On Mon, Jul 10, 2017 at 9:49 AM, Przemysław Sobala <[email protected]> wrote: > W dniu sobota, 8 lipca 2017 00:42:31 UTC+2 użytkownik Ben Noordhuis napisał: >> >> On Fri, Jul 7, 2017 at 10:50 AM, Przemysław Sobala >> <[email protected]> wrote: >> > W dniu piątek, 7 lipca 2017 10:29:13 UTC+2 użytkownik Przemysław Sobala >> > napisał: >> >> >> >> W dniu czwartek, 6 lipca 2017 17:30:05 UTC+2 użytkownik Ben Noordhuis >> >> napisał: >> >>> >> >>> On Thu, Jul 6, 2017 at 12:21 PM, Przemysław Sobala >> >>> <[email protected]> wrote: >> >>> > Hello >> >>> > [In the preface I want to say that it can be valgrind's false >> >>> > positive]. >> >>> > I'm building a C++ class that uses libuv (1.13.0) and libcurl >> >>> > (7.54.1) >> >>> > for >> >>> > non-blocking file downloading. UV loop is being initialized in >> >>> > object's >> >>> > constructor, deinitialized in destructor and it's being kept a class >> >>> > member. >> >>> > When it's a pointer type class member (uv_loop_t *), initialized via >> >>> > malloc: >> >>> > loop = (uv_loop_t *) malloc(sizeof(uv_loop_t)); >> >>> > if (loop == NULL) { >> >>> > throw std::bad_alloc(); >> >>> > } >> >>> > uv_loop_init(loop); >> >>> > and deinitialized via uv_loop_close(loop) there's no memory leak. >> >>> > >> >>> > But when it's a struct type class member (uv_loop_t) it's >> >>> > initialized >> >>> > automatically while object construction, then in constructor I call >> >>> > uv_loop_init(&loop) and uv_loop_close(&loop) in destructor, valgrind >> >>> > reports >> >>> > a memory leak: >> >>> > ==2337== 256 bytes in 1 blocks are definitely lost in loss record >> >>> > 790 >> >>> > of 840 >> >>> > ==2337== at 0x4C2FC47: realloc (vg_replace_malloc.c:785) >> >>> > ==2337== by 0x816560: maybe_resize (core.c:808) >> >>> > ==2337== by 0x816560: uv__io_start (core.c:847) >> >>> > ==2337== by 0x8181EB: uv_poll_start (poll.c:120) >> >>> > ==2337== by 0x417ECE: >> >>> > imageresizer::engine::FileDownloader::handle_socket(void*, int, int, >> >>> > void*) >> >>> > (FileDownloader.cpp:298) >> >>> > ==2337== by 0x417E28: >> >>> > imageresizer::engine::FileDownloader::handle_socket_cb(void*, int, >> >>> > int, >> >>> > void*, void*) (FileDownloader.cpp:277) >> >>> > ==2337== by 0x7D1798: singlesocket (in >> >>> > >> >>> > /opt/WP/imageresizer-worker/dist/Debug/GNU-Linux/imageresizer-worker) >> >>> > ==2337== by 0x7D4AD9: multi_socket (in >> >>> > >> >>> > /opt/WP/imageresizer-worker/dist/Debug/GNU-Linux/imageresizer-worker) >> >>> > ==2337== by 0x7D4C86: curl_multi_socket_action (in >> >>> > >> >>> > /opt/WP/imageresizer-worker/dist/Debug/GNU-Linux/imageresizer-worker) >> >>> > ==2337== by 0x417D59: >> >>> > imageresizer::engine::FileDownloader::on_timeout(uv_timer_s*) >> >>> > (FileDownloader.cpp:248) >> >>> > ==2337== by 0x417D25: >> >>> > imageresizer::engine::FileDownloader::on_timeout_cb(uv_timer_s*) >> >>> > (FileDownloader.cpp:242) >> >>> > ==2337== by 0x81D4B4: uv__run_timers (timer.c:165) >> >>> > ==2337== by 0x8159AB: uv_run (core.c:353) >> >>> > >> >>> > Can you help me with getting rid of that memory leak? >> >>> > >> >>> > -- >> >>> > regards >> >>> > Przemysław Sobala >> >>> >> >>> >> >>> Check the return value of uv_loop_close(). My guess it's UV_EBUSY, >> >>> indicating the event loop can't be closed yet because there are open >> >>> handles or requests. >> >> >> >> >> >> Yes, but uv_loop_close() returns UV_EBUSY in both cases. >> >> Should I wait some time for all handles to close or iterate over all >> >> handles inside loop and call uv_close() ? >> > >> > >> > I've configured my asynchronous loop stop callback as this: >> > uv_async_init(&loop, &loop_close_event, [](uv_async_t* handle) { >> > uv_stop(handle->loop); >> > uv_walk(handle->loop, >> > [](uv_handle_t *handle, void *arg) { >> > uv_close(handle, NULL); >> > }, NULL); >> > }); >> > >> > And now uv_loop_close returns 0 and valgrind reports no memory leak. Is >> > that >> > a correct approach? >> >> Yes, that's one way to do it, with two caveats: >> >> 1. You probably don't need to call uv_stop(). >> 2. Closing handles indiscriminately is usually not a good idea unless >> you own every handle. > > > I think I get it. As soon as I close all loop's handles the loop will quit > itself without the redundant uv_stop() call. > > What is the most preferable way to asynchronously stop and gracefully close > the loop while it's working (iterating) e.g. inside the signal handler?
Depends on what you mean by asynchronous. If you mean from inside uv_run(), then uv_stop() is not your worst option. You still need to close open handles afterwards, though. -- You received this message because you are subscribed to the Google Groups "libuv" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To post to this group, send email to [email protected]. Visit this group at https://groups.google.com/group/libuv. For more options, visit https://groups.google.com/d/optout.
