evhttp_request_own does not work as expected about libevent

[nmathewson]

Hm. It could be that there's another evhttp_request_free somewhere in here happening first. Can you use a debugger or valgrind or something to see where the first call to evhttp_request_free is happening?

[pphaneuf]

A request still in progress when an evhttp_connection gets freed will get blown away:

https://github.com/libevent/libevent/blob/patches-2.0/http.c#L1142

Is that expected? I'd think that if you used evhttp_request_own, the user should be the one doing the evhttp_request_free?

[pphaneuf]

Got started on a pull request to fix this (it fixes the case that the example program above triggers), but I think we should go over the calls to evhttp_request_free in http.c and review whether they should be conditional or not (which I could add to this pull request).

[errzey]

Just for my own information: we have made no claims that after evhttp_request_own() we do further processing on it, correct? in evhtp, the function evhtp_take_ownership() relinquishes all further processing from the library.

If this is the case, maybe a simple macro for internal use only:

#define evhttp_request_free_(req_)  do {           \
    if ((req_->flags & EVHTTP_USER_OWNED) == 0) {  \
         evhttp_request_free(req_);                \
    } } while(0)                                   \

[pphaneuf]

My understanding is that this function only changes the memory ownership (in other words, processing continues, but now the user is responsible for freeing the request, as maybe they want to access the header/body after returning from the callback).

[azat]

So I decided to use inline function instead of macros (since it is easy to debug), other then this it is the same as @ellzey suggested. Plus add a regression test for this.