Gecko certainly ignores open() calls on random documents that have already been unloaded or are the result of a DOMParser parse invocation or whatnot. Not sure what other UAs do.

The spec has some efforts at doing this already (e.g. not allowing it during unload); but I'm happy to add more. Do you have any test cases showing this?

  var doc = new DOMParser().parseFromString("<span>", "text/html")
  alert(doc.open());
  alert(doc.querySelector("span"));

Or:

  var ifr = document.createElement("iframe");
  document.documentElement.appendChild(ifr);
  var doc = ifr.contentDocument;
  doc.open();
  doc.write("<span>");
  doc.close();
  ifr.parentNode.removeChild(ifr);
  alert(doc.open());
  alert(doc.querySelector("span"));

Or:

  var ifr = document.createElement("iframe");
  document.documentElement.appendChild(ifr);
  var doc = ifr.contentDocument;
  doc.open();
  doc.write("<span>");
  doc.close();
  setTimeout(function(){
    ifr.onload = function() {
      alert(doc.querySelector("span"));
      alert(doc.open());
      alert(doc.querySelector("span"));
    };
    ifr.contentWindow.location = "http://w3.org";
  }, 100);


Note that for all three of these the current spec makes no sense, since it talks about navigating and whatnot, but that assumes a browsing context. In particular, navigating in the third testcase would be pretty odd.

Note also that UA interop might not be great here.  ;)  In particular, I think Opera throws on the open() in all three cases, Gecko returns null and does nothing, WebKit clears out the DOM but doesn't do any obvious navigating.

Yeah, the spec makes no sense here. It assumes the doc is active.

Looking at tests: Wow, there's just no interop here.

e.g. http://software.hixie.ch/utilities/js/live-dom-viewer/saved/2006
 Opera: replaces the currently rendered page with the new text (!)
 WebKit: nukes the doc, writes to it off-screen
 Gecko: does nothing (ignores .open and .write calls)

Should I just abort if the doc is not the active document of its browsing context, or if it has no browsing context? Or just wrap the stuff that only makes sense for active docs with a check that the doc is active?

I should probably test IE before deciding, based on the total lack of interop I've seen so far.

IE10RP Win7:
log: object "[object Document]" (259 props: doctype="null", documentElement="null", implementation="[object DOMImplementation]", inputEncoding="UTF-8"...)
log: undefined
log: object "null" (0 props)
log: object "[object HTMLPhraseElement]" (272 props: cite="", dateTime="", currentStyle="[object MSCurrentStyleCSSProperties]", runtimeStyle="[object MSStyleCSSProperties]"...)
IFRAME says "hello!".

In IE10 mode.

For http://software.hixie.ch/utilities/js/live-dom-viewer/?saved=2007

Just tried this again in the same browser and it says "error: Permission denied on line 10".

And / is displayed.

> Should I just abort if the doc is not the active document of its browsing
> context, or if it has no browsing context?

That's my preference, obviously, but I might be biased by that being what Gecko does and simple to implement.

> Or just wrap the stuff that only makes sense for active docs with a check that
> the doc is active?

That would probably be OK too, but implementing in practice might be a pain.

IE apparently does what Opera does. I don't understand how that happens.

It seems to depend on timing.

Checked in as WHATWG revision r7626.
Check-in comment: Make document.open() do nothing on non-active documents
http://html5.org/tools/web-apps-tracker?from=7625&to=7626