fix(page): drop cached node ids after DOM.getDocument

CHANGELOG.md

@@ -18,6 +18,9 @@
 - A `Ferrum::Node` found before a navigation or reload kept its cached CDP node id, which Chrome reuses for the new
   document, so `focus`, `click`, `scroll_into_view` and friends silently hit another element. Node ids are now dropped
   on `DOM.documentUpdated` and such nodes raise `Ferrum::NodeNotFoundError` [#86]
+- Replacing an iframe's content with `frame.content=` or querying the whole page with `accessibility.query` broke every
+  node found before it with `Ferrum::NodeNotFoundError`, even though the nodes were still on the page. Their ids are
+  requested again now. A node used while the page is still loading a new document says so in its error [#650]
 - Behind a proxy with credentials, `request.abort` in `on(:request)` and `network.blacklist`/`whitelist` were ignored
   because Ferrum's own proxy handler continued every request first. It now only continues requests nobody else handles [#332]
 

lib/ferrum/node.rb

@@ -35,7 +35,9 @@ def node_id
       document_version = page.document_version
       @node_id = begin
         id = page.command("DOM.requestNode", objectId: @object_id)["nodeId"]
-        raise NodeNotFoundError, { "message" => "node is not trackable" } if id.zero?
+        if id.zero?
+          raise NodeNotFoundError, { "message" => "node is not trackable, the page is still loading a new document" }
+        end
 
         id
       rescue NoExecutionContextError

lib/ferrum/page.rb

@@ -87,8 +87,8 @@ class Page
     # @return [Downloads]
     attr_reader :downloads
 
-    # Bumped every time Chrome reports `DOM.documentUpdated`, which
-    # invalidates every node id handed out before it.
+    # Bumped every time Chrome reports `DOM.documentUpdated` and after every
+    # `DOM.getDocument`, both invalidate every node id handed out before them.
     #
     # @return [Integer]
     attr_reader :document_version
@@ -521,9 +521,13 @@ def use_authorized_proxy?
     #   The root node id, or `true` when sent asynchronously.
     #
     def document_node_id(async: false)
-      return client.command("DOM.getDocument", async: true, depth: 0) if async
-
-      command("DOM.getDocument", depth: 0).dig("root", "nodeId")
+      result = if async
+                 client.command("DOM.getDocument", async: true, depth: 0)
+               else
+                 command("DOM.getDocument", depth: 0).dig("root", "nodeId")
+               end
+      @document_version += 1
+      result
     end
 
     private

spec/node_spec.rb

@@ -61,6 +61,30 @@
       expect(browser.evaluate("document.activeElement.id")).not_to eq("customer_email")
     end
 
+    it "keeps working after an iframe's content is replaced" do
+      browser.go_to("/form")
+      node = browser.at_css("#customer_email")
+      node.focus
+      node.blur
+
+      browser.frame_by(name: "frame").content = "<p>new</p>"
+      node.focus
+
+      expect(browser.evaluate("document.activeElement.id")).to eq("customer_email")
+    end
+
+    it "keeps working after an accessibility query of the whole page" do
+      browser.go_to("/form")
+      node = browser.at_css("#customer_email")
+      node.focus
+      node.blur
+
+      browser.accessibility.query(role: "textbox")
+      node.focus
+
+      expect(browser.evaluate("document.activeElement.id")).to eq("customer_email")
+    end
+
     it "raises an error if the element is not visible" do
       browser.go_to("/index")