def safely_parse_json(data)
       rescue JSON::NestingError
         # nop
       rescue JSON::ParserError
-        safely_parse_escaped_json(data)
+        safely_parse_lone_surrogates(data)
       end
 
-      def safely_parse_escaped_json(data)
-        unescaped_unicode =
-          data.gsub(/\\u([\da-fA-F]{4})/) { |_| [::Regexp.last_match(1)].pack("H*").unpack("n*").pack("U*") }
-        escaped_data = unescaped_unicode.encode("UTF-8", "UTF-8", undef: :replace, invalid: :replace, replace: "?")
-        JSON.parse(escaped_data, max_nesting: false)
+      def safely_parse_lone_surrogates(data)
+        replaced = data.gsub(ESCAPE_OR_LONE_SURROGATE) { |escape| ::Regexp.last_match(1) ? "\\ufffd" : escape }
+        JSON.parse(replaced, max_nesting: false)
       rescue JSON::ParserError
         # nop
       end

spec/network_spec.rb

@@ -227,6 +227,18 @@
       expect(page.body).not_to include("Disappearing header")
     end
 
+    it "continues requests with a binary body" do
+      network.blacklist = /jquery/
+      page.go_to("/")
+
+      size = page.evaluate_async(<<~JS, page.timeout)
+        const bytes = new Uint8Array([0x22, 0x5c, 0x00, 0x0a, 0xed, 0xa0, 0x80, 0x41]);
+        fetch("/upload", { method: "PUT", body: new Blob([bytes]) }).then((r) => r.text()).then(arguments[0]);
+      JS
+
+      expect(size).to eq("8")
+    end
+
     it "blocks with single pattern" do
       network.blacklist = /jquery/
 

spec/node_spec.rb

@@ -482,7 +482,7 @@
       browser.go_to("/unicode")
       node = browser.at_css("#invalid")
 
-      expect(node.text).to eq("Havregrynskake med marengs og nøtter ???")
+      expect(node.text).to eq("Havregrynskake med marengs og nøtter \uFFFD")
     end
 
     it "works with curly utf8" do

spec/support/application.rb

@@ -88,6 +88,10 @@ def authorized?(login, password)
       "Current host is #{request.scheme}://#{request.host}:#{request.port}"
     end
 
+    put "/upload" do
+      request.body.read.bytesize.to_s
+    end
+
     get "/unicode" do
       File.read("#{FERRUM_VIEWS}/unicode.html")
     end