and furthermore error out of one of them isn't a DNS option.
Test case by Ben Lund <ben.lund@gmail.com>; additional help from Tim Caswell
<tim@creationix.com>.
Changed ReallyEmit so that it clones the Array of listeners before
processing the emit. Added better tests to make sure that modifying
listeners inside event handlers doesn't cause later listeners to be skipped
or added.
After getting some feedback from Mikeal Rogers and Tim Smart, it was decided
that evalcx should not try to do any fancy security stuff, and instead leave
that in the hands of the user. To comply more with spidermonkey, everything
is passed in, and objects are passed in by reference rather than being
cloned.
1. Move the context->Enter() call so that the global obj is available for writing.
2. On success, copy the modified global out to the sandbox object.
3. Don't copy functions in either direction. They have scope and closures, and make for craziness when trying to keep contexts separate.
4. Only do the ->ToObject->Clone() on objects, so that simple values stay simple.
5. Update the test so that it tests all this stuff.
Update the ini parser to support some more whitespace cases, turn lines
without an equal sign into a "flag" that's just true if set, and support
comments.
Allows for more fine graining, especially finding out about an individual
chunk of data being flushed in a write stream rather than the whole queue.
This commit also fixes a bug causing forceClose to fail on a readStream that
did not finish opening yet.