Re: [Libevent-users] free bufferevents on all data is sent in filters chain

[Nick Mathewson <nickm@xxxxxxxxxxxxx>]

On Thu, Feb 6, 2014 at 2:42 PM, Artem <ru.netz@xxxxxxxxx> wrote:
> Greetings,
> I'm trying to understand how to properly free bufferevents when all data is
> finished sending using filters chain.
> I.e. I have one bufferevent and bufferevent_filter on top of it. I'm writing
> some data into filter, and on some event I'm deciding that all data is sent
> and I want to close it. If I close it immediately with bufferevent_free I'll
> loose data that is in output_buffer, but is not sent to the client. I there
> is no filter I'm setting write callback and, there I check if there is still
> data in output_buffer and call bufferevent_free if it's empty. With filters
> it doesn't work because filter write_callback is called when underlying
> buffer can accept data AND there is data in filter buffer (BEV_NORMAL), so
> if there is no data in filter buffer, but there is some in underlying
> buffer, it will not be called.
>   To fix this there should be got_out_eof in struct bufferevent_filtered,
> that should be set when bufferevent_flush is called with EV_WRITE, after
> that be_filter_process_output should call user write callback no mater what.
> After that in user write callback we check if all output buffers in filters
> chain are empty and if yes, then free the filter bufferent. Am I missing
> something?

So, the current workaround is to free the outer bufferevent without
freeing the inner one, and then set your own callbacks on the inner
one so you can close it once it's flushed.

I think that more generally, we need a "flush and then free" call for
the various bufferevent types, possibly with a callback to invoke
before freeing.  It seems to be a common request.
***********************************************************************
To unsubscribe, send an e-mail to majordomo@xxxxxxxxxxxxx with
unsubscribe libevent-users    in the body.