The send_output method which writes to the buffer triggers a write attempt. So that should not be a problem.
I see. Then I guess it works, although personally I think it's a bit obscure to rely on the internal write cb blocker. Anyway, I've applied your patch. Thanks.