On Mon, Sep 14, 2020 at 05:36:30PM +0200, Thomas Huth wrote: > On 14/09/2020 15.46, Eduardo Habkost wrote: > > On Mon, Sep 14, 2020 at 07:39:09AM +0200, Thomas Huth wrote: > >> On 13/09/2020 04.51, Eduardo Habkost wrote: > >>> On Sat, Sep 12, 2020 at 08:45:19AM +0200, Thomas Huth wrote: > >>>> On 11/09/2020 22.06, Eduardo Habkost wrote: > >>>>> On Fri, Sep 11, 2020 at 08:06:10PM +0100, Peter Maydell wrote: > >>>>>> On Fri, 11 Sep 2020 at 19:49, Eduardo Habkost <ehabk...@redhat.com> > >>>>>> wrote: > >>>>>>> > >>>>>>> I'm wondering: do our supported build host platforms all include > >>>>>>> compilers that are new enough to let us redefine typedefs? > >>>>>>> > >>>>>>> The ability to redefine typedefs is a C11 feature which would be > >>>>>>> very useful for simplifying our QOM boilerplate code. The > >>>>>>> feature is supported by GCC since 2011 (v4.6.0)[1], and by clang > >>>>>>> since 2012 (v3.1)[2]. > >>>>>> > >>>>>> In configure we mandate either GCC v4.8 or better, or > >>>>>> clang v3.4 or better, or XCode Clang v5.1 or better > >>>>>> (Apple uses a different version numbering setup to upstream). > >>>>>> So you should probably double-check that that xcode clang has > >>>>>> what you want, but it looks like we're good to go otherwise. > >>>>> > >>>>> Can anybody confirm if the following is accurate? > >>>>> > >>>>> https://gist.github.com/yamaya/2924292#file-xcode-clang-vers-L67 > >>>>> # Xcode 5.1 (5B130a) > >>>>> Apple LLVM version 5.1 (clang-503.0.38) (based on LLVM 3.4svn) > >>>>> Target: x86_64-apple-darwin13.1.0 > >>>>> Thread model: posix > >>>>> > >>>>> If we know we have GCC 4.8+ or clang 3.4+, can we move to C11 and > >>>>> start using -std=gnu11? > >>>> > >>>> You don't have to switch to gnu11, redefintions of typedefs are already > >>>> fine in gnu99, they are a gnu extension there to the c99 standard. > >>>> > >>>> See also: > >>>> https://git.qemu.org/?p=qemu.git;a=commitdiff;h=7be41675f7cb16b > >>>> > >>>> https://www.mail-archive.com/qemu-devel@nongnu.org/msg585581.html > >>> > >>> They still trigger a warning with gnu99 on clang: > >>> > >>> $ clang --version > >>> clang version 10.0.0 (Fedora 10.0.0-2.fc32) > >>> Target: x86_64-unknown-linux-gnu > >>> Thread model: posix > >>> InstalledDir: /usr/bin > >>> $ cat test.c > >>> typedef struct A A; > >>> typedef struct A A; > >>> $ clang -std=gnu11 -c test.c > >>> $ clang -std=gnu99 -c test.c > >>> test.c:2:18: warning: redefinition of typedef 'A' is a C11 feature > >>> [-Wtypedef-redefinition] > >>> typedef struct A A; > >> > >> Ah, right, I forgot about that ... so for clang, we silence that warning > >> via CFLAGS in the configure script. See commit e6e90feedb706b1. > > > > Nice, I hadn't seen that. This means we don't need C11 for > > supporting redefinition of typedefs. > > > > Now, do we have other reasons for not moving to C11? It would be > > nice to make QEMU_GENERIC unnecessary and just use _Generic, for > > example. > > See https://www.mail-archive.com/qemu-devel@nongnu.org/msg585581.html > ... c11 is still "experimental" in GCC 4.8, so I think we likely have to > wait 'till summer next year - then we do not have to support > RHEL7/CentOS7 anymore according our support policy, and thus we can bump > the minimum required compiler versions.
Thanks Thomas and Daniel for the pointers. Staying with gnu99 for a little longer sounds reasonable, now that we typedef redefinitions are allowed. -- Eduardo
