On 2011-08-25 15:07, Anthony Liguori wrote: > On 08/25/2011 08:02 AM, Avi Kivity wrote: >> On 08/25/2011 03:38 PM, Jan Kiszka wrote: >>> >> >>> >> What a mess. Do we really have to go through all 257 packed data >>> structs >>> >> in QEMU and add these MS compat bits to all potentially affected >>> ones? >>> > I prefer to detect -mms-bitfields and remove it in configure. >> >> Can use -mno-ms-bitfields later to override it. > > No, we can't do that. > > The reason glib uses -mms-bitfields is that you need to use it in order > to call Windows APIs which is does. We will eventually need to do it > anyway. > >> >>> /me too - if that is possible, ie. if the glib bits we are using doesn't >>> require us to apply that mode. Can anyone comment on this? >>> >> >> I'd shy away from that. Even if it works now it may break later. >> >> We should simply avoid bitfields on externally-defined formats; >> meanwhile we can use __attribute__((gcc_struct)) and keep using those we >> have already. > > +1
We aren't discussing bitfields anymore but essential unaligned (and therefore packed) protocol structs. And I'm not that much worried about going through the few slirp structs but all the other 250 packed ones in entire QEMU. Jan
signature.asc
Description: OpenPGP digital signature
