Hi Nicholas, I've looked at the code, and I don't think this is a bug in MHD. Basically, the issue is that you're telling MHD that you cannot immediately process all POST data (by partially handling the data). But MHD never does busy waiting, and we have now no way of telling when your application might be ready to handle the rest of the POST.
MHD's behavior in this case is to avoid the busy loop, and instead simply retry at the next opportunity, which as you rightly observe might be 'never'. I've clarified that we consider this an application bug by printing a warning in SVN 37545 whenever an application behaves "badly" in this sense. The correct behavior (for your code) would be to call MHD_suspend_connection() as you are returning from handling the POST partially (with more than 0-bytes left in upload data). Then, once you are ready to handle the rest of the data, you should call MHD_resume_connection(). Alternatively, in "thread-per-connection" mode, you should block until you've processed all the data. Only in "external" select mode, you don't have to suspend the connection as you can force MHD to re-trigger the handler by calling MHD_run(). As of SVN 37545, MHD now logs a warning if applications keep data unprocessed while also not suspending the connection and also not running in external select mode. I hope this will help everyone in (1) finding these corner cases, and (2) understanding how to handle them properly. Happy hacking! Christian On 07/14/2016 06:34 AM, Nicholas Cooper wrote: > Hi, > > Let's say the access handler callback passed to MHD_start_daemon() is > answer_to_connection(). According to the comment in microhttpd.h, the > parameter upload_data_size shall be updated to the number of bytes not > processed. However, it seems that it only works perfectly if the > *upload_data_size is updated to zero. If its value is set to non-zero, > answer_to_connection() won't be called again, which leaves the HTTP > client waiting forever. To be more exact, unless the client sends > another (unrelated) request, MHD calls answer_to_connection() for the > suspending request again. Don't know if this is a bug, or I'm misusing > MHD. Having tried this on both Linux (Debian) and Windows, and both > don't work. The attached file demonstrates the problem. > > Thanks. Have a nice day. > > Nicholas Cooper.
signature.asc
Description: OpenPGP digital signature
