Created attachment 107183 [details]
Patch which fixes the issue

There is a special case in getTokenId() that looks like this:

case 65535: // SIGN ERRORS! Why does this happen?

The way I see it the problem seems to be this line in nextToken():

token = tokenStream.getToken() & Parser.CLEAR_TI_MASK;

where if tokenStream.getToken() returns org.mozilla.nb.javascript.Token.ERROR which is actually -1 (the only negative token id), in two-s complement it has all the bits on 1. So -1 & Parser.CLEAR_TI_MASK (which is 0xFFFF) gives unsurprisingly the 65535 value.

PS: The good thing is that the 'case 65535' is treated as JsTokenId.ERROR anyhow, as it should.