test failure after issue29 - 3 test on IE11

[jmesserly]

Running IE11 on the "issue 29 - 3" test it passes, but there's a failure on cleanup afterwards.

Not entirely sure if it's that test, the results seem timing dependent. Some of the early investigation notes are here: #43 (comment)

(edit: it's 29-3 not 29-2. can be seen by running http://localhost:8000/polymer-expressions/tests/index.html?grep=issue-29 - 3)

[jmesserly]

I think it's probably a race condition, since adding some logging code made it pass. But I don't know for sure yet.

[jmesserly]

possibly related to googlearchive/TemplateBinding#184

[jmesserly]

I found a "fix" -- add this print statement to the end of insertInstanceAt:

console.log(previousInstanceLast.nextSibling);

Naturally, the console needs to be open for it to work. So this is pretty deep into heisenbug territory.

My hypothesis: under certain conditions IE loses the identity of the Text node in the tree. Perhaps this happens because we're inserting from a DocumentFragment, or because of the adoptNode we've done before, or just some random bug (refcount? GC?) that we happen to be hitting.

This appears to not be the first time, @arv worked around a similar issue here: https://github.com/Polymer/ShadowDOM/blob/master/src/wrappers/Node.js#L436 though I'm surprised this kind of thing is still happening in 11 :|

I found a not-totally-terrible workaround we can do in TemplateBinding, by reassigning the terminator_ property. On most browsers it will be a no-op, whereas on IE it will ensure we get the new Text node instance, instead of the zombie one. Will post the code change shortly.

[jmesserly]

I found a not-totally-terrible workaround we can do in TemplateBinding

Scratch that. The no-op workaround makes that particular test pass when run it isolation, but moves the problem to another place. Drat.

[jmesserly]

landed a workaround, but still need a simplified repro for https://connect.microsoft.com/IE