Re: [libmicrohttpd] Expose the fdset events

[Simon Newton]

On Sun, Nov 25, 2012 at 10:43 AM, Christian Grothoff
<address@hidden> wrote:
> On 11/25/2012 05:07 PM, Simon Newton wrote:
>>
>> I use microhttpd with the Open Lighting Architecture. As I've posted
>> before we have a C++ wrapper around libmicrohttpd (HTTPServer), which
>> then integrates with OLA's event management system.
>>
>> While this works, it's rather messy. The wrapper sets up a call to a
>> HTTPServer::UpdateSockets() method into the event loop, and
>> UpdateSockets() calls MHD_get_fdset to find out what FDs microhttpd is
>> interested in.
>>
>> It then loops over all the FDSETs and tries to reconcile the event
>> manager's state with the desired state from microhttpd. This involves
>> adding and removing Socket objects from the event manager.
>>
>> I'd prefer it if I could set a callback to be run each time a FD
>> should be added or removed from the event manager. Is that easy enough
>> to do?
>
>
> "Easy enough" is hard to say, what I can say is that it would not be
> trivial.
>
> However, that doesn't mean it will never happen --- something similar would
> be needed to make MHD work with epoll/kqueue-based event loops, which is in
> general a nice system call that in a perfect world MHD should be able to
> utilize.

Yes, that's why I brought it up. The event manager started as a
wrapper around select() but has grown more complex over time. I'm
looking at changing it over to epoll.

>
> Nevertheless, the change would be quite invasive and would require work that
> I won't be able to put in anytime soon.  So I guess what I'm saying is that,
> while I'm not opposed, I don't see this happen anytime soon unless I
> magically get a patch ;-).

Ok, if I get some free time I'll have a go.

>
> Also, have you considered "improving" your event manager by extending its
> interface to allow it to work also with fdsets directly?  Depending on the
> system call that are you using internally in your event manager this might
> be trivial.

I'd have to have a way to associate callbacks with the fdsets, and
then that starts to look a lot like the Descriptor objects we have
now, only they'd have multiple fds associated with a single callback.
Maybe it's worth adding, it's hard to say.

Simon

>
> Happy hacking!
>
> Christian
>
>