Re: [PATCH] Fixed Android support
On Wed, May 31, 2017 at 8:15 PM, Marc Lehmann wrote: > On Tue, May 30, 2017 at 05:28:07PM -0700, enh wrote: >> The attached patch fixes Android NDK bug >> https://github.com/android-ndk/ndk/issues/396 --- basically the >> presence of EPOLL_CLOEXEC doesn't necessarily imply the presence of >> epoll_create1 on Android. This is because (as of NDK r15) we're using > > If android uses header files that don't match it's libc *on purpose* then > android is way beyond broken. no, these are the Android libc header files. >> Gingerbread, you'll have all the enums and macros and structs from >> Marshmallow and O. so you actually need a proper configure test for >> epoll_create1. > > configure is optional for libev, so a "proper" test wouldn't work. i'm not sure what you're saying here: you already use configure and have correctly used HAVE_ tests for other functions. > This is clearly a bug in the android headers, and since it apparently only > affects old versions, I don't think it makes sense to make things worse > for users of higher quality platforms that have correct header files. no, it's the opposite way round. older versions of Android didn't have epoll_create1. newer versions do. if you want to keep building on NDK r15 and later, you'll need to check for the function, not the constant. (the constants come from the upstream kernel uapi header files, and a given Android release doesn't mandate any corresponding kernel version, so there's no way to version that.) > I would suggest opening a bug against android so these obvious bugs get > fixed instead of trying to push patches to the rest of the world to work > around it. i manage the Android libc/NDK team :-) this change is a result the NDK catching up with the platform headers after being years behind. > Unrelated, the __ANDROID__ change will be in the next release. > > Greetings, > > -- > The choice of a Deliantra, the free code+content MORPG > -==- _GNU_ http://www.deliantra.net > ==-- _ generation > ---==---(_)__ __ __ Marc Lehmann > --==---/ / _ \/ // /\ \/ / schm...@schmorp.de > -=/_/_//_/\_,_/ /_/\_\ ___ libev mailing list libev@lists.schmorp.de http://lists.schmorp.de/mailman/listinfo/libev
Re: [PATCH] Fixed Android support
On Tue, May 30, 2017 at 05:28:07PM -0700, enh wrote: > The attached patch fixes Android NDK bug > https://github.com/android-ndk/ndk/issues/396 --- basically the > presence of EPOLL_CLOEXEC doesn't necessarily imply the presence of > epoll_create1 on Android. This is because (as of NDK r15) we're using If android uses header files that don't match it's libc *on purpose* then android is way beyond broken. > Gingerbread, you'll have all the enums and macros and structs from > Marshmallow and O. so you actually need a proper configure test for > epoll_create1. configure is optional for libev, so a "proper" test wouldn't work. This is clearly a bug in the android headers, and since it apparently only affects old versions, I don't think it makes sense to make things worse for users of higher quality platforms that have correct header files. I would suggest opening a bug against android so these obvious bugs get fixed instead of trying to push patches to the rest of the world to work around it. Unrelated, the __ANDROID__ change will be in the next release. Greetings, -- The choice of a Deliantra, the free code+content MORPG -==- _GNU_ http://www.deliantra.net ==-- _ generation ---==---(_)__ __ __ Marc Lehmann --==---/ / _ \/ // /\ \/ / schm...@schmorp.de -=/_/_//_/\_,_/ /_/\_\ ___ libev mailing list libev@lists.schmorp.de http://lists.schmorp.de/mailman/listinfo/libev
[PATCH] Fixed Android support
The attached patch fixes Android NDK bug https://github.com/android-ndk/ndk/issues/396 --- basically the presence of EPOLL_CLOEXEC doesn't necessarily imply the presence of epoll_create1 on Android. This is because (as of NDK r15) we're using the same headers for all API levels, so even if you're targeting Gingerbread, you'll have all the enums and macros and structs from Marshmallow and O. so you actually need a proper configure test for epoll_create1. It also seems that the existing Android support had bitrotted over the years, so this patch fixes that too (you need __ANDROID__, not ANDROID). With this patch I can build for either old or new releases (I tested with JellyBean and yet-to-be-released O), and I correctly get a reference to epoll_create1 or not, depending. I also checked that this didn't break glibc. Index: ev.c === RCS file: /schmorpforge/libev/ev.c,v retrieving revision 1.480 diff -u -r1.480 ev.c --- ev.c18 Feb 2016 04:48:05 - 1.480 +++ ev.c31 May 2017 00:27:06 - @@ -365,13 +365,10 @@ # define EV_HEAP_CACHE_AT EV_FEATURE_DATA #endif -#ifdef ANDROID -/* supposedly, android doesn't typedef fd_mask */ -# undef EV_USE_SELECT -# define EV_USE_SELECT 0 -/* supposedly, we need to include syscall.h, not sys/syscall.h, so just disable */ -# undef EV_USE_CLOCK_SYSCALL -# define EV_USE_CLOCK_SYSCALL 0 +#ifdef __ANDROID__ +/* Android doesn't typedef fd_mask. */ +# undef EV_SELECT_USE_FD_SET +# define EV_SELECT_USE_FD_SET 1 #endif /* aix's poll.h seems to cause lots of trouble */ Index: ev_epoll.c === RCS file: /schmorpforge/libev/ev_epoll.c,v retrieving revision 1.71 diff -u -r1.71 ev_epoll.c --- ev_epoll.c 18 Feb 2016 04:48:05 - 1.71 +++ ev_epoll.c 31 May 2017 00:27:06 - @@ -239,7 +239,7 @@ int epoll_init (EV_P_ int flags) { -#ifdef EPOLL_CLOEXEC +#ifdef HAVE_EPOLL_CREATE1 backend_fd = epoll_create1 (EPOLL_CLOEXEC); if (backend_fd < 0 && (errno == EINVAL || errno == ENOSYS)) Index: libev.m4 === RCS file: /schmorpforge/libev/libev.m4,v retrieving revision 1.16 diff -u -r1.16 libev.m4 --- libev.m428 Oct 2013 12:36:44 - 1.16 +++ libev.m431 May 2017 00:27:06 - @@ -4,7 +4,7 @@ dnl libev support AC_CHECK_HEADERS(sys/inotify.h sys/epoll.h sys/event.h port.h poll.h sys/select.h sys/eventfd.h sys/signalfd.h) -AC_CHECK_FUNCS(inotify_init epoll_ctl kqueue port_create poll select eventfd signalfd) +AC_CHECK_FUNCS(inotify_init epoll_create1 epoll_ctl kqueue port_create poll select eventfd signalfd) AC_CHECK_FUNCS(clock_gettime, [], [ dnl on linux, try syscall wrapper first ___ libev mailing list libev@lists.schmorp.de http://lists.schmorp.de/mailman/listinfo/libev