From: Takashi Iwai Subject: Re: [alsa-cvslog] alsa-lib: Add struct timeval and timespec definition when _POSIX_C_SOURCE is not defined Date: Mon, 25 Jun 2007 17:31:17 +0200 Message-ID: References: <20070624133525.GA24739@procyon.home> <20070625152456.GC24551@newmaster.mivlgu.local> Mime-Version: 1.0 (generated by SEMI 1.14.6 - "Maruoka") Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Return-path: Received: from mx1.suse.de (ns.suse.de [195.135.220.2]) by alsa0.perex.cz (Postfix) with ESMTP id 21FFD24357 for ; Mon, 25 Jun 2007 17:31:18 +0200 (CEST) In-Reply-To: <20070625152456.GC24551@newmaster.mivlgu.local> List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: alsa-devel-bounces@alsa-project.org Errors-To: alsa-devel-bounces@alsa-project.org To: Sergey Vlasov Cc: alsa-devel@alsa-project.org, Jaroslav Kysela List-Id: alsa-devel@alsa-project.org At Mon, 25 Jun 2007 19:24:56 +0400, Sergey Vlasov wrote: > > On Mon, Jun 25, 2007 at 03:15:39PM +0200, Takashi Iwai wrote: > [...] > > > #include > > > #include > > > > > > $ gcc -c --std=c99 x.c > > > In file included from x.c:2: > > > /usr/include/alsa/global.h:134: error: redefinition of 'struct timeval' > > > /usr/include/alsa/global.h:139: error: redefinition of 'struct timespec' > > > > > > (Without #include only 'struct timespec' is redefined.) > > > > > > IMHO attempts to define structures which should be defined by system > > > headers are too dangerous - if someone is using preprocessor flags > > > which hide required parts of system headers, it is their problem, and > > > the risk of mismatch is too high. > > > > I agree. > > > > Jaroslav, which system has no definition of these types? Maybe it'd > > be better to check it in configure script, rather than the simple > > ifdef. > > This won't help here - the above problem happens when alsa-lib headers > are used by other programs, not when alsa-lib is built itself. But, it can check at least whether the system-wide struct timeval / timespec definitions. It wasn't clear to me for which system these definitions were needed... > > > BTW, does not provide 'struct timeval' at all, even with > > > _GNU_SOURCE - it defines only 'struct timespec'; however, this does > > > not cause problems, because ALSA headers currently use only pointers > > > to snd_timestamp_t, which works even without a complete definition of > > > 'struct timeval'. We can try to include to get 'struct > > > timeval', or even (which currently seems to provide > > > both 'struct timeval' and 'struct timespec' even with -ansi, and > > > http://www.opengroup.org/onlinepubs/009695399/basedefs/sys/select.h.html > > > tells that it should always provide their definitions). > > > > A patch is welcome :) > > Just not sure which one... > > The situation on my system (with glibc-2.5) is as follows: > > - defines 'struct timespec', unless -ansi (or -std=c99) > option is used; it never defines 'struct timeval', even with > -D_GNU_SOURCE (which should enable everything). This means that if > only is used, alsa/timer.h will not compile with -ansi due > to the 'snd_htimestamp_t tstamp' member in snd_timer_tread_t, > unless the program includes some other header which defines 'struct > timespec' before alsa/timer.h. > > - defines both 'struct timeval' and 'struct timespec', > even with -ansi; however, according to the current Open Group spec > (http://www.opengroup.org/onlinepubs/009695399/basedefs/sys/time.h.html), > must provide only 'struct timeval' and is not required > to provide 'struct timespec', so the behavior may change in a > future glibc release - again, a missing 'struct timespec' > definition is fatal for alsa/timer.h. > > - defines both 'struct timeval' and 'struct timespec' > even with -ansi. In the Open Group spec is > specified to define 'struct timeval', and also define 'struct > timespec' "as described in " (where it is described as > optional), so it should be not worse than . > > Would it be OK to include instead of and > remove attempt to define system structs from alsa/global.h? Another > option is to include both and . Yes, sys/select.h sounds reasonable. thanks, Takashi